Professional Documents
Culture Documents
Handbook of Antenna Design2
Handbook of Antenna Design2
The
Handbook of
Antenna
Design
VOLUME 2
Editors
A.W. Rudge K. Milne
A.D. OIver P. Knight
Peter Peregrinus Ltd on behalf of the Institution of Electrical Engineers
IEE ELECTROMAGNETIC WAVES SERIES 16
SERIES EDITORS: PROFESSOR P.J.B. CLARRICOATS
E.D.R. SHEARMAN
AND J.R. WAIT
The
Handbook of
VOLUME 2
Previous volumes in this series
VOLUME 2
Editors
A.W.Rudge K. Milne
A.D.OIver P.Knight
Peter Peregrinus Ltd on behalf of the Institution of Electrical Engineers
List of Contributors
VOLUME 2
ISBN 0-906048-87-7
List of contributors iv
Preface xiii
9 Linear arrays - R. C. Hansen 1
Introduction 1
9.1 Design of array excitation 2
9.1.1 Narrow beam, low sidelobe pattern 4
9.1.2 Pattern synthesis 42
9.1.3 Superdirective arrays 54
9.1.4 Thinned arrays 68
9.1.5 Array tolerances and limits 73
9.2 Array feeds 79
9.2.1 Resonant arrays 81
9.2.2 Travelling-wave arrays 87
9.2.3 Corporate line and distributed feeds 98
9.2.4 Phaser and sub-array lobes 104
9.3 Array elements 109
9.3.1 Dipoles 109
9.3.2 Waveguide slots 113
9.3.3 Scan compensated elements 121
9.3.4 Printed-circuit antennas 126
9.3.5 Moderate gain elements 133
9.4 References 134
been nowhere more evident that at the shorter wavelengths, where the combination
of digital computer and mathematical method has found numerous applications in
the design of quasi-optical antennas. Computer codes based upon geometric optics,
physical optics, the geometric theory of diffraction, Fourier transform theory, arid
spherical-wave analysis now provide essential tools in modern antenna design.
Thoughtful application of these analytical methods has lead, not only to improved
computations of antenna radiation characteristics, but also to improved insight into
the underlying physical phenomena.
The first volume of this text provides the mathematical background and a large
number of examples of the use of mathematical methods in the design of reflector
and lens antennas, including their primary-feeds. Quasi-optical antennas, including
hybrid configurations, where the reflector or lens is used in conjunction with a
complex array feed, are particularly well suited to the use of computer-aided design
methods and major performance improvements can be realised. These topics are
dealt with in Chapters 2-6 of the first volume.
Low and medium gain microwave antennas have many applications, of which
serving as a primary-feed is only one. Chapter 7 provides design data for a diverse
range of radiators including helices and spirals, slots, microstrip antennas, backfire
antennas, dielectric rods and horns.
Advances in antenna performance must be matched by improved electrical
measurements and there have been a number of significant developments in antenna
metrology during the past decade. The key developments in this important area are
described in the last chapter of the first volume.
The fundamental principles and design of antenna arrays has been the focus of a
great deal of attention during the past decade. Although cost still remains a barrier
in the wide scale deployment of large phased-arrays a continued growth in the
applications of antenna arrays can be anticipated. The importance of array tech-
nology has been recognised here in the dedication of the first five chapters of the
second volume to this theme. The subject matter covered includes linear, planar,
conformal and circular geometries in addition to array signal processing. It is
evident that the combination of antenna arrays with integrated-circuit technology
will have much to offer in the future in terms of both performance flexibility and
lower cost.
In practice many antennas are operated either behind or within a radome cover.
In such cases the electromagnetic characteristics of the radome must be considered
as a factor in arriving at the overall system performance. In Chapter 14 this
important topic is considered and design data are provided for a variety of radome
types.
The last few chapters of the second volume are concerned with the design of
antennas and coaxial components in the LF to UHF frequency bands. In general
design data for these bands are less well documented than their counterparts in
the microwave region, and the opportunity has been taken here to bring together
design information from a number of sources including a rather sparse and widely
dispersed literature.
Preface xv
In preparing these volumes the editors are indebted to the authors, all of whom
are well known experts in their fields, who have somehow found time in their very
busy schedules to organise and prepare their contributions. The authors have
exhibited a remarkable degree of patience in seeing through a project which has
taken several years to complete. The work described here has been performed by a
great many engineers and scientists over a period of many years and although a
sensible effort has been made to ensure that the references and credits are accurate,
the editors offer apologies for any errors or omissions in this respect.
For our part as editors we have sought to bring together the contributions into
a cohesive whole and have on occasions made modifications or deletions to achieve
this end. Any resulting omissions, misprints or errors in the text remains our
responsibility and we shall be glad to have notice of them to correct any future
editions. Although some effort has been made to standardise the notations
employed throughout the text, the two volumes represent the efforts and individual
styles of 28 contributing authors and, in view of the excessive work entailed, a
comprehensive cross-correlation between chapters has not been undertaken. Vector
notation^ have been indicated by the use of either bold type as in E, H, or by use of
bars as E, H. References, equations and figures are numbered in sequence on a
chapter basis, and the references are listed at the end of each chapter. A compre-
hensive index is provided at the end of each volume.
It is the editors hope that this handbook will be of value both to practising
design engineers and to students of antenna theory. In the preparation of the text
it has been assumed that the reader has at least a working knowledge of electro-
magnetic theory and antenna technology, however, the material provided should
be sufficiently detailed to guide the reader from this stage. The review of the basic
properties of antennas contained in Chapter 1 provides an indication of the
technical level required and includes appropriate references to further background
reading.
Finally the editors wish to acknowledge with thanks the many organisations and
institutions for their kind permission to use figures, photographs and other
information.
Chapter 9
Linear arrays
R. C. Hansen
Introduction
The radiated fields from a linear array are a superposition (sum) of the fields
radiated by each element (in the presence of the other elements). The field of an
element is proportional to the excitation parameter, which is current for a dipole,
voltage for a slot, and mode voltage for a multiple-mode element. Self and mutual
impedance (or admittance) relates the applied voltage (current) to the excitation
current (voltage). This Section is concerned with the design of the set of excita-
tions, a complex number for each element.
A common notation will be used throughout this Chapter; it is that most often
used by array papers. The excitation vector will consist of an amplitude and phase
at each element. This discrete distribution is often called an aperture distribution,
where the discrete array is the aperture. The far-field radiation pattern, also called
the space factor, is just the Discrete Fourier Transform of the array excitation.
When the array is assumed to have isotropic elements, this DFT is the sum of exci-
tation coefficients times phase-path exponentials. The overall pattern is the product
of the isolated element pattern and the isotropic array space factor. This is the
'forced excitation' problem: the drive voltages are individually adjusted so that each
excitation current (using dipoles as an example) is exactly the value desired. In
practice many feeds produce a 'free excitation', that is the drive voltages are all
fixed and the excitation currents are those allowed by the 'active' element
impedance. Active implies a variation with observation angle 6. Ways of calculating
active array patterns using active impedances or active element patterns are discus-
sed in Section 9.4 and in Chapter 10. Returning to notation, the element spacing is d
where each d may be different, angles are measured from broadside, X is wave-
length, and the convenient array variable is u:
d n
u = - sin d
A
n = 1 2 3 L . . . N
Uniform (equal) spacing is assumed in what follows. If the co-ordinate system starts
at one end of the array, as in Fig. 9.1, the space factor is, for an N element array:
iV
F(u) = I An exp; 2n(n - l)u (9.1)
Linear arrays 3
Again forced excitation is assumed, and An is the complex excitation, which may
include phasing. Later the phasing terms will be written explicitly for narrow-beam
and for shaped-beam patterns. Also it will be convenient later to move the co-
ordinate centre to the array centre.
The unit circle approach of Schelkunoff135 replaces the exponential by:
w = exp (/ 2m)
Real (physical) space corresponds to a portion of the unit circle in the complex w
plane. The polar angle in the w plane is \jj — 2mi, and the portion of the unit circle
covered is for:
-kd<\p<kd (9.3)
Thus half-wave spacing allows w to traverse the unit circle once, wavelength spacing
twice, etc. The w polynomial has N— 1 roots, but they do not have to lie on the
unit circle.
} main beam
w = 1 region
sidelobe
region
The pattern is given by the product of the distances from the observation point
(w) to each of the zeros (see Fig. 9.2). As the point moves around the circle, lobes
build up then decay, and, as mentioned, zeros located on the circle give pattern
nulls. Zeros off the circle may give pattern minima. The uniform array is an
especially simple case in that:
F{u) = (9.4)
sin nit w-1
4 Linear arrays
Here the exponential term corresponds to the phase at the array centre and the
roots are equally spaced on the circle by 2n/N. This unit circle analysis is exactly
the same as the z-transform, which has gained some currency in circuit analysis in
recent years, z=w. Zero (root) placement will be discussed further in Section
9.1.1.3.
* This implies a linear phase progression over the array, corresponding to main beam angle 0O
Linear arrays 5
to the desired continuous space factor (pattern). If the calculated pattern is still not
sufficiently close to the desired pattern, a perturbation technique may be used. For
example, with a narrow beam pattern with a shaped sidelobe envelope where side-
lobe heights are slightly off, it may be assumed that the position of each sidelobe
peak is unchanged (valid since the peaks are rounded on top). Then each element
excitation may be perturbed and an error signal may be derived from the difference
between each lobe peak in the original pattern and in the pattern produced by the
perturbed excitations. Solution of the resulting equations yields values for the exci-
tation perturbations. Several iterations may be necessary for satisfactory pattern
adjustment.
Before computers
Linear arrays
sin Nrni
F(u) = f (9.5)
sin nu
where now
u = -(sin0 — sin0o)
A
The interelement phase shift is kd sin0o. When this is zero the beam is broadside.
By varying the interelement phase shift the beam position is controlled to produce
a main beam scan. Fig. 9.3 shows the variation in pattern that can be produced by
a two-element array by varying spacing and phase (Southworth141). Before taking
up the behaviour of these canonical uniform arrays when scanned, it is useful to
evaluate broadside performance.
no o Q
(0 (d) (e)
Fig. 9.3 Directivity patterns of 2-element array vs. spacing and phase ( T = 2-n) {courtesy:
Southworth)
Beamwidth: The half-power points on the beam of a uniform array are found
simply by putting sin Nm3/(N sin nu3) = \/o3. The solution of this for various N is
given in Fig. 9.4, where it may be seen that the variation inNu 3 is less than 1% for
N>1. In fact the error is only 5% for N= 3. For large arrays, the half-power
points can be written simply: Nu3 = ± 0.4429. For a beam scanned at angle 0O> this
gives the 3 dB beamwidth 03 (angle between the 3 dB points on the main beam as:
/ 0.443 X\ / 0.443 \\
0$ 2£ arc sin I sin d0 H — arc sin I sin 0o 1
y Nd ] \ Nd ]
or, for TV large,
0.866 X
(9.6)
Nd cos 0A
Linear arrays 7
It is useful here to consider how the beam changes with scan. At broadside the
beam is a rotationally symmetric disc (omitting element patterns) as sketched in
Fig. 9.5. As the beam scans toward endfire, the disc folds forward to make a cone.
0.52 i
0.46-
0.42 -
0.40
6 10 100
number of ele
A singular situation occurs when the 3 dB point gets to 90°. Beyond this scan angle
the beam has two peaks and the 'beamwidth' will double as the outside 3 dB points
are used. Finally at endfire a pencil beam results. Thus a linear array at broadside
yields directivity in one dimension, while at endfire it yields directivity in two
8 Linear arrays
10 20 50 100
Accuracy of this is better than 1% for Ndfk > 2. The beam collapse, where the 3 dB
point is at 90°, occurs for a scan angle of:
0.443X\
0 = arc sin II — (9.8)
and the accuracy of this is better than 1% for Nd/\ > 4. This beamwidth is larger
than the broadside value by 2.14 \/Nd/\. Thus for uniform linear arrays these
formulas are all valid forN> 20; for smaller arrays the value of Nu3 from Fig. 9.4 is
substituted in place of 0.443 in eqn. 9.6 and 03 can be determined for any scan
angle. Fig. 9.6 shows the beam broadening vs. angle. When the inner 3 dB pattern
point reaches 90°, the pattern cone is filling in and the usual definition of beam-
width is not useful.
Sidelobes: Sidelobes of the uniform array are well behaved; nulls of F(u) occur
at u — n/N, n= 1 to N — 1, and the peaks of F(u) occur at u that are solutions of
TVtanrrw = X^nN-nu. For large N, the equation becomes tan7V~7rw =Nnu, and the
Linear arrays 9
first solution is Nu = 1.4303. The sidelobe ratio for large arrays is the same as that
for uniform line sources: 13.26 dB, independent of main beam angle. For smaller
arrays the location (value of u) of the first sidelobe and the sidelobe ratio are shown
in Fig. 9.7 where it appears that arrays of less than 8 elements experience a signifi-
cant sidelobe-level degradation. Of course, the relation of sidelobe peak angles and
array length depends upon main beam position through u. The uniform linear array
has a sidelobe envelope that decays as nu approaches 7r/2, and as mentioned in Sec-
tion 9.1.1.3, this decay is important to maintain low aperture Q and tolerance in-
sensitivity.
47n 15
1 46-
sidelobe ratio, dB •
1 45- 13
12
• sidelobe position Nu
142 10
141
140
10 20 40 60 100
Fig. 9.7
Grating lobes: From the expression for the array pattern (eqn. 9.5) it is
obvious that maxima of unit value occur whenever u = «, an integer, because then
both sine terms are zero and the limit is unity. If d/X and 0O are chosen properly,
only one main beam exists in Visible' space, — 90° < 0 < 90°. Large spacings will
produce one or more additional main beams called grating lobes (GL) (see Fig. 9.8
for a sketch of a typical pattern). Physically the larger spacing allows the wave from
each array element to add in phase at the GL angle. From the condition given above,
u =n9 the location of grating lobes can be easily determined. The GL equation is
usually written as:
10 Linear arrays
(9.10)
X sin 60 — sin 6g
When the main beam is broadside, the first grating lobe is just appearing at 6g =
— 90°, when d = A. A symmetric GL appears at 8g — 90°. Beam scan to 0O = 90°
will produce a GL at —90° when d = A/2; this is the oft quoted rule that half-wave
spacing precludes grating lobes. Actually as d -> A/2 the GL approaches visible space,
with half of the GL visible at d = A/2. Higher-order grating lobes occur when n =
2, 3 or more. Radio-astronomy interferometers use spacings of several wavelengths
or more to get narrow beamwidth with few elements; many grating lobes result.
Fig. 9.9 shows the onset of the first grating lobe for various scan angles (main beam
angles), and GL angles vs. d/X. For example a 30° scan requires a spacing of 0.667A
or less to keep grating lobes at the edge of visible space or beyond.
-90 90
Placement of GL null: It is often desirable to keep all of the grating lobe out of
visible space. One might in principle adjust d/X so the grating lobe at — TT/2 is equal
to the sidelobe envelope near there. Since the sides of the grating lobe are steep,
small changes in frequency or manufacturing tolerances may bring in a sizable por-
tion of the grating lobe. A better scheme puts the pattern zero adjacent to the grat-
ing lobe at — TT/2, SO that the entire grating lobe is comfortably excluded. This can
be done for uniformly illuminated arrays directly, but, as mentioned elsewhere,
there are no universal array distributions with adjustable sidelobe level and Q. Such
arrays are usually designed using continuous distributions such as the Taylor one-
parameter modified sinx/x line source. Thus this distribution will be used here to
find the reduction needed in d/X to put the grating lobe null at —nil* Unlike the
line source pattern, which has no grating lobes, the array pattern (for large d/X) is a
repetition of part of the line source pattern, as shown in Fig. 9.10. The sidelobes
This analysis is due to David Munger of Motorola, Phoenix, AZ, and Richard Phelan of Harris
Corp., Melbourne, Fl
Linear arrays 11
are lowest halfway between the main lobe and grating lobe. Using the Taylor one-
parameter distribution (see Section 9.1.1.3) the part of the pattern adjacent to the
main lobe is given by:
diiin^/u MJ Nd , ^N
F(«) = f
•z « = — (sm 0O ~ sin 6) (9.11)
TT\ U —B* A
where B is the parameter which is selected to determine the sidelobe level. The first
null occurs for ux — \/l + B\ and placing the grating lobe at — TT/2 gives:
Nd
ug = — (sin 0 O + 1) (9.12)
A
12-1
grating lobe angle
11-
10 20 30 40 50 60 80 90
main beam angle 90
Using the grating lobe equation (eqn. 9.10), ug = N. Now the portion of the pattern
adjacent to the grating lobe is given by:
sin 7T\/(Ug — u)2 — B2
F(u) = (9.13)
The null adjacent to the grating lobe is at u = ug — uu and when it is now placed at
-TT/2:
12 Linear arrays
Nd . __ Nd,
8 l 8 (9.14)
X ° X
But from above ug — Ui=N — \ / l + B29 so the modified grating equation becomes:
d_ _ N-^/l+B2 i 1
(9.15)
X N sin 0O + 1
4n\H(do,<t>o)\2\F(uo)\2
(9.17)
Jo J-*a WW* <t>)\2\F(u)\2cosd dd d<t>
For isotropic elements Hid, 0) = I and the integration with respect to 0 may be
performed immediately since the array factor F(u) is rotationally symmetric (Fig.
9.12), giving:
G = (9.17a)
1
J-7T/2
cos Odd
10-
09-
08-
07-
SLR=13dB 20 25 30 35^0
0-5-
03
6 10 20 30 L0 60 100
number of element^ N
In the case of a uniform broadside array F(u) is symmetric and is given by equation
9.5, so that
N2
G = (9.17b)
cos0 dd
14 Linear arrays
Changing the variable to u gives:
N2d/X
r
G = (9.18)
sin2N7tu
2
sin 7rw
Jo
This can be integrated with the help of an expansion (Whittaker and Watson176):
I^d/X _ rd,
: £ (/V — «)cos 2nmt du
G ~ Jo n=l J
NdMN^\
__ — + — KT
2* {N — n) sine nkd (9.19)
A A n=l
The peak directivity occurs just before the onset of the grating lobe and the minor
oscillations below the peak depend upon whether the boundary of real space
(sin# = 1) coincides with a peak sidelobe or with a zero. This behaviour can also be
interpreted as the effect of the mutual coupling between elements varying as the
spacing is varied. In fact, the summation in eq. 9.20 can be recognised as a weighted
Linear arrays 15
summation of the mutual resistances, since the normalised mutual resistance
between a pair of isotropic radiators spaced by 5 is simply sine (ks).
At this stage, it is worth recalling that the assumption underlying the curves shown
is that the currents in all elements have been forced to be equal. Achievement of
10 12
equal currents depends upon the design of the feeding network. Simply connecting
all the elements in parallel will not achieve equal currents, since the end elements
'see' a different mutual impedance than do the centre elements. The feeding network
has to allow for the effect of self-reactance and mutual reactance as well as the
mutual resistance of the elements.
16 Linear arrays
For a scan off broadside, the directivity result is:
N2
G = spi (9.21)
N + 2 £ (iV — «) sine wfaf cos («M sin 0O)
N
G = N.x (9.22)
1+- £(#-«)sine2n*rf
N n=i
This duplicates the broadside array results, but with spacing halved. Thus G = TV for
d/X = 0.25, 0.5, 0.75, . . . The beam at endfire as mentioned earlier is a pencil
beam, but the beamwidth is much broader than the broadside disc beam. Using the
general formula with half-wave spacing inserted, it is interesting to see how the
effective aperture (directivity divided by broadside directivity N) varies with scan
angle. Since the sine argument is nn, the directivity is N for any scan angle, and the
effective aperture is unity. This is in contrast to a uniformly excited continuous line
source (Hansen69) which has an endfire directivity of 4L/X, but a broadside direc-
tivity of 2Z//X. (Although the endfire 3 dB beamwidth is again much broader than
the broadside beamwidth, the total solid angle is smaller, cf. Fig. 9.5.)
For N large, maximum directivity occurs with a spacing of just under X for
broadside operation and just under X/2 for endfire operation, with a value
approaching 2N. Expressed in terms of aperture length L = (N—l)d, it is evident
that these maximum directivities approach the same values as those of the continu-
ous line source, viz. 2L/X at broadside and 4L/X at endfire.
Turning now to arrays of actual elements such as short dipoles or half-wave di-
poles (slots have similar patterns), the overall radiation pattern is given by the
element pattern multiplied by the array-factor pattern (isotropic elements), but the
overall directivity is not the product of the element and array directivities.
Parallel short dipoles provide directivity in the 0 dimension and the appropriate
element pattern to use in eqn. 9.17 is:
>) = V I ~ cosz0 sin> (9.23)
whilst collinear short dipoles provide directivity in the plane of the array with an
element pattern:
The analysis proceeds along the same lines as before. The resulting directivity for a
uniform broadside array of parallel short dipoles is:
Linear arrays 17
G = 3N/2 ,923b)
. . 3*L* \sinnkd cos nkd sin nkd]
N
For both parallel and collinear short dipoles the mutual resistance is given by the
bracketed trig terms times a constant. Short elements are seldom used in arrays
because of their high reactance (susceptance) and narrow bandwidth; half-wave or
resonant elements are usually preferred. One might infer that directivity of half-
wave dipole arrays could be calculated using mutual resistance, and this proves to
be the case. The directivity will be derived for a uniform broadside array of half-
wave dipoles in any orientation. Directivity is the ratio of peak field power density
to input power:
G = "-^ (9.27)
where the dipole currents (excitation coefficients) are In. The numerator is then,
for uniform excitation (/„ = / 0 for all n):
IN \2
£ // JJ == UQN2^
num = 120 (( £ (9.29)
collinear case exhibits less grating-lobe directivity drop around d = A, but the peak
value is also lower. This is because the mutual coupling effect is greater with parallel
dipoles. The parallel array gives roughly 3 dB more directivity at the peak, due to
narrowing the beam in the transverse plane.
For fixed array spacing and number of elements, and for uniform amplitude of
excitation, the maximum directivity does not occur at broadside or endfire. This
case is included in Section 9.1.3 on superdirective arrays.
20 Linear arrays
An array with tapered excitation has a space factor that can be written as:
Inserting this into eqn. 9.17J and integrating gives, for a tapered array of isotropic
elements:
' N
G
=
lm=l
Here it is assumed that the coefficients are real, but this restriction is removed
simply by using AnA*m,
When all ^4n = 1 this reduces to eqn. 9.20. If the spacing is d - A/2, the direc-
tivity becomes:
G = i '- (9.34)
Symmetric arrays allow eqn. 9.33 to be written with sums going to N/2 (oxN/2 +
1). In this case two sine terms are involved. As an example, the result for TV even is
shown:
G = (9.35)
N/2 N/2
Z L^n^m[ s i nc ( w + m
~ l)kd 4- sinc(« —rri)kd]
n = im = l
However, a direct correspondence between this polynomial and the array poly-
nomial is not feasible, because the main beam must be symmetric and have zero
slope at its centre. The TV-element array has N— 1 zeros while the Nth order
Chebyshev polynomial has N zeros. Thus an N— 1 degree Chebyshev is used. Part
of the x > 1 region is mapped onto one side of the main beam while the oscillatory
portion of the Chebyshev is mapped out once onto the sidelobes on one side of the
main beam. The transformation from T^_l{x) to space factor F(\p), with i// =
kd sin 6, is x = XQCOS (I/ / /2). The voltage sidelobe ratio is given by:
SLR = (SLR>1) (9.37)
or inversely,
arc cosh (SLR)
x0 = cosh
N-l
and the space factor is simply
FOJO = 7WJOCOS0/2) (9.37a)
As B varies from —7r/2 to 7r/2, X varies from (x 0 cos nd/X) up to x0 at 6 - 0 and then
back to (x0co$ird/\). For half-wavelength spacing, the minimum value ofx is zero.
The maximum possible spacing is determined by the need to prevent* falling below
— 1. This gives the maximum element spacing for a broadside array as
22 Linear arrays
= arccos(-l/* 0 )
n
which approaches one-wavelength spacing for large arrays.
The excitation An of the nth element in the array may be obtained by choosing
the centre of the array as the phase reference and writing the space factor as
JV
Fty) = I Anew\j(2n-N- 1)^/2] (9.37c)
Since this Fourier series has a finite number of terms, its coefficients may be
obtained from the inverse relationship:
An=Z
4M \^M2-mTn{xn)[TM{axn + b)^(-\)nTM{b-axn)] (9.42)
where et = 1 for / = 0 and et — 2 for i > Q;xn = cos wnjM. The integers Mx and Af2
are, respectively, the integer parts of Af/2 and (M+ l)/2. This result is valid for
d < X/2. Small spacings (highly superdirective arrays) may require double precision
arithmetic due to the subtraction of terms. Many arrays are half-wave spaced; for
these the a and b of eqn. 9.40 reduce to:
^ , b = ^ (9.43)
For half-wave spacing the two approaches give identical results! In fact, due to the
properties of the Chebyshev polynomial, the two space factors, in precursor form,
are equal:
It was mentioned earlier that the Chebyshev distribution tended toward large peaks
at the array ends. To give an idea of the array lengths and sidelobe ratios for which
this occurs, the sidelobe ratio for the centre and end excitations equal is given
versus N, and also the value that ensures a monotonic distribution, in Table 9.1. For
a 20-element array, the centre equal-end situation occurs for SLR greater than
20 dB. Arrays larger than 28 elements do not allow a monotonic distribution at all
for SLR up to 40 dB. Fig. 9.17 shows array amplitudes for SLR = 20, 30 and
40 dB. Adjacent element amplitudes are connected by straight lines; but these
distributions are typical for Chebyshev arrays. In practice, Chebyshev distributions
are not often used, as the Taylor h distribution (see a later Section) offers more
flexibility.
SLR = 20dB
centre
array
Du Hamel43 extended the Chebyshev design principle to endfire arrays, but only
for d < X/2. In fact, to avoid a back lobe, spacing is customarily made < A/4. To do
this for any scan angle, \j/ is modified as usual to:
Linear arrays 25
b = (V^TT + f C ° S ^ (9.47)
2 sin kd
Drane42 has shown that a large scanned Chebyshev linear array has a directivity
independent of scan, for any spacings between half and full wave, for that scan
range for which only one main beam exists (no grating lobes). Directivity can be
calculated using the formulas of Section 9.1.1.2. Beamwidth is best calculated
directly from the space factor using a root finder such as Wegstein, with F(w3) =
F(0)/\/2. Approximate values may be obtained from calculations of Drane;42 Figs.
9.18 show directivity and beamwidth vs. SLR and L[\. It is interesting to note that
for each L/X there is an SLR giving maximum directivity. Smaller SLRs (higher
sidelobes) represent significant power contained in the equal-level sidelobes, while
larger SLRs yield lower efficiency due to the wider beam associated with lower
sidelobes.
That the sidelobe envelope does not decay for large u is indicative of trouble, and
the aperture distribution corresponding to F(u) has a singularity (infinite value) at
each end; thus the appellation 'ideal'. Taylor invented a clever solution to this
dilemma. Close-in zeros should maintain their spacings to keep the close-in side-
lobes suitably low, thus keeping efficiency high. However, to reduce aperture
26 Linear arrays
energy storage Q, the sidelobe envelope for large u should taper as 1/w. To combine
these, the u scale is stretched slightly by a factor o slightly greater than unity, thus
maintaining close-in zero locations. At some point a zero can be made to fall at an
2OOO
1000
10 100 1000
sideiobe voltage ratio
10 100 1000
sidelobe voltage ratio
Fig. 9.18 Chebpshev array performance vs. SLR and LA. After Drane
a Directivity
b Beamwidth
integer; from this transition point the zeros occur at ±n. The pattern then has n
roughly equal sidelobes; beyond u — n the sidelobe envelope decays as 1/w.
Actually the first n lobes are not precisely equal; the transition region allows some
Linear arrays 27
decay of the sidelobes near n, and the envelope for u > ft is slightly different from
\ju there. Fig. 9.19 shows a typical pattern for SLR = 25 dB, and h = 5.
The zeros un become:
zn = ± (9.51)
Fiu) = ^ ri (9.52)
nu n=i
Beamwidth is very closely given by oul3i where ul3 is the 'ideal' beamwidth. Table
9.2 shows the^l, w3 and o for various sidelobe ratios and values of n.
SLR A u3 n=2 4 6 8 10
dB 0
where the aperture variable p is zero at the centre and ± 1 at the ends. The coef-
ficient is:
\(n-\V]2 "-1 n2
F(n9A,n) = , . / ; w - 1—TT n i - 4 - (9-55>
(n — 1 +n)l(n — 1 —ri)\ m = i z^
Tables of aperture distribution and coefflcents are given for SLR = 20(5)40 dB and
« = 3(1)10 by Hansen.69 More extensive tables have been calculated by Spell-
mire.142
It is necessary to discuss the appropriate range of values of n. Too large a value
of h will give a non-monotonic aperture distribution and may even produce a large
peak at the ends of the aperture. On the other hand, too small a value will not allow
the transition-zone zeros to behave properly. Fig. 9.20 shows distributions for the
largest values of n that allow a monotonic distribution: SLR = 25 dB, ft = 5; 30dB,
n = 7; and 40 dB, h — 11. A comparison of maximum efficiency and monotonic
distributions for SLR = 25dB is given in Fig. 9.21, where it can be seen that
n = 12 distribution is not only peaked at the end but has small oscillations on
either side of the dip. The difference in aperture efficiency, as will be shown next,
is only 1.5%.
The Taylor n distribution is widely used because it gives slightly better efficiency
and beamwidth than the Taylor one-parameter distribution, for the same sidelobe
level.
Linear arrays 29
Note that this equation, which is often used, is valid only for distributions with
linear phase (Friis and Lewis58). It is not correct for difference-pattern distributions
for example. Using eqn. 9.54 for the distribution, the numerator is unity since the
cosine terms integrate to zero. In the denominator the product #2 before integration
gives unity plus two single series plus a double series. Again, each term of the single
series integrates to zero. The double-series integral is zero for n i= m, and gives \ for
n = m. Thus the excitation efficiency becomes:
1
T? = (9.57)
n-l
where the F(n9A9ri) coefficients are given by eqn. 9.55. Even large values of n
allow easy calculation of r\. There is, for eachSLR, a value of n that gives maximum
17. This is because, as it increases from a small value, the main beam energy decreases
30 Linear arrays
faster than the sidelobe energy increases. For larger n, however, the main beam
changes more slowly. Table 9.3 gives these peak efficiency values of n. These distri-
butions are not monotonic, and have significant edge excitation. Also shown in the
Table are the largest values of n that give monotonic distributions and the corre-
sponding efficiencies. It may be seen that the efficiency penalty is small.
max efficiency, n = 12
max monotonic, n r 5
(9.58)
n i+- 2
n=l n
This can be more simply written as:
(9.59)
where the constant C has become sinh TTB/TTB. Thus this pattern is a modified
sinjc/x pattern; the single parameter B controls all characteristics: beamwidth,
efficiency, sidelobe level etc. A transition from the sine pattern to the hyperbolic
form occurs at u — B, on the side of the main beam. For smaller u the hyperbolic
form provides the central part of the main beam, with the peak value (u = 0) of
sinh-nB/irB. The sidelobe ratio is immediately that of the sinx/x times the beam
peak value. In dB,
fF(0)]2 (9.62)
2
j_*[F(u)] du
Due to the rapid sidelobe decay with u, the integral may be approximated with
infinite limits even for modest-length apertures. This allows the integration to be
reduced to a tabulated integral (Rothman;130 Abramowitz and Stegun1).
2sinh27r£ , ^x
(963)
)
where Jo is the integral of IQ from 0 to the argument. All characteristics of the one-
parameter distribution are thus readily available.
Linear arrays 33
Table 9.4 Taylor one-parameter line source characteristics
E 0.4
h M
-U> (9.64,
Table 9.5 gives values of normalised beamwidth for various SLR and L/X. For long
sources normalised beamwidth is independent of L/X, and for this case:
^ * 2a 3 (9.65)
34 Linear arrays
SLR 00 L/X
dB
3 10 30 100
13.26 0.4429 0.8891 0.8862 0.8859 0.8859
15 0.4615 0.9267 0.9233 0.9230 0.9230
20 0.5119 1.0288 1.0242 1.0238 1.0238
25 0.5580 1.1225 1.1165 1.1160 1.1159
30 0.6002 1.2085 1.2011 1.2004 1.2004
35 0.6391 1.2880 1.2790 1.2782 1.2781
40 0.6753 1.3622 1.3515 1.3506 1.3505
1-5
13
12
1-1
10
09
0-8
0-7
0-6
0-5
10 15 20 25 30 35
sidelobe r a t i o . dB
Fig. 9.24 Normalised beamwidth vs. sidelobe ratio for linear one-parameter source
An SLR of 40 dB, for example, gives a beamwidth about 50% larger. Fig. 9.24
shows normalised beamwidth for sources 3 and 30 wavelengths long.
A comparison of Taylor h = 5 and Taylor one-parameter distributions for 25 dB
SLR is interesting. Figs. 9.19 and 9.22 show the space factors; the far-out sidelobes
of the one-parameter are roughly 5 dB lower. Excitation efficiency is, as expected,
higher for the h: 0.91 for n = 5 and 0.85 for one-parameter. Fig. 9.25 shows the
distributions. Surprisingly, one may note that these differ only in the 'tails', where
Linear arrays 35
the h has a much higher pedestal. Thus, the n distribution should be used where
efficiency (directivity) is crucial, and the one-parameter distribution should be used
where far-out sidelobes are important.
taylor n = 5
center end
Beam Efficiency: For radiometer and some mapping applications, the fraction
of energy radiated in the main beam, null-to-null, is important. This quantity is
called the beam efficiency r?b. This Section determines how beam efficiency varies
with other design parameters such as sidelobe ratios, beamwidth, taper efficiency,
edge illumination etc. The vehicle for doing this is the one-parameter line source
developed by Taylor. The geometry is sketched in Fig. 9.26 where the line source
is along the z axis in spherical co-ordinates. The narrow pattern is rotationally sym-
metric about z, giving an integrand factor of cos0. Using the one-parameter space
factor F(u) developed previously, the beam efficiency is:
n
|F| 2 cos0tf0
(9.66)
Normally the evaluation of this would require numerical integration over all the
pattern sidelobes, an expensive and relatively inaccurate task. However, the taper
efficiency rjt, which is the ratio of directivity to directivity of the same source uni-
formly illuminated, can be accurately evaluated. These values can be used to evalu-
ate the denominator of the expression, leaving only a numerical integration over the
main beam. The latter is a simple and accurate task. The taper efficiency is:
36 Linear arrays
FIX sinhTri?
Vt = n/2
Fa = (9.67)
rn/2 nB
2L \F\2cosdd6
Jo
This gives:
where the first null occurs at 0x. The dependence upon L/X can be removed for long
sources. ForX > X, dt < 1 and
lFldtt (9.69)
* FS J o >
Figure 9.27 and Table 9.6 give the values of beam efficiency vs. sidelobe ratio.
Other parameters are found readily from the value of B.
SLR B 7]f Vb
dB
13.26 0 1 0.9028
15 0.3558 0.9931 0.9364
20 0.7386 0.9329 0.9820
25 1.0229 0.8626 0.9950
30 1.2762 0.8014 0.9986
35 1.5136 0.7509 0.9996
40 1.7415 0.7090 0.9999
0-9-
0-8
20 25 30 35
sidelobe ratio, dB
These excellent results reinforce the concept of designing space factors by proper
placement of zeros.
sin m/m of eqn. 9.59. The parameter C now controls initial sidelobe level, while v
controls envelope decay. Note that these two-parameters are independent, unlike
the obsolete multiple-parameter distributions mentioned earlier. The Bickmore-
Spellmire aperture distribution is:
-40
0° 30° 60° 90° 120° 150° 180°
angle from endfire
-10
-20
-30
-40
0° 30° 60° 90° 120° 150° 180°
angle from endfire
g(p) = (9.71)
Linear arrays 39
Aperture end singularities occur for v < \, while v > \ gives envelope tapers more
rapid than 1/w, but commensurately less efficient.
Low-sidelobe designs: In the last decade low sidelobes (roughly —30 to — 60 dB)
have become of interest for several reasons: increasing spectrum congestion in
broadcast and in satellite transmissions, desirability of reducing radar intercept
probabilities, and reduction of radar clutter/jammer vulnerability. From the
previous Sections one may infer that a good low sidelobe distribution should be
heavily tapered in amplitude. Examination of the one-parameter space factor shows
very low pedestals for 40-60 dB SLR; from Taylor's observation that distributions
with pedestals are more efficient, the n space factor might appear more attractive.
This is indeed the case. To design a high-performance low-sidelobe space factor,
again the emphasis should be on the pattern zeros. Here the Taylor h space factor
is almost uniquely properly designed. Recapitulating, close-in zeros are adjusted to
obtain a few nearly equal sidelobes at the design sidelobe level, while farther-out
zeros are properly placed to give a \\u envelope. One may contrast this with a
Hamming distribution, which is a cosine of a doubled argument, on a pedestal. The
argument produces a zero contribution from the cosine at the centre of the half
aperture. This distribution is (Blackman and Tukey16):
g(p) = 0.54 + 0.46 cosTrp = a + b cosnp (9.72)
resulting in a centre value of unity and a pedestal (end value) of 0.08. Excitation
efficiency of the Hamming is:
= °- 7338
and the half-power beamwidth is u3= 0.6515. The corresponding space factor has
zeros at u = 2, 3, 4, . . . plus another zero at u = yja/(a — b) = 2.5981. The space
factor is:
= [(abWaUinnu
nau(u2- 1)
The close spacing of the first two zeros produces an irregular convex sidelobe enve-
lope where the highest sidelobe is the fourth at — 42.7 dB. First and second side-
lobes are —44.0dB and — 56 dB. Moving the salient zero to 2.5 to centre it does
lower the first and raise the second, but the overall SLR is now 1 dB worse.
A comparison of several low sidelobe distributions is illuminating. Take SLR =
42.7 dB for an example, since this is the Hamming value. Table 9.7 gives the first
ten pattern zeros for the Taylor one-parameter, the Hamming, and Taylor n = 6
and h = 10. All the Taylor distributions exhibit a relatively smooth progression; the
Hamming has the salient zero at u = 2.6. This is better seen in Table 9.8, where the
spacings between adjacent zeros are shown. Clearly, the Hamming is wasting zero
spacing on the first several sidelobes. In this Table can also be seen the effect of too
large an n for the SLR. n = 10 gives a slight oscillation in spacing. Finally Table 9.9
40 Linear arrays
Table 9.7 Zeros for SLR = 42.7 dB
n Taylor Taylor Hamming
one-parameter
fl = 6 « = 10
1 2.112 1.894 1.897 2
2 2.732 2.398 2.396 2.598
3 3.550 3.173 3.166 3
4 4.412 4.069 4.056 4
5 5.335 5.020 5.002 5
6 6.282 6 5.978 6
7 7.243 7 6.970 7
8 8.214 8 7.974 8
9 9.190 9 8.984 9
10 10.172 10 10 10
gives beamwidths, efficiencies and products for these cases. As expected, the Taylor
h is the best. A choice of h can be made on the basis of aperture distribution shape
and pattern. The h pattern also is convex, but less so. Too large an n will also
increase the excitation tolerance requirements.
Average sidelobe level: In the last several years there have been new restrictions
on antenna sidelobes, from the standpoint of personnel hazards in the path of high-
power transmitting antennas, and to reduce interference between closely located
antennas. Thus both the sidelobe envelope and the average sidelobe level are
important antenna design parameters. It is desirable to be able to estimate the aver-
age sidelobe level for an antenna, and to know what sidelobe envelopes are reason-
able. This can be done as the average sidelobe level and envelope are approximately
related to the antenna directivity. This approximate relationship will be derived for
the Taylor one-parameter source. Since most antennas (linear) have a sidelobe
envelope that asymptotically decays as 1/w, the Taylor one-parameter source is an
excellent model. The average power sidelobe level is defined here as the integral
from the first null to TT/2:
L/K d
- _A_ [ "
where the first null is at u0 = \/l +B2. For long sources, L/\ > B, and
SLL Sl l n (976)
- ^ "^f
The actual sidelobe power is less because the sidelobes fit under the envelope.
Taking a sidelobe (power) shape of sin2?rw, independent of envelope, gives an
integral of 1/2. So the sidelobe power is 3 dB below the envelope. If the directivity
is G = 2T?Z,/A, with 77 the taper efficiency, the average sidelobe level may be written
in terms of the directivity:
SLLave
1 2i (9 77)
~ n^G W '
For a given initial SLR, B is fixed, and the right term can be calculated. This term is
shown in Fig. 9.29 vs. SLR and it can be noted that it varies from 0 to 2.7 dB for
SLR from 13 to 40 dB. The average SLR is then:
i? iIn
SLRave ^ G ^ + (SLR - 13.26 dB) - 10log ^ + 9.95 dB
2/
(9.78)
42 Linear arrays
Thus the average sidelobe ratio is 10 to 34 dB lower than the directivity (in minus
dB).
Edge pattern amplitude, at 6 = 90°, is also of interest. For large L/X the edge
field is simply X/nF0L, or in terms of directivity:
27?
(9.79)
-2-
dB
-3-
-5
10 20 30 z.0 50 60
SLR.dB
9.1.2.1 Shaped beams: Shaped beams are usually used to cover an area; exam-
ples are the radar antenna with cosecant elevation pattern to compensate reduction
of range with elevation, or the satellite antenna pattern shaped to fit a given geo-
graphical area. Almost always it is the pattern amplitude that is important; the
phase is not prescribed. Assuming a phase distribution may require an inefficient
distribution, so only the amplitude (or power) pattern should be synthesised. A
series of orthogonal functions could be used, e.g. Fourier series, but many terms are
required as the functions do not fit the array physics gracefully (Jordan and
Balmain87). Weighting functions have been used to improve convergence for shaped
Linear arrays 43
beams (Bricout21) and for pencil beams (Jaeckle84). Chebyshev polynomials have
also been adopted for sector-pattern synthesis (Evans50). Where maximum flatness
is important the Butterworth-function method of Ksienski97 may be used. For
pencil-beam patterns the zero placement technique described earlier is preferable. A
better way of handling shaped beams analytically is by using orthogonal beams, fit-
ting a series of these beams to the desired shaped beam. The Woodward and Lawson
technique,180'181 although developed for continuous apertures, is equally useful for
arrays. Each beam is formed from a uniform-amplitude and linear-phase distribu-
tion, where the nth beam is given by sinNm/N sin m, where the array hasiV ele-
ments and u — (d/X) (sin0 — sin# n ). Here the array spacing is d and the beam peak
is at Bn. If the beams are spaced apart (in sin#) by nX/Nd, then all beams except the
nth will have zeros at the nth beam peak. This orthogonality allows the desired
pattern to be 'sampled' at approximately NX/2d points. For example, a half-wave
spaced array has zeros in sin 6 at 2n/N. A ten-element array can have ten beams
positioned, one at each null. Since the beams are orthogonal, the desired pattern at
each sampling point is exactly equal to the amplitude of the beam that peaks at
that angle, giving the amplitude immediately. The phase tilt for that beam corre-
sponds to the beam peak angle. And the total array distribution is simply the sum
of these for each constituent beam. If the desired pattern is F(w), and if the beam
peaks are at un and of amplitude F(un), the actual pattern is simply:
The amount of ripple will be controlled by the element spacing. Fig. 930a from
Woodward and Lawson181 is an example of a cosecant synthesis using 21 beams
(a 21-element array at half-wave spacing) where one zero is placed at Q = 0. Smaller
ripples will be obtained if the zeros are shifted to allow the cosecant drop-off to
cross the 0 = 0 axis; (see Fig. 930b). The element-spacing tradeoff involves pattern
ripple; larger spacing means fewer sampling points, but, more important, spacing
larger than half-wave will allow grating lobes (see Section 9.1.1) to appear if the
desired pattern must be sampled at wide angles. Spacing smaller than half-wave gets
into the realm of superdirectivity (see Section 9.1.3) and is generally undesirable.
Half-wave spacing, then, is a good choice for cosecant-type patterns.
The Woodward-Lawson synthesis used constituent beams that were derived
from a uniform distribution, giving for long arrays a —13.3 dB sidelobe level. It is
possible to start with the zeros of a low sidelobe distribution, then to move the
zeros off the unit circle (in or out) to shape the beam.
However, the orthogonal beam spacing here is larger; so the cost of achieving
lower sidelobes (and higher directivity) on the opposite side of the cosecant
pattern, for example, is larger pattern ripple or a larger aperture.
Since analytical synthesis does not control pattern ripple in the desired region,
and does not allow sidelobes outside the desired region to be independently speci-
fied, numerical methods have been developed which allow optimisation of both
features. In fact, it is not necessary that the pattern exhibit zeros (nulls). The array
44 Linear arrays
180°
amplitude phase
(a)
csc6
-10 -0-5
180°
amplitude 90°
-5 -3 -1
y/x 90 (
-180
- 5 - 3 - 1 0 1 3 5 phase
y/X
1-0J
(b)
csc8
-1-0 -0-5 0 05 10
sinG
— Fd(u)
— Fd(u)
(b)
— Fd(u)
(c)
0i
A
-10
l\
[
-20
db
AHA
1 I
-30
flAn/w
if
-40
Rfi
-U-12-10-8 -6 -L -2 0 2 4 6 8 10 12 U
z
1-0
0-8
T3
I 0-6
4, 0-4
>
? 0-2
-TT TT
-n 2 p=0 2
TT
20n
argf (p) phase
20
When large arrays are used a line-source distribution can be sampled to obtain
element excitation values. For small arrays, the sidelobe envelope shaping must be
done directly on the array polynomial zeros, TV— 1 of them for an TV element array.
48 Linear arrays
These are complex zeros in the variable w = exp (j2nu). The complex coefficients
of this polynomial are just the element excitations. Newton-Raphson iteration is
again used to adjust each sidelobe to the desired peak value by perturbing the com-
plex zeros. However, only the sidelobe magnitudes can be specified, so the
complex-array polynomial must be normalised so that it is real for N odd (for N
even the complex polynomial is imaginary). Using this method questions of sampl-
ing accuracy are obviated, with excellent agreement-with actual arrays (Elliott46).
Another application of sidelobe envelope shaping through zero placement is the
production of a broad null, a few sidelobes wide. To accomplish this, the sidelobes
around the null portion are progressively lowered to simulate a broad null
[Tseng163]. Fig. 9.33 shows the modified pattern along with the starting 40 dB
Taylor space factor.
exp(—jkri sin0)
F = (9.83)
where rn is the distance from the reference point to the /tth element. For a uni-
formly spaced array, rn~(ji— \)d. Now define matrices^ and B, where A is:
\A\ = \F*)(F\ (9.84)
and the elements of B are:
Linear arrays 49
Fig. 9.33 Broadsidelobe null (solid); 40dB Taylor (dashed) {courtesy: Tseng)
a Space factor
b Distribution
50 Linear arrays
The pattern of the zth element isft. Isotropic elements and a uniformly spaced array
allow simplifications of A and B. The elements then become:
(9.86)
Bnm = sine (n — m) kd
Now the directivity Q and sensitivity can be written in abbreviated form:
The tolerance sensitivity S is defined as the ratio of variance of peak field strength
produced by errors of variance o\ (Uzsoky and Solymar168):
(AE)2 oj(J\J*)
" 2 2 2
Oj OJ OJ
It may be noticed that Q = SG. The directivity is a ratio of two Hermitian quad-
ratic forms,^ with B positive definite and A at least positive semi-definite. Thus all
the eigenvalues of the associated equation
\A\J) = Gmax\B\J) (9.89)
are zero or positive real. Since A is a single-term dyad there is one non-zero eigen-
value. The eigenvector is found to be:
\Jmax) = \B\~lF) (9.90)
The corresponding maximum directivity is given by:
Gmax = (FlBr'F*) (9.91)
These compact results are the same as those developed in Section 9.1.3.3.
U W r >
(9.92)
where ^ o is the specified value of the constraint. The variation of this expression
is now taken with regard to |/> and A, and set to zero. The result is:
|/> = q\KYl (9.95)
The matrix AT is:
|tf| = \B\+p$f\D\-p\C\ (9.96)
and the constants p and q are given by:
p = KU\B\J*)2U\A\J*rl{J\D\J*Tl
q = (J\B\J*)(J\A\J*Tl(F\J) (9.97)
with \A\ = \F*)(F\ as before. When these are substituted into the ^ equation the
result is the characteristic equation in p:
<F\K\'l( %f*\D\ - \C\)\KYlF*) = 0 (9.98)
The equation indicates that (F\ is orthogonal to the vector:
\K\~\ Mi\D\ - \Q)\KrlF*) (9.99)
A complete set of functions can be constructed with F as its first element. If
F= [/i,/ 3 ,...,/ w ] (9-100)
105
then the set is simply (Lo et al. ):
F, = [/I,/„ • • • ,fN]
= [-7,7,0,...,0
h h
52 Linear arrays
1
(9.101)
Now the vector (eqn. 9.99) can be expanded in the functions F2 through FN:
N
l
\K\-\ ^0\D\-\C\)\K\~ Ft) = £ ht\Ft) (9.102)
i=2
This can be written as a matrix times an a vector, and the determinant of the
matrix set to zero. The resulting polynomial in p has one root which can be found
by a rooter such as Newton or Wegstein. A faster procedure, since rooters only
work properly within a narrow range of starting values, is to separate the matrix
real and imaginary parts (Winkler and Schwartz 178 ) into a double-size matrix where
eqn. 9.102 is written as
Wllr-Wm h \r
Wl2i~Wl2r hH
W21r-W21i • h2r = 0
W2U-W21r hti
\V\X> = -IX]
P
where V is a 4N x 4N real matrix given by:
0 /
W\ = (9.105)
/ is the identity matrix as usual. Numerically it is much easier and faster to deter-
mine the eigenvalues of a 4N x 4/V real matrix than to find the roots of aniV-degree
polynomial. The set of values of p (the eigenvalues) must now be tested to find
which one maximises G subject to <3f= J%^. The process of determining the ps
and finding the correct one is lengthy but straightforward. Since most arrays to be
Linear arrays 53
optimised are not large, the matrix inversion and eigensolution processes are not
burdensome.
The generalised constraint (eqn. 9.94) can utilise Q, the supergain ratio, toler-
ance sensitivity S, sidelobe envelope etc. The sidelobe envelope constraint is
handled by defining a desired sidelobe-level vector (S\ which relates the amplitude
of each sidelobe to the beam peak:
(J\Fp(SL\ = [</|Ff>, U\F$\ . . . , </|F£>] (9.106)
To find the angles corresponding to the sidelobe peaks, values from the closest
canonical distribution are used. After the optimisation it may be necessary to re-
determine sidelobe peak angles and to re-run the optimisation. The general con-
straint matrix is now used with ^ = 0 .
Minimax optimisation has also been used (Mucci et a/.112), but these algorithms
are usually extremely slow when the number of variables is more than a few. Also
the least Pth gradient method is nearly equivalent and much faster as previously
mentioned.
Directivity is the ratio of peak field intensity at any far field radius from the antenna to the
integral of field intensity over a sphere of that radius.
f See Bloch et al.17 for a list of early references.
Linear arrays 55
74
monotonic phase function was accomplished by Hansen and Woodyard. Another
early contributor was Franz.56 Schelkunoff,135 in a classic paper on linear arrays,
discussed, among other topics, array spacings less than A/2, showing how equal
spacing of the array polynomial zeros over that portion of the unit circle repre-
sented by the spacing gives superdirectivity. The field received wide attention when
La Paz and Miller100 purported to show that a given aperture would allow a maxi-
mum directivity, and when Bouwkamp and De Bruijn19 showed that they had made
an error and that there was no limit on theoretical directivity. Thus the important
theorem: a fixed aperture size can achieve (in theory) any desired directivity value.
This theorem is now widely recognised, but the practical implications are less well
known. Bloch, Medhurst and Poole17 say that the theorem has been rediscovered
several times; the practical limitations of superdirectivity occur as a surprise to sys-
tems engineers and others year after year! In 1946, a burst of wartime research
reporting occurred. Reid123 generalised the Hansen-Woodyard endfire superdirec-
tivity to include an element pattern. Uzkov167 derived the endfire directivity as
d -> 0. And Dolph39 invented the widely used Dolph-Chebyshev array distribution
wherein the equal-level oscillations of a Chebyshev polynomial are used to produce
an array pattern with equal-level sidelobes (see Section 9.1.1). To follow this last
development, Riblet127 developed Dolph-Chebyshev arrays for spacing below A/2,
i.e. superdirective. DuHamel43 and Stegen146 developed complementary advances
in the computation of Dolph-Chebyshev coefficients and directivity. Chu30
developed fundamental limitations on electrically small antennas, deriving Q and
directivity from a consideration of spherical modes. Maximum directivity for an
array with fixed spacings was derived, for acoustic arrays, by Pritchard.120
I (2« + l)Qn
Q (9U0)
* "'W + 2)
For electrically small entannas or arrays, we are concerned primarily with one
mode, for which the overall Q is half the mode Qn (equal power in TE and TM
modes)* The single mode antenna is given by:
which is plotted in Fig. 9.34. Note that for k < 1, the Q varies inversely as the cube
of ka or array length in wavelengths. This result corresponds to a single antenna
element. Exciting the element equally with electric and magnetic mode halves the Q
for a given ka. Also shown are curves with efficiencies below unity. Since the loss is
just a resistance in series with the radiation resistance, its effect on Q is apparent.
From these curves then it can be seen that a single mode element (as most are)
cannot be much below half wave in size for an octave bandwidth (Q = 1.414). And
small elements will have a narrow bandwidth especially since most elements do not
use the fundamental spherical volume efficiently. Size reduction of elements is thus
usually impractical.
Bandwidth is derived from Q by assuming that the antenna is a resonant circuit
with fixed values. Then bandwidth is:
Maximum directivity for uniform amplitude, fixed spacing: In 1938 Hansen and
Woodyard74 developed an endfire line source with modest superdirectivity. This is
of interest because the distribution can be sampled to get array excitations and
because the amplitude is constant, a feature that is attractive for arrays. They
observed that, if the free-space phase progression along the array were increased,
the space-factor power integral decreased faster than the peak value; thus the direc-
tivity increases up to a point. The endfire pattern is:
where 0 is now measured from endfire and j3 is the wave number over the aperture.
Inverse directivity is proportional to:
Linear arrays 57
1
— oc *- s f sine2 -(fccos0—0) sinBdd
G 2
sine -<k-8)\
-(* ° L J
1 l^.o^. cos0-1
(9.114)
2
Here Si is the sine integral, and <f> = L(k — p)9 the additional phase along the aper-
ture (in addition to the progressive endfire phase). Maximum directivity was deter-
mined to occur for $ — 2.922 radians. In many books it is carelessly stated that n
extra radians of phase are needed, but there is no physical reason for this; a better
approximation to 2.92 is 3. Directivity increase over normal endfire is 2.56 dB, and
the sidelobe ratio is 9.92 dB. This distribution is suitable for long arrays; for short
arrays a computer optimisation of phase is recommended.
n W2 w 3w/2 k o d=2 7r
2-25
20
-TT/2
1-75 20
b A/=3
The Hansen-Woodyard distribution is endfire. In general, the maximum direc-
tivity does not occur there. The most general solution for uniform amplitude would
allow any element phases needed to maximise directivity. Such a solution could be
formally realised for a given number of elements and spacing, but the equations
would require a numerical solution. A slightly simpler problem was worked by Bach
and Hansen;9 they started with a uniform-amplitude array that was phased to pro-
duce a main beam at 6Q. The interelement phase factor is 6:
5 = kd sin 60 (9.115)
and the directivity (see Section 9.1.1.3) is:
Linear arrays 59
s-y N
(9.116)
1 + 7; Z (W — n) sine (wfaO cos (nb)
Calculations were made for 2-, 3-, 4- and 10-element arrays for all beam angles, and
for spacings up to X. Fig. 9.35 shows their results.
Fig. 9.3Sd is striking in that high directivity occurs along a line roughly for
37T/2 k O d=2rf
-it/2 - '
ir\l •T 3fr/2
]
v*—"^^ y3
1 '
^7
9
-Tt/2-
12
8 ^ ^ S8 10
8=-Tt I
(d)
A;d + 5 = 0, or 0O = ~TT/2, with peak directivity near endfire at X spacing. Along the
line roughly for kd = 5 + 2TT directivity is changing rapidly, perhaps due to appear-
ance of another lobe. Directivity values are shown at 0O> but in some cases a 'side-
lobe' may have higher amplitude. Thus even for uniform amplitude an array is
complex. Next variable amplitude excitation is allowed.
60 Linear arrays
Maximum directivity for fixed spacing: For an array with a given number N of
elements and a fixed spacing, it is clear that there is a maximum directivity. The
array coefficients that produce this maximum are found by the Lagrange multiplier
method (Courant31). Since the set of coefficients can be scaled by any common
factor, it is convenient to make the sum unity. Then for an array of an even number
of elements:
N/2
Y.An = 1 (9.H7)
n=l
This allows the inverse directivity of a broadside array (which will be minimised) to
be expressed as:
i N/2 N/2
~ = 2 I I AnAm[fnm+fn.m] (9.118)
tr n-im = l
where fn = sine (nkd). An odd number requires a change of limits and introduction
of the Neumann number.
Applying the variational method gives a set of equations, where the Lagrangian
multiplier is/*:
N/2
I i4B [ / „ • „ - ! + / » - m ] + P = 0, m=l,2,...,Nl2 (9.119)
The set of equations hasjV/2 + 1 unknowns andiV/2 equations; again the sum equa-
tion completes the set. The multiplier can be eliminated by substituting from the
m = 1 equation.
N/2
M = - lAn[fn+fn-i] (9.120)
n=l
This gives a set of N/2 simultaneous equations in the array coefficients. These were
solved by hand for iV = 3, 5, 7 by Pritchard,120 and can readily be solved by com-
puter for any reasonable N. Writing the set of equations:
N/2
I An = 1
n=l
N/2
YAn[fn + m-l+fn~m-fn~l-fn] = 0, m = 2, 3, . . . ,M (9.121)
n=l
The RHS vector is (1, 0 , . . . ) and the matrix of unknown coefficients is:
"J"/n-3~~/n-l ~~ fnl
03 CU 05 06
For endflre beams, in the directivity expression (eqn. 9.118)/ n is replaced byf2n
and this carries over to the simultaneous eqns. 9.119. The element coefficients will
now be complex, due to the Hansen-Woodyard endflre aperture. For progressive
phasing (phasing matches space-wave phasing), the directivity is N for spacing of
both X/4 and X/2. However, at X/4 spacing, the optimum excitation in amplitude
and phase yields a directivity of iV2.
(9.123)
where 6 is the angle of the main beam from broadside. Tai154 independently
developed this for broadside arrays. Here cos0 = 1. Since the Legendre polynomial
at argument 1 can be written as a product of factors, the result for maximum direc-
tivity is simply
I . 3. 5 . . . E N+ 1
where E = 2 x AINT -1 (9.124)
Here E is the number of odd elements or the number of even elements minus 1.
Thus the interesting result that 3-and 4-element arrays have the same limiting value,
5- and 6-element arrays have the same value, etc. After reflection this is not surpris-
ing as both 3- and 4-element arrays have two variables, both 5- and 6-elements have
three variables, etc. This maximum directivity is plotted in Fig. 9.37. The circles
show the corresponding limiting directivity for Chebyshev arrays with lOdB SLR
and the squares show for the 20 dB SLR. The Chebyshev directivity is less, as the
maximum directivity pattern does not have equal-level sidelobes, and even in the
case where there is only one sidelobe (backlobe), the Chebyshev result would be
equal to the maximum value only if the sidelobe ratio were properly chosen. See
the next Section for a discussion of Chebyshev arrays.
9 10 11 12 13
For endfire arrays, the maximum directivity in the limit of zero spacing is:
Uzkov also shows that the maximum endfire directivity at X/2 spacing is N. For
X/4 spacing, the maximum directivity isN2.
(9.126)
where xn = cos (nir/M), m = 0 is the centre element and Ml = INT(M/2.)9 Ml =
INT((M + l)/2.)
This is used in double-precision form due to the partial cancellation that occurs in
superdirective arrays.
Using the formula for array directivity and the Chebyshev coefficients, direc-
tivity for arrays of 3, 5, 7 and 9 elements have been calculated, for sidelobe ratios
of 10 and 20 dB. The superdirectivity can be seen in Fig. 9.38 and 9.39 for spacing
below 0.5X, as the ordinary directivity (using the Chebyshev coefficients that are
independent of spacing) goes smoothly to zero dB at zero spacing. The figures dis-
play these calculated directivities vs. element spacing. Thus a 3-element array
offers roughly 3 dB extra directivity for small spacings, 5 elements offers roughly
5 dB. Next the disadvantages of superdirectivity will be discussed.
01 02 03 0U 05 06 07 08 09 10
ttn
efficiency), sensitive mechanical and electrical tolerances, and high Q (narrow band-
width). First simple calculations are made using Dolph-Chebyshev arrays.
linear up to spacings of roughly 0.1X. For N = 3 the slope is 2:1, and for TV = 5 the
slope is 4:1. Thus for TV = 3, if the element spacing is halved, the tolerances must
be 4 times tighter. The percentage tolerance for the centre element, to maintain
close to (0.5 dB) the expected directivity, is approximately:
100
% tolerance ^ —==
10?
105
10*-
io 2 H
10-
1
0.01 0.02 0.04 0.1 0,2
d/A
Fig. 9.41 Q of broadside arrays
9.1.4.1 Background: Large arrays with half-wave spacing contain many ele-
ments: the associated cost, weight, etc. may be unacceptable. Also, mutual coupling
may seriously degrade scanning performance. Since half-power beamwidth is
approximately proportional to X/L, array length L cannot be reduced without
increasing beamwidth. In some cases, array size is dictated by beamwidth, but the
directivity of a filled array is not needed. Thus a thinned array can offer essentially
the same beamwidth with less directivity and fewer elements. Directivity is approxi-
mately equal to the number of elements N. Mutual coupling effects are also signifi-
cantly reduced (Agrawal and Lo2). The average sidelobe level (power) is l/N. Regu-
lar thinning produces grating lobes, but these can be suppressed by randomising*
the element spacings. There is then the question of how to select the 'random'
spacings. Most of the analytical work started with a desired pattern expressed as a
Fourier series, then devised various ways of fitting the array sum to the Fourier
series. Sandier134 expands in terms of a set of uniformly spaced arrays; Harring-
ton75 considers an array with small departures from uniform; Unz166 expands each
array term in a Bessel function series; Ishimaru83 expands in a delta-function series.
All of these approaches suffer from a common difficulty: in order to expand the
desired pattern into a Fourier or any other series, the pattern must be known in
detail. If only amplitude is known the problem is much more difficult; ideally pat-
tern amplitude and phase are given. Rarely, however, is phase specified, and for
narrow-beam, low-sidelobe arrays, the amplitude cannot even be specified. Rather
the sidelobe envelope is usually specified, or the envelope below which all sidelobes
must lie is specified. No Fourier series can be fitted to a sidelobe envelope as the
series represents the actual lobes. Thus, analytical efforts have been found to be of
little use for narrow-beam patterns. Shaped beam patterns have been synthesised by
Stutzman152 using a Legendre polynomial expansion of the array pattern, analogous
to a Gaussian quadrature.
Computer selection of element spacings has also been pursued. However, the
problem is enormously difficult as a simple calculation will show. Take an array
100 wavelengths long, and allow possible element positions every eighth wave-
length. Now take 40 elements, a 20% filled situation. There are:
/800\ IQ70
\40J "
possible patterns to be evaluated if a direct search is used. Dynamic programming
was tried as a search speedup, but since the optimum position of an element
If the array spacings are selected by a deterministic algorithm, the array can be called
aperiodic instead of random.
Linear arrays 69
depends upon the positions of elements on both sides of the element, the problem
does not meet the necessary criteria, and some empirical adjustments are needed to
get useful results. Other search speedups have proved unsatisfactory. Optimisation
algorithms using conjugate gradient methods, such as Fletcher-Powell or Fletcher-
Reeves*, will yield an optimum thinned array, say minimum sidelobe envelope for a
given array length, but are exceedingly slow. Most of the random-array computer
studies have been on small arrays (King et al.;92 Andreasen;8 Redlich;121 Stein-
berg148). Use of such arrays is limited by the type of sidelobe structure that usually
appears. High sidelobes occur at random angles, with a wide range of sidelobe
heights in between these high lobes. Thus, the computer design usually attempts to
keep all sidelobes below a fixed value. Algorithmic comparisons have been made by
Steinberg.148
a = — (9.129)
SLR
Here the sidelobe ratio SLR is in power. The result is:
p = [l-e-*]e-(2L/\)*j7rot/3exp(-a) (9.130)
Thus sidelobe level and number of elements can be directly traded. Fig. 9.42 shows
P vs. a for L/X = 10, 30, 100, 300, 1000. For example, mSLR of 20dB is achieved
with an 80% probability for L = 1000X withTV = 1028; or an average element spac-
ing of one wavelength. For large N the formula may be simplified (Lo104):
p = [ 1 - <,-<*]6L/\ (9.131)
This gives a reasonably close fit for P>0.5 and L/X> 100. Fig. 9.43 shows the
approximate relationship for P = 0.8, 0.9 and 0.99 vs. L/\. The quantity plotted is
\0\ogN —SLR and again these results are only valid for large arrays. All of the
results are for an unsymmetrical array. If the array is symmetric about its centre
some degrees of freedom are lost, and the sidelobes are higher.
09-
08-
07
0 6
!5 0 5
0-4.
03
01
U
N/SLR
In using the curves of Fig. 9.43, care must be taken that the resulting TV is not
greater than that of a filled array, 2L/\. For example, a filled array with/, = 100X
has 7V= 200 (half-wave spacing) which gives a = 9.44 for SLR = 13.26 dB. From
the curve the probability is near unity, which means that a 100X aperture can
achieve an SLR fo 13.26 dB or better. However, for L = 30X, N = 60 the value of
a for that SLR is 2.83, which from the curve gives a probability near zero. This
means that a thinned 30A aperture will probably have sidelobes worse than — 13 dB.
Linear arrays 71
Thus, for each aperture length there is a sidelobe ratio that is difficult to exceed,
with longer apertures yielding better SLR capability. For example, for 80% prob-
ability, Table 9.11 gives these values. Thus, for reasonably low sidelobes very large
arrays are required to achieve a significant thinning. But these very large arrays have
very many sidelobes; only a small fraction of these sidelobes need to be at the
envelope level to greatly reduce directivity and beam efficiency. This is expected
since directivity is roughly proportional to the number of elements. Beam
efficiency is important for radiometer and power-transfer situations. A tapered
sidelobe envelope will reduce the sidelobe power and increase beam efficiency. This
is especially important for planar arrays since there are so many more sidelobes at
large angles from broadside.
P = 0-99
dB
09
08
100 1000
L/X
(g 1321
This is only valid for W/X > \, due to approximations used in evaluating an integral.
An example will show the effect of element directivity. TakeL/X = 1000, P = 80%
and SLR — 15 dB. Then a thinned array of isotropic elements needs 325 elements.
Keeping the SLR and the number of elements fixed, consider half-wave dipole-
type elements (W/X = ^) and 2X sub-arrays. Table 9.12 gives the probabilities of
sidelobe ratio of 15 dB or greater. As expected then, directive elements reduce the
sidelobes.
9.1.4.4 Space tapering: A special kind of thinned array uses variable element
spacing to produce an equivalent amplitude taper. The goal is to: (a) produce a side-
lobe envelope that tapers down, and (b) produce lower initial sidelobes (Willey177).
As in all thinned arrays the number of array elements must be large. It is particu-
larly attractive in a distributed amplifier array where each element has its own
transmitter/receiver module. Since it is much easier to make all modules alike, space
tapering allows lower sidelobes to be realised, with all elements equally excited.
Experience has shown, however, that the sidelobes do not fall off quite as fast as
predicted by theory. Also an occasional high sidelobe is observed. It may be
necessary to make adjustments to the spacing or to the filling ratio.
Some recent calculations show the degree of thinning vs. sidelobe level. The
Taylor one-parameter distribution (see Section 9.1.1.3) is used. The number of
elements relative to the X/2 spacing filled array number No is:
Linear arrays 73
This can be integrated (using eqns. 11.4.10 and 10.2.13 of Reference 1) to give:
N= (9J34)
~^dm
Table 9.13 gives the values and also shows the relative beamwidths (taken from the
Taylor space factor). Even for SLR = 40dB, the thinning is less than half. So,
although space tapering does allow simulation of tapered amplitude distributions,
the percentage of elements saved is modest. For rectangular arrays, the values for
each co-ordinate are combined.
Now let the amplitude error be An and the phase error be 6 n ; the coefficient with
error is:
An(l + K)eiSn
The power pattern is now:
74 Linear arrays
E\B) = 1 1 ^ n ^ ( l + A n ) ( l + A
(9.136)
Under the assumption of zero means, and by using the Central Limit Theorem, that
the mean of the sum of independent variables is the sum of the means, the average
pattern with errors becomes:
^ ( 0 ) = £3(0) + £ - (9.137)
where E% is the error free pattern, Go is the directivity, and
o 2 = o2A + o28 (9.138)
This interesting result, first developed by Ruze,131'132 shows that on the average a
small constant amount is added to the sidelobes, and that this effect is less for long
arrays (high directivity). Also for low-sidelobe designs the effect is more pro-
nounced as addition of a small error field affects a low sidelobe more than it would
a higher sidelobe. The main beam peak is reduced a small amount. The probability
density function PDF of the field is a modified Rayleigh (Bennett14), because the
field at any angle is the vector sum of the no-error field and the x and y error
components.
(9,39)
In this expression An represents the peak of the unperturbed nth sidelobe, and the
product is over the desired range of sidelobes. Using a recursive algorithm developed
by McGee,110 values of probability have been calculated for arrays of 20 and 100
elements; these numbers represent the range of many actual arrays that are being
used. Figs. 9.44 and 9.45 are for uniform excitation and show probability of side-
lobes from 10° to 90° exceeding — 15 dB, and of sidelobes from 30° to 90° exceed-
ing — 20 dB. 10° is between the first and second sidelobes; the latter has a nominal
level of —17.7 dB. And 30° is between the fourth and fifth sidelobes, with the
latter nominally at —23.6 dB.
- 2 0 , dB beyond fourth sidelobe
1.0 -,
—15 dB, beyond first sidelobe
0.8-
0.6 -
0.4 -
0.2 -
As expected the larger array is less sensitive to errors. Arrays of 20 and 100 ele-
ments using a Taylor one-parameter 25 dB distribution have been similarly calcu-
lated, with the results shown in Figs. 9.46 and 9.47. Figs. 9.48 and 9.49 are for a
low-sidelobe 40 dB Taylor one-parameter space factor. Again, the cases are for side-
lobes past the first and past the fourth. Note scale changes as the design SLR
changes. In each case the dB level chosen is roughly the same number of decibels
above the nominal second sidelobe level, and similarly for the level above the fifth
sidelobe level, thus allowing the effect of initial sidelobe ratio to be observed.
This theory assumes that the perturbed sidelobe peak position is unchanged,
which is valid for small to modest errors. Thus, to calculate probability of sidelobes
exceeding a given level, only the levels of the unperturbed sidelobes need to be
known. The examples given here are not sufficiently extensive to cover all cases
that might arise. However, the Q function is easily put into subroutine form and the
calculation of probability is quite simple.
Directivity reduction is found by inserting the modified power pattern into the
directivity expression. The result, for small errors, is:
76 Linear arrays
Go
/~% (9.144)
2
l+o
Thus a one sigma phase error of 30° gives a 1.05 dB directivity loss. A one sigma
amplitude error of 1 dB yields a 1 dB directivity loss. Note that directivity loss, to
first order, is independent of array size, directivity or sidelobe level.
10-
1-4
Other errors than those in amplitude and phase can be simply related to ampli-
tude and phase if they appear singly. Combinations of various types of errors can
be analysed if the relative values of the variances of each error are known
(Elliott44).
0.30
I.O-i
0.25 0.30
- 4 8 dB
beyond fourth
l.O-i
' -42 dB
beyond first
0.8-
0.6-
/ /
0.4- /
u
0.2-
/ /
0-
0.01 0.02 0.03 0.04 0.05
O"
I.Oi
0.8-
0.6-
0.A-
0.2-
1.L
uniform
1.0 -
\30dB
0.6 -
^ \
0.4 - \25dB
\
^ \
0.2 - 20 dB
1 1 —I ! I
0- 400 600 1000
40 60 80 100 200
Linear array feeds in general are either series or shunt, which refers to the geo-
metry rather than to an equivalent circuit. Series arrays, sometimes called line
arrays, have the elements disposed along a transmission line. The line can be a
standing-wave (resonant) feed, described in Section 9.2.1, or a travelling-wave feed,
described in Section 9.2.2. Some other aspects of line feeds are discussed in Section
9.2.3, along with shunt feeds where each element is fed independently, either by a
divider network (corporate feed) or from its own receiver/transmitter module
(distributed array). Most of the discussion is centred on dipole or slot elements.
80 Linear arrays
12-1
10-
sidelobe ratio
35dB
0-8-
06-
0 4-
02-
12-,
10-
sidelobe ratio
35dB
08-
0 6-
04-
02-
Microstrip patch elements are used both for resonant and for TW arrays; see Section
9.3.4. Another type of TW array is the serrated guide with closely spaced transverse
broadwall slots (Hyneman81). This is a type of leaky-wave antenna (Tamir155).
To scan an array each element must be excited with the proper progressive
phase; a phase shifter or phaser is used. However, these devices are not covered here
(see Ince and Temme82).
Quantisation effects on pattern and gain are discussed in Section 9.2.4, where
the array is quantised into sub-arrays, or where the phasers are quantised.
Y0 = ' % Yn
o o 6
pi
Fig. 9.53 Prototype sect/on
Calling C = cos j3/, S = sin j3/, the A-B-C-D matrix of the section (Altmann5) is:
C jS\(l 0\ lC + jYnS jS
(9.145)
cj\rn 1/ \Ync+js c;
82 Linear arrays
The overall matrix is then the product of the prototype matrices;one is used for each
element in the array, starting at the feed end. A shorted quarter-wave section with
C2 = cosj3//2 and large Ys is added at the end, giving:
N jS2
- (9.146)
c D Ysc2+jS2 c2
Here C is used both as part of the A-B-C-D matrix and also as a trig term, but the
usage should be clear. Either reflection coefficient or input impedance can be calcu-
lated from A-B-C-D. The latter is:
A +B
(9.147)
3-
to
2-
and from this the VSWR is obtained. Fig. 9.54 is for an end-fed uniform array of 20
elements, showing VSWR vs. normalised frequency. Only f/fo> 1 is shown as the
curve is symmetric. The nearly linear part of this curve, with a similar curve for a
25 dB Taylor one-parameter distribution are in Fig. 9.55 plotted against NB, where
N is the number of elements and B is the total percentage bandwidth. The Taylor
array is half of a symmetric array; centre feed is assumed with each half matched.
Linear arrays 83
The curve is for N elements in the half array. From these the values of NB for
VSWR = 2 are 66 for the uniform case and 50 for the 25 dB Taylor. Tapered distri-
butions have narrower bandwidth when the strongly coupled elements are near a
feed point. This can be seen by visualising the array on a Smith chart. Starting near
the top (slightly off resonance) the first load moves the point along a constant B
line toward the centre. Next the line length gives a rotation that is nearly 360°.
Then the next load moves to a larger B, etc.
3.0
25 dB Taylor
2.5 -
2.0 -
a:
3; 1.5 -
1.0
20 60 80
Fig. 9.55 VSWR of half of resonant array; number of elements times % bandwidth
When a uniform excitation is used, the largest constants movements (for small G)
are nearly radial, which moves the admittance toward match. The last movements,
which are not radial, are small on the chart. In contrast with a tapered distribution
the small G occur in the radial movement region, while the large G values (near the
feed) move to a larger B. Thus the real part of impedance is closer to unity, but the
reactance is larger than that for uniform excitation. For all distributions there is a
negligible change in NB withiV.
Returning to Fig. 9.54, it is interesting to note that the VSWR oscillates with
frequency for wider bandwidths. A simple physical explanation of this can be
developed using simplified theory developed by Watson.173 He set up a finite
product of factors, where each factor contained a length of transmission line and a
shunt load admittance. By neglecting multiple reflections between loads, the
product could be approximated by a sum. His often quoted result is:
N
. N
T = ^r N
r- 1 1 ^N (9.148)
2+ Yne J2n(3d
I I
n=l
84 Linear arrays
At centre frequency 0d = n and the formula reduces to the simple form:
N
(9-149)
From this arises the simple rule that at centre frequency the sum of the slot sus-
ceptances should be zero and the sum of slot admittances should be unity. Fry and
GowardS9 have plotted eqn. 9.148; the VSWR rises with frequency to a normalised
frequency just above 1 + 1/2N (for uniform illumination) and for higher frequen-
cies the-VSWR oscillates about 2. From this simple theory the near linear region
gives NB = 89 for VSWR = 2 and N = 20. Note that this result is more generous
than the exact result quoted earlier. When the excitation is uniform, Yn = l/N and
the series can be summed exactly, giving:
1 KN-i
r . ^ ( 9 . 1 5 o,
Nsmpl
The sin factor is recognised as the space factor of aniV-element array, and this fac-
tor produces the oscillations in F and in VSWR as j3/ changes. The oscillation is
about F = 1/3 or VSWR = 2 since the centre of the sin factor oscillations is zero.
An even more simple derivation observes that at a frequency where the last ele-
ment is n out of phase, N$l = Nn + n/2 or/// 0 = 1 ± 1/2N, which gives a product of
NB = 100 (full bandwidth). From eqn. 9.150 this corresponds to T = 0.3475 and
VSWR =2.07.
All of the results given so far are too optimistic. When the changing element
impedance (both self and mutual) with frequency is included, the bandwidth drops
significantly, perhaps to 2/3 or less. With waveguide broad-wall shunt slot or dipole
elements, the element impedance can be included in the calculation. For waveguide
edge slots it is necessary to measure the actual bandwidth. Another practical factor
is the beam splitting that occurs at roughly twice the NB for VSWR = 2. The two
travelling waves that at centre frequency combine to give a standing wave are now
of unequal amplitude and out of phase, so that one travelling wave produces a beam
left of centre and the other produces a beam right of centre (Kummer98). However,
this effect is not noticed for bandwidths that keep VSWR < 2.
broad-wall displaced slots across the wall centre-line, while inclined broad wall or
edge slots are reversed in inclination. As a result, all elements are excited in phase
and principal radiation occurs in a broadside direction. Resonant arrays are most
often constructed of waveguide slots due to the precision with which such slots can
be milled. Due to the half-wave element spacing (in feed guide wavelengths) the
input admittance is simply the sum of the slot admittances plus the end admittan-
ces. A short is properly located at the end so that the input admittance depends
only on the slot admittances. Such a linear resonant array may be fed from one
end, but more often both ends are shorted, with a cross-guide coupler used as feed.
The flat plate array of Chapter 10 is of this type. The resonant nature of the feed
introduces bandwidth constraints.
The slot amplitude distribution can be obtained from the techniques of Section
9.1; next is the question of determining the set of slot offsets (or angles) and
lengths, assuming that slot width and waveguide dimensions have already been
selected. However, an important factor makes the simple calculation of slot para-
meters using the universal slot data of Section 9.3 incorrect. This is the ubiquitous
mutual coupling which has a significant external component, due to coupling
between slots in the region outside the waveguide, and may have a significant
internal component due to slot coupling inside the guide. The latter is implemented
via the excitation of higher modes which, though evanescent, have small decay from
a slot to its neighbours. Internal coupling is severe for edge slots and is second order
for broad wall slots. Due then to mutual coupling, which in practice couples each
slot to several slots on each side, the slot-array design problem must be solved
simultaneously instead of slot by slot.
Two relationships allow such a solution. First the sum of slot admittances,
including mutual admittances, equals the specified input admittance. For a single-
line array this is usually 1 + / 0 , but for planar arrays the real part may be different
from 1. Secondly, all mode voltages in the waveguide at each slot are equal. How
then to incorporate mutual coupling? Broad-wall slots can be replaced by Babinet
equivalent strip dipoles if the waveguide face is assumed to be a flat ground plane.
The remainder of the discussion applies to broad-wall slots. Most slots are narrow so
that the strip dipoles can be replaced by cylindrical dipoles of diameter half the
strip (and slot) width (Tai153). Then Carter's 1932 mutual impedance formulas
using zero-order current distributions on the dipoles can be used. A convenient
formulation of these for computers has been made by Hansen and Brunner.73 Use
of more accurate current distributions is not warranted as the errors introduced via
the ground-plane approximation are more significant. In designing a resonant array
of dipoles, the Carter mutual impedances, suitably weighted by the excitations, are
simply added to the dipole self impedances. Slot self admittance, however, includes
the effects of energy storage inside the guide in the vicinity of the slot, and thus the
waveguide slot is not the Babinet equivalent of the strip dipole for self-admittance
purposes. A clever solution to this dilemma was produced by Elliott and Kurtz.49
Slot self-admittance is determined by measurement or by calculation from the uni-
veral slot data. Then it is inverted to slot impedance and the mutual-impedance
86 Linear arrays
terms from the strip dipoles are added. This is again inverted to produce active slot
admittance. Turning now to the design equations for longitudinal displaced shunt
slots, slot admittance, using the modified Stevenson result of Section 9.3, and the
mutual impedance process just described, becomes:
(9.151)
Here Zn is the combined inverse of waveguide slot admittance and dipole mutual
impedances; Yn is the active admittance of the nth slot and the coefficient Cn is:
In this formulation the series does not include n = m as the waveguide slot self
admittance Y® has been written separately. Vn is the slot voltage and Znm is the
mutual admittance between the nth and rath dipoles. Thus, if the slot offsets (or
angles) and lengths are known, eqn. 9.151 with the help of eqns. 9.152 and 9.153
can be used to obtain the active slot admittances. Then the sum should be:
N y
£ -n = i + / o (9.154)
n-l ^0
This equation is complex, and hence gives two real equations. Additional equations
arise from the equality of mode voltages mentioned earlier. Mode voltage vn at the
nth slot is:
When each mode voltage is equated to the next, vt = z>2, v2 = v3,... pn_ t = vn, a set
of 2N — 2 real equations is produced. Thus, there are now 2N total equations relat-
Fig. 9.56 Resonant slot array (courtesy: Antenna Dept., Radar Systems Group, Hughes,
El Segundo)
Linear arrays 87
ing the 2N unknowns (slot offsets and lengths), so the problem is, as expected,
determinate. Unfortunately the equations are nonlinear and transcendental, so an
iterative solution must be employed. A satisfactory solution uses the Newton-
Raphson method (Stark145). Here a gradient matrix is formed, of derivatives of the
RHS of each equation, for each variable. This matrix is calculated using the vector
of variables; these are a set of 2N real simultaneous equations whose solution is the
delta vector to be added to the vector of unknowns. Through iteration a set of un-
knowns is found that satisfies the mode-voltage and input-admittance equations. A
starting vector must be provided initially. Convergence is best indicated by the
gradients becoming small. This technique has been used with success for the design
of a variety of linear and planar slot arrays. Fig. 9.56 shows a typical waveguide
resonant array.
9.2.2.1 Single-beam condition and frequency squint: The feed line may be
TEM or dispersive, and may have phase velocity above (waveguide), equal to, or
below (dielectric loaded coax) that of free space. For what range of spacings will
only one beam exist? Two examples are representative. First is a TEM line, where
the element couplings are n out of phase. Why out-of-phase coupling will be
explained later. With 6 measured from broadside and d element spacing as usual,
the interelement phase shift needed for a main beam at 60 is kd sin d0. Phase along
the feed line is kd + rr, where the IT is the out-of-phase term. Thus the phase
equation is:
kdsindo + 2nn = kd-n (9.156)
Each n > 0 represents a real beam, with the nth beam starting at d/\ = (n 4-§)/2.
Fig. 9.57 shows beam angles vs. d/X for the first three beams. No beam with com-
plete addition exists for d/X<0.25, and for d/X= 0.75 two beams exist. So the
useful range is 0.25 <d/X< 0.75. Either forward squint (d/\> 0.5) or backward
squint (d/\ < 0.5) is possible. Beam-squint change with frequency is indicated by
the slope of the n = 0 line. Spacing of 0.5X represents a special case of broadside
radiation. However, this spacing allows all element admittances to add together,
with the resulting narrow-band performance of the standing-wave array of Section
88 Linear arrays
9.2.1. Now it will be evident why TW arrays with in-phase coupling are undesirable.
The phase equation in this case is:
= kd (9.157)
This allows a forward fire (endflre) main beam for n = 0. (sin0o = 1) a* all spacings.
Although beams for n = 1,2 etc. can be positioned at a desired angle, the endflre
beam is always present.
The second example is that of a waveguide array with out-of-phase coupling.
Again, the in-phase coupling allows a beam for all spacings, at sin#o = &/k- F ° r the
out-of-phase case the phase equation is:
0d-ir (9.158)
Now both plus and minus values of n give real beams. Fig. 9.58 shows beam
position vs. d/X for the first three beams. The graph is plotted for P/k = 0.6. For
P/k = 0.5 both n = ± 1 modes start at d/X = 1. Since waveguides are usually
operated with higher ($/k, the range of operations is:
0/*>O.5 (9.159)
1+P/k X 1 + PII
For the example chosen, the n = — \ beam starts at d/X = 1.25 where the n = 1
beam starts at d/X = 0.9375. Not shown is the n — 2 beam starting at d/X = 1.563.
The Figure gives a picture of the mode appearance but is not useful for design since
it is for P/k = 0.6 only. Fig. 9.59 plots beam angle vs. spacing for various values of
P/k: P/k = 0.5(0.1)0.9. Using this graph the tradeoff of p/k and d/X to give a fixed
beam angle may be made.
Linear arrays 89
1-0-,
n=-1
08
Change of beam angle with frequency is important, as can be seen from the slope
of the curves. An explicit formula for slope is useful. Taking the derivative of
sin QQ w.r.t. frequency, and multiplying by frequency to normalise the slope, gives:
tfsinflo X2Ac , 1
(9.160)
df p/k 2d/X
This contains three interrelated variables, but fortunately it simplifies to:
d sin 60 1
— sin #o (9.161)
df
Normalised slope is plotted vs. sin#0 in Fig. 9.60. As expected, P/k nearer to unity
gives lower slopes, as do larger values of sin d0. The curves stop at the angle where a
second beam emerges; thus the practical design range is contained in these curves.
The beam angle, waveguide P/k, element spacing and normalised slope can all be
traded off for an optimum design given a particular application.
F F F F
1 2 N-1 N
t f ft
in5 l (9.165)
L =p~ (9.166)
The input power is the load power plus the sum of coupled powers, with each
multiplied by the appropriate number of segment loss factors s~l\
Gn = sr^^ s (9.169)
p «-
m-\
For negligible loss, of course, s = 1. Loss generally needs including only when it is
large (e.g. at 94 Ghz) or when the loss between elements is comparable to the small-
est conductance. Usually the Fn excitation coefficients are obtained from a distri-
bution such as Taylor one-parameter or Taylor/!, and the scale factor of theFn will
make Pin ¥= 1. Note that at each slot the fraction of power coupled out is
G/(l + G). The continuous version of eqn. 9.169 is often seen (Kummer98).
Here R(y) is the loss per unit length of guide. However, as might be expected,
sampling the continuous result to get slot coupling is less accurate than using the
sum formula (eqn. 9.169). For small arrays, or for low sidelobe arrays, it is import-
ant to use this more exact equation.
Fig. 9.62 shows the conductance values for a 29-element uniformly excited
array, for load power percentages of 5, 10 and 25%. It may be seen that the 5 and
10% curves go off scale, while the 25% curve peaks just above 0.1. It is always
necessary to keep all conductances small; 0.1 is a good maximum. Not only are
larger values often difficult to realise, but the impedance mismatch effect may no
longer be negligible. Larger arrays, of course, will have lower conductances per slot.
The curves foriV = 29 may be converted to general-purpose curves (approximately)
by multiplying values by N/2 = 14.5. Conductance values for different N are then
obtained by dividing by N/2. However, for precise results, a new calculation should
be made after the number of slots is established based on the approximate scaling
just described.
Figs. 9.63 and 9.64 give conductances for the Taylor one-parameter distribution
and for the Taylor n distribution (n = 5). For both figures N = 29 and SLR =
25 dB. Note that the heavy taper of the one-parameter distribution gives curves
with a single peak, whereas the n distribution, which has a higher pedestal (see Sec-
tion 9.1), tends to rise again near the load, especially for small load power. Since all
the curves for 5 and 10% load power have peak above 0.1, it is in practice necessary
to either increase load power, thereby reducing gain, or to increase the number of
slots. Again the curves can be scaled approximately using N/2.
Linear arrays 93
9.2.2.3 TW slot array design: Without the effects of mutual coupling the design
of a travelling-wave slotted waveguide array is simple: a table of conductances is
obtained (see previous section) and the universal slot curves of Section 9.3 are used
to find length and position of each slot to give the proper resonant conductances.
0.28
0.2-4-
0.05/ /0.1
0.20-
S 0.16-
c
o
o
8 0.12-1
P, = 0.25
0.08-
0.04-
17 21 25 29
slot number
With mutual coupling the slots interact strongly so that it is necessary to consider
the entire array at once. This has been successfully done for broad-wall shunt (dis-
placed) slots by Elliott,47 and the procedure can be applied equally well for any
array in which mutual coupling between elements can be accurately computed. A
salient exception is the inclined-edge slot waveguide array, as the replacement of
slots by Babinet equivalent dipoles for the computation of mutual impedance is not
sufficiently accurate.
The design procedure for travelling-wave arrays is similar to, but more compli-
cated than, that for resonant arrays. As is the case there, the mode voltage at each
slot is proportional to the slot voltage divided by the active impedance:
94 Linear arrays
(9.171)
Unlike the resonant array, the mode voltages are unequal, since the slots are spaced
an irrational number of guide wavelengths apart. Notation is that of Section 9.2.1,
except the mode voltages are subscripted here. In fact, the nth mode voltage is
related to the n — 1st mode voltage and the n — 1st admittance, where elements are
numbered starting at the load. This relation is:
028
P, =005
024
13 17 25 29
slot number
Yn is defined below, and d is the slot spacing along the guide. Combined with eqn.
9.171 an equation is obtained for the ratio of two successive slot voltages:
0-28-j
024-
P, =0 05
13 17 29
slot number
Starting at the slot nearest to the load where Yx = Y* + Go, eqn. 9.174 is applied
successively until the input admittance is obtained. This should be 1 + / 0 if the cor-
rect slot lengths and offsets were chosen. These two real equations, with the 2N — 2
96 Linear arrays
mode voltage equations give the IN real equations necessary to allow a definitive
problem. These also can be solved iteratively using a Newton-Raphson solution.
Fig. 9.65 shows a typical slot array.
Fig. 9.65 Travelling-wave edge slot array {courtesy: Antenna Dept, Radar Systems Group,
Hughes, El Segundo)
9.2.2.4 Fixed squint feed: As shown at the beginning of Section 9.2.2, the
beam angle changes with frequency because of the series nature of the feed. This
beam scan can be removed by use of an equal-path-length feed, as sketched in Fig.
9.66, (Cheston and Frank28). Unfortunately, much of the simplicity and low cost of
the series TW feed is lost. The series feed slots in this scheme do not radiate
directly, but each feeds a waveguide. Radiating slots or horns are arranged at the
waveguide end. The geometry is arranged so that the waveguide path length from
the input port to each radiating slot is the same; thus no change in beam position
with frequency will occur. Another advantage is that all radiating slots experience
the same amount of waveguide dispersion. The radiating waveguides should be
along the squint angle so that the aperture is broadside. This specifies the angle
between feed and radiating guides.
7 < , • \ a (9176>
A 1 + sin 6m
Linear arrays 97
radiating elements
Clearly larger s/X, and correspondingly larger m, gives a faster change of beam angle
with frequency. If the frequency scan passes through broadside there will be a sig-
nificant change in impedance due to the addition of all the element conductances.
Element conductances, or couplings, are determined just as for a TW array. If the
waveguide bends needed to make a snake are not well matched over the frequency
band, it may be necessary to calculate the input admittance using the techniques
developed for resonant arrays. The coupler reflections tend to produce a sidelobe at
the conjugate beam direction; the level of this reflection sidelobe depends upon the
coupling conductances, and is higher when the scan is closer to broadside.
In a straight TW feed losses are usually not important. But the extra length of
98 Linear arrays
the sinuous feed often introduces non-trivial loss. Maximum gain occurs when
as = 1, where s is the total snake feed length (Begovich13). The corresponding feed
loss is 4.3 dB. Maximum radar target tracking accuracy, a combination of gain and
beamwidth, occurs for as = 2, with 8.6 dB feed loss (Skolnik139).
radiating
slots
load
waveguide '
Fig. 9.67 Snake feed geometry
Dispersion in the snake feed will distort short pulses if the pulse length is less
than, or comparable to, the feed travel time. Detailed calculations have been made
by Bailin,10 and Tseng and Cheng.164
As an example, assume each serpentine loop to be five free-space wavelengths,
s = 5X0, and a waveguide wave number ratio of p/k = 0.6 at centre frequency, and a
slot spacing of half free-space wavelength. The beam angle is given by:
= 10[Vl-0.8/o//)2-0.6/o//] (9.178)
This is shown in Fig. 9.68. A 10% bandwidth covers a 117° scan.
Fig. 9.69 shows a typical folded waveguide snake feed, while Fig. 9.70 depicts a
linear feed used in a cylindrical parabolic reflector. Another way of packaging the
snake is in a helix. Here a helical waveguide with the outside wall open is machined
from a solid block; a cover is later fastened around the open top (Croney and
Foster32). Higher dispersion, and hence a larger slope of degrees per MHz, can be
obtained from a dispersive, periodically loaded waveguide, analogous to the wave-
guide delay lines used in travelling-wave tubes (Hockham and Wolfson80).
9.2.3.1 Corporate feeds: Corporate feeds are named after the structure of
organisation charts: a single port divides into two or more ports, each of which then
divides into two or more, etc., until the desired number of ports (equal to the num-
ber of array elements) is reached. Most often each port is split into two ports; for
this type of divider the number of elements must be a power of two. Bifurcated T
waveguide or coaxial-line T junctions, or hybrid junctions can be used. The former
are simpler, but hybrid junctions reduce the effects of element impedance mis-
match. Typical hybrids are the short-slot hybrid (Riblet;128 Levy102) which uses two
Linear arrays 99
90-|
0--
-90
0.95 1.0 1.05
f/fO
guides with a contiguous narrow wall, the top-wall hybrid (Hadge64), the magic tee
(Young,185), and the ring hybrid (Tyrell165). Power dividers and hybrids can also be
implemented in stripline or microstrip (Shimizu and Jones;138 Levy;102 Young186).
separation
of parabolic
cylinder common horn over slots to
direct power from feed into
parabolic reflector
-focal lines
-line source sinuous
feed array
Fig. 9.70 Frequency scanner: Linear array feed a parabolic cylinder {courtesy: Begovich)
power
dividers
elements
phasers
time delays
9,2.3.2 Line Feeds: Linear feeds are simpler than corporate feeds, and phasers
for beam scanning can be placed in either the branch lines or in the main line. Fig.
9.73 shows these configurations. In the branch-line phaser connection the advan-
tage is that each phaser handles only 1/7V of the transmitted power. However, at
any scan angle, the basic interelement phase shift <i> = kd sin6 must be multiplied
by the number of the element. Thus left to right the phasers are set to modulo 2n
values of 0, 2$, 30 etc. With digital phasers and a microprocessor control this is not
a disadvantage. Or an adder chain can be set up to add the phase increment from
each phaser to the next. For the main-line phaser connection the advantage is that
all phasers have the same setting of (f>. But the first phaser handles essentially all the
power and the losses are in series. The last elements have a reduced excitation then,
unless the couplings are adjusted for cumulative power loss.
Mutual coupling between array elements changes with scan angle, producing im-
pedance mismatches except at the design angle. These mismatches may upset the
operation of the feed network, and in some cases may introduce systematic errors
which produce a set of sidelobes of predictable location and height. For example,
take an array with phasers in the branch lines (see Fig. 9.73a) and consider the
array to be transmitting. Some energy will be reflected from the element due to the
impedance mismatch with scan; this energy goes back through the phaser. Each
elements
phasers
v
in o—
main feed line
V V V V
Similar effects can be observed with phasers in the main feed line (see Fig.
9.73Z>), where element mutual impedance changes with scan are reflected into the
main line (Spradley and Odium143). For a detailed discussion of this work, see
Kummer.98
0 10
0-010-
0001
30 60 90 120 150 180
angle from broadside,deg
A scanned array experiences similar problems even when path lengths are
adjusted so that the path length from the feed port through each element to a wave
normal are equal. The entire aperture is not excited until the pulse has reached the
104 Linear arrays
furthest element of the array. A simple but powerful analysis of antenna transients
has been given by Tseng and Cheng.164
exp[jmp+jp(2n-l-N)-jWp]dp (9.179)
Since L =NW, u = Nw, and |3 = nv0. The integral and summation can both be per-
formed in closed form yielding:
\KK|
1 N 1 \ \| \J \l
\
0
/
desired phase /
b
W
— sin 60 < \
A
Locations of the QL are closely at the grating-lobe locations:
ink
sin0 Q j = s i n 0 o - — > m=l,2,... (9.182)
W
The second QL is always smaller than the first and has a peak value of 0.2172 at
j3 = 1.7989. For small j3, the QL amplitude is approximately 0/TT. The QL amplitude
is with respect to the main beam amplitude at broadside. At a scan angle, since the
main beam is reduced with the effective aperture, the relative QL increases. Calcu-
lations have shown that tapered (Taylor) distributions give essentially the previous
result as long as N > 20.* Values of J3/TT for various QL levels are given in Table
9.14; since j5/n = (W/X) sin $0, various combinations of W/X and 60 can be used.
For example, a QL no higher than — 15dB (with respect to main beam broad-
side) can be produced by a sub-array width of 0.9X for ± 10° scan. Suppression of
the QL occurs when the sub-array widths are slightly randomised.
9.2.4.2 Digital phasers: A digital phaser allows the desired phase to be approxi-
mated within the least count of the phaser. With M bits the least count is 2n/2M.
Thus the use of digital phasers for scanning also approximates to the ideal linear
phase over an aperture by stair steps, and the error curve is a sawtooth. However,
only the discrete points corresponding to the array element positions are important.
Two cases can be distinguished. If the element spacing and scan angle are such that
the number of elements is at least twice the number of steps, each step will have
two elements or more, and the discrete case approximates to the continuous saw-
tooth. For this case a discrete QL will be produced as in the sub-array situation.
The other case is where the number of elements is less than the number of steps,
which gives the discrete phase distribution a random character. First the oversampled
The QL amplitude result obtained previously can now be used to find QL in terms
of $. Table 9.15 gives the results and it can be seen that 4 to 5 bits are needed for
low sidelobes, but 3 bits are often adequate. The QL amplitude can be significantly
reduced if a small random phase is added to each phaser. The QL is then de-
correlated and the overall sidelobe level is raised.
Case two, with random errors, can be evaluated by using an approximate vari-
ance for the phase of |$ 2 :
2
o2 = — (9.186)
3(4 M )
The RMS sidelobe level for uniform illumination referred to broadside is o2/N (see
Section 9.1.5). At a scan angle #0, the main beam energy is reduced approximately
eos2#0; so the RMS sidelobe level with respect to the main beam is increased. Fig.
9.76 shows the RMS sidelobe level referred to broadside. Use of an aperture taper
will also increase the sidelobe level. Along with the increased sidelobes due to
108 Linear arrays
(9.187)
Table 9.16 gives this loss for various numbers of phaser bits. From this Table, three
or more bits should be adequate.
phaser bits = 2
-20-
dB
-30-
-60
too
number of elements
Fig. 9.76 RMS sidelobe level due to random phaser quantisation error
Uniform array at broadside
The quantisation lobe reduces the main beam gain, and this gain can be shown
(Allen et al.4) to be approximately:
^ sine (9.188)
Values are given in Table 9.16 where it can be seen that 3 bits are adequate for
most arrays.
Linear arrays 109
9.3.1 Dipoles
Because of its simple construction from two rods or tubes with a feed in between,
or two printed-circuit strips, the thin dipole is popular when the bandwidth is
roughly 10-20% or less. A strip dipole, where strip thickness is small compared to
strip width, is equivalent to a cylindrical dipole of radius a equal to one-fourth the
strip width w (Lo103). The effects of a dielectric sheet used as a substrate will be
discussed later. A half-wave dipole has a pattern symmetric about the dipole axis,
and with Q measured from the axis it is:
sin 6
The half-power beamwidth is 78.1°, and the directivity is 1.64 = 2.15 dB. As the
dipole length shortens, the pattern approaches that of a short dipole, sin#, with
half-power beamwidth of 90° and directivity of 1.5 = 1.76 dB. For lengths longer
than half wavelength, the pattern sharpens, then breaks up. At / = 0.625 X, the main
lobe is broadside, and the two sidelobes are small. But at / = 0.75X, the sidelobes
(at 45°) are larger than the main beam. Finally the full-wave dipole with sinusoidal
current distribution has a null at broadside. Pattern bandwidth then goes from a
short dipole (limited by impedance) to roughly 0.6X length.
Impedance bandwidth is limiting for thin dipoles. It was shown by Abraham in
1898 that a vanishingly thin dipole has a sinusoidal current distribution. The Carter
zero-order impedance theory discussed in Section 9.4 is adequate for mutual-
impedance calculations, and for self impedance of very thin (//a> 1000) dipoles.
Full-wave dipoles according to this theory have zero feed current, so a higher order
theory is needed for them. However, owing to their high input impedance and multi-
lobe pattern, full-wave dipoles are rarely used. For antennas in the vicinity of half
wave in length, the King second-order theory (King94) agrees well with experiments
as do the moment-method calculations (Harrington75). Unfortunately, no satisfactory
simple function fit to the impedance vs. frequency has been possible. The designer
can start with curves such as Fig. 9.77 from King,94 but should finish with
110 Linear arrays
6000-1
a = 2 In (I/a)
vQ = 20
0 1 2 3 4 5 6 7
(b) Poh
Fig. 9.77 Second-order resistance of cylindrical antenna king-Middleton expansion
b Dipole reactance (after King)
0-98-
0-96-
0-9/.
1
0-92-
100 103
10 n
path length and dividing by the path width. For sinusoidal current the path length
is A/4 while the width is the dipole circumference or width:
RL = te = M (9.190)
Sna Aw
For example, a printed circuit dipole at 6000 Mhz made of aluminum, and with a
length/width ratio of 62.5 has a terminal resistance of 0.50O. There is then a
spacing between dipole and ground screen that maximises the gain.
Direct (metallic) coupling of a dipole to a transmission line is not necessary;
electromagnetic coupling may be used (Forbes55). Here the shorted dipole is placed
above a two-wire feed line at an angle to the line.
Dipole bandwidth increases as the dipole becomes 'fatter'; for very fat configur-
ations a bow-tie shape has demonstrated excellent bandwidth. Only numerical
methods such as the moment method have proven useful in analysing such fat di-
poles.
Figs. 9.80 give impedance design data for bow-tie monopoles vs. electrical half
length in degrees as calculated by Butler et al2S Much broader bandwidth can be
achieved by utilising a three-dimensional structure, the open sleeve dipole (King and
Wong93). The sleeve is open as it does not surround the dipole, but consists only of
two tubes on opposite sides of the dipole. Fig. 9.81 sketches a linearly polarised
open sleeve dipole. Crossed dipoles with crossed sleeves for CP can also be construc-
ted. The crossed sleeves on each side can be replaced by a metallic circular disc.
Bandwidth, for VSWR<2 can be nearly an octave. These dipoles have been
arrayed over a ground plane with excellent results.
One of the most widely used radiators for arrays is the waveguide slot, simply a nar-
row slot cut into the broad wall or edge of a rectangular waveguide. Slots can be
accurately milled, especially with numerically controlled machines, and the wave-
guide provides a linear feed that is low loss. The precise control of aperture distri-
bution afforded by a slot array has led to their replacing reflector antennas in many
missile and aircraft radar systems. A narrow slot along the centre line of the
waveguide broad wall does not radiate. To produce radiation the slot must be dis-
placed (toward the edge) or rotated (about the centreline). Similarly, a slot in the
narrow wall does not radiate if it is normal to the edge; rotating the slot couples it
to the waveguide mode. Since an edge slot must usually be 'wrapped around' to get
resonant length, displacing such a slot is impractical. Fig. 9.82 shows the three most
important types. The displaced broad wall slot is often called a shunt slot as its
equivalent circuit is a shunt admittance across the feed line. The rotated series slot,
centred on the broad wall, is a series slot, as its equivalent circuit has a series im-
pedance. And the edge slot is a shunt-type slot. The edge slot and displaced broad-
wall slot are most often used, with polarisation often the deciding factor. In an edge
slot the E polarisation is along the guide axis, while for displaced slots it is across
114 Linear arrays
the axis. Pattern behaviour of waveguide slots is close to that of slots in an infinite
ground plane, except at angles near grazing where edge effects are important. How-
ever, in a linear array of waveguide slots spaced in the rough range of 0.5 to 0.8
600- 80
500-
400-
300-
200-
100-
0
0 30 60 90 120 150 180 210 240 270 300 330 360
A,deg
300-
-300-
J
A.deg
free-space wavelength, the slot phases must usually be alternated to correct for the
waveguide phase. For example, in a resonant array with elements spaced X^/2 apart,
the elements will alternate phase unless every other element has coupling reversed.
Linear arrays 115
Thus such arrays have slots alternating on each side of the broad-wall centreline; or
edge slot angles alternating from positive to negative. Now the array pattern is com-
posed of an array factor with double spacing, and an element consisting of a pair of
slots. For longitudinal slots the lobes can be controlled through element spacing.
With edge slots, however, cross-polarised lobes are produced at certain angles. These
lobes can be reduced by replacing each slot by a closely spaced pair of slots with
smaller inclination angle.
0-U1 in.SEMIRIGID
COAXIAL CABLE
REFLECTOR
SURFACE
f COAXIAL
INPUT
Fig. 9.81 Open sleeve dipole {after King and Wong)
longitudinal
displaced
inclined
edge
- = —~^-sin2—cos2— (9.191)
where Ro is the dipole resistance and x is the slot offset. This formula resulted from
the assumption of a cosine field distribution in the slot. Das and others33 observed
that a sinusoidal distribution is more appropriate for a slot, which yields the
modified Stevenson formula:
=
v ^-sin2-tcos^/-cosM]2 (9.192)
The unmodified formula is less accurate when the slot length / is not close to A/2.
Variational formulation of the susceptance problem was made by Oliner.114'11S He
obtained closed-form simplifications which are of use, but not sufficiently accurate
for array design. With the availability of powerful computers, a more complete
evaluation of variational forms can be made. Yee184 developed such a formulation.
A different type of solution was developed by Khac,90'91 who wrote coupled
integral equations representing external and internal electric fields. The coupled
integral equations are then solved by the moment method, using pulse-expansion
functions and delta-testing functions. In both, wall thickness is included via Oliner's
method of coupling external and internal fields by a waveguide transmission line,
where the waveguide cross section is the slot and the waveguide length is the wall
thickness. Satisfactory agreement with measured data has been realised with both
approaches.
To understand the behaviour of slots, and to put slot data in a format useful for
array design, the 'universal' curves can be utilised. Longitudinal broad-wall slots are
again used as they can be more accurately calculated. Using the Yee variational
method, corrected and improved by Goodyear Aerospace Corp.,* admittance of
slots in WR-90 waveguide was calculated. First, resonant slot length vs. slot offset
and resonant conductance are calculated. Offset x is conveniently normalised to
guide width a, while resonant length lr is normalized to free-space wavelength Xo.
Fig. 9.83 shows curves fora = 0.9, wall thickness = 0.05 and slot width = 1/16, all
in inches. Frequencies of 9.375 and lOGhz are shown, with b = 0.4 and 0.2. Stand-
ard WR-90 has 0.4 height, but the half-height guide is of considerable interest for
receiving arrays. Resonant conductance is shown in Fig. 9.84 for only one case; it
matches very well the Stevenson formula given earlier.
Next the calculated admittance is normalised by resonant conductance Gr with
slot length normalised by resonant length. The result is the 'universal' curves
developed by Kaminow and Stegen,88 (see Figs. 9.85 and 9.86). The curves are uni-
versal in the following sense. To first order the variation of Y/Gr with l/lr and with
x/a is independent of 0/fc, that is of frequency. Thus measurement or calculation of
data for a given waveguide size at one frequency is adequate. For small offsets (0.05
and below) the normalised admittance variation is also independent of a/b. How-
ever, for larger offsets data should be obtained for the exact a/b. The variational
calculation of admittance is much too slow to use with iterative methods such as
those described in Section 9.2; polynomial fits are more suitable. However, the uni-
versal curves are difficult to fit with one expression. Dividing up the curve allows a
more accurate fit, but care must be taken to avoid slope discontinuities at the joins
as these affect the gradient methods used in Section 9.2. An alternative method uses
impedance rather than admittance. When slot impedance is normalised to resonant
resistance, the curves are as shown in Figs. 9.87 and 9.88. Resistance is a straight
line (to first order), with the slope independent of a/b. There is a slope change with
fi/k as shown. The reactance curve is a slightly curved line, which again is independ-
ent of fi/k9 and for small offsets is independent of a/b also. For larger offsets the
reactance curves depend on bothx/a and a/b. Thus the impedance curves are just as
'universal' as the admittance curves, and are much easier to fit with polynomials.
05161
9375 MHz, o = 0 - 9 , b = 0 2
0508-
0-500-
0492-
9375 MHz, a = Q 9 , b = 0 4
0-484-
10 GHz. a = 0 9 , b = 0-4
0476-
0 468
0 460
0 005 010 015 020 0-25
x/a
If data are not available on a particular size and wall thickness of waveguide to be
used in an array, it is necessary to either calculate or measure sufficient data to plot
universal curves. The impedance curves are preferable as they are slowly varying.
Edge slots are difficult to analyse due to their wrap-around nature. For the same
118 Linear arrays
0 1-
001
0001
005 010 015 0-20 0 25
x/a
reason the wall thickness has a significant effect on admittance. For reduced-height
guide, where the wrap-around is severe, even the pattern is difficult to calculate. At
present, no theories for edge slots have appeared, using either variational methods
or moment methods. Array design is based on measurements, and, even here, edge
slots are difficult. Because of the strong mutual coupling between edge slots (as
compared with displaced broad wall slots) incremental conductance is usually
measured.
0.6
x/a=0.05
1.0
0.8
0.2
-0.6 "
0.90 0.92 0.94 0.96 0.98 1.00 1.02 1.04 1.06 1.08 1.10
That is, a series of resonant slots is measured; then one slot is taped up and the
remaining slots are measured again. The resulting incremental conductance
(Watson173) is that of a resonant slot in the presence of mutual coupling. Trial and
error is necessary to find the resonant length for a set of slots for a given angle,
however. For inclination angles below 15°, the resonant conductance developed by
Stevenson (Watson173) varies as sin2:
4S0a/b
(9.193)
7T$/kR0(2a/X)4
This is shown in Fig. 9.89, the error in this curve for larger angles is not important,
as incremental conductance should be used. It appropriately follows a sin20
behaviour, especially for small angles, with the incremental value larger than the
120 Linear arrays
0.6 -
*^\\x/a=0.05
0.4 0.2 \ X
0.2
/A V
// \ - QA
0 -
// \ \ \x/a = 0.2
V \v
0.05 \v \v - O.I
02 / /0.05 \
-0.2 /
- 0.
-a*
• • ' ' — " ~~
- o.
-0.6
0 i i i i l 1 ( 1 1
0.90 0.92 0.94 0.96 0.98 1.00 1.02 1.04 1.06 1.08 1.10
12-
0-8-
(U-
-0-4-
-08-
-092
0-90 092 0-94 096 0-98 100 102 1-04 106 108 1 10
The wrap-around nature of edge slots can be avoided by using an H slot (Chignell
and Roberts29), where a slot normal to the guide edge is augmented by slots at each
end that are parallel to the edges. Thus the slot looks like an H. By making the out-
side arms asymmetric the coupling can be varied, and in fact circular polarisation
can be achieved (Hill79).
Other types of slots such as probe coupled, iris excited, and crossed are discussed
by Oliner and Malech.115
For a linear array the pattern is not a pencil beam, and thus the element orientation
0-1 -
001
10 15 20 30
inclination angle, deg
is important. From the grating lobe series (see Chapter 10) the active element
pattern can be written as a series containing ideal element patterns evaluated at the
grating lobes. When this is applied to a linear array of short dipoles (current line),
Linear arrays 123
171
the following ideal element patterns result (Wasylkiwsky and Kahn ):
collinear F(d,(p) = 1
(9.195)
parallel /f-plane F(0,0) = 1/(2 - cos20)
Little has been done on approximating the parallel //-plane pattern. However it is
possible to better approximate the planar array ideal cosi/2# pattern by using TE
and TM waveguide modes instead of electric and magnetic dipoles. This multi-mode
element is simply an open-ended rectangular or round waveguide arranged to sup-
port the proper mix of TE and TM modes. Such elements are able to greatly reduce
mutual-coupling impedance changes with scan.
Compensation of active impedance changes with scan angle can be done several
ways. The three most important methods are baffles and interconnections in the
feed network (see Chapter 10) and multiple mode elements. Knittel95 has surveyed
wide-angle impedance matching (WAIM) techniques. One way of looking at the
multiple mode element is that the combination of modes approximates the ideal
element pattern. Another viewpoint is that one group of modes has a mutual coupl-
ing scan behaviour opposite to that of another group of modes. The proper mix of
modes is produced by one or more dielectric sheets placed in front of the array
face and a dielectric plug in each element. The external dielectric sheet for scan
matching was developed by Magill and Wheeler107 and analysed by Wu and
Galindo182 and by Lee and Mittra.101 Use of a dielectric plug with two modes for
wide-angle matching was developed by Tang and Wong156 and Galindo and Wu.60
The mix of modes that is needed depends upon the waveguide dimensions and on
the lattice parameters. A typical mode set might be TE 10 , TE 2 o,TE 3O ,TE 11 , TE01,
TM n . With a dielectric plug in the waveguide the TE2o mode will likely be pro-
pagating; all other higher modes will be evanescent. The plug should be half wave
long (in loaded guide wavelengths) at centre frequency to properly locate the active
admittance on the Smith chart. Then the guide dimensions are adjusted to give a
high susceptance for the TE2o mode looking into the matching network. Steps in
guide height can then be used for centre-angle impedance matching; additional
dielectric plugs can also be used for matching. Fig. 9.90 sketches a typical guide
configuration, including two sheets over the array face. The design process for
WAIM elements is based on work by Diamond,37 Knittel, Hessel and Oliner.96
Additional references are Amitay and Galindo;7 Stark, Tang and Wong;144 Diamond
and Knittel;38 Wong et al;179 Hansen;70'72 and Amitay, Galindo and Wu.6
The analysis uses the unit-cell concept pioneered by Oliner116 (see Chapter 10).
Since the unit cell at scan angles other than broadside has impedance walls, the
modes are USE and LSM rather than TE and TM. It is necessary to assume values of
all physical parameters to start the process: slab dielectric constants and thick-
nesses, plug dielectric constants and lengths, waveguide heights and widths and, of
course, the element lattice dimensions. A mode set for the waveguide element is
then assembled. Choosing a reasonably good set of all these parameters requires con-
siderable experience, gained mostly by running various cases and then examining
124 Linear arrays
the results. With all parameters set, the unit-cell admittance is calculated for each
waveguide mode, assuming unit amplitude of the latter. Many modes will be
required in the unit cell to match the waveguide mode field at the aperture (the
waveguide open end), and for each unit cell mode the admittance must be calcu-
lated. Starting with the value for large distance (from the slabs), this admittance is
transformed through the dielectric slabs, and then combined with the other unit
cell modes. Numbers of modes typically are 100 to 300. Longitudinal-section
modes must be converted to TE and TM modes so that mode admittances can be
matched. Next, starting with an incident TE10 mode in the waveguide, the aperture
admittance of each waveguide mode is calculated. Although the mode admittances
must be equal across the aperture, the mode amplitudes depend upon the total
admittance. Thus, a set of simultaneous equations, one for each waveguide mode,
can be solved. Typical results are shown in Fig. 9.91 which is for 0.403X0 square
waveguides in an equilateral triangular lattice (Tsandoulas and Knittel161). Band-
width for VSWR<3 is 25%, and the scan range is roughly a quarter hemisphere.
An interesting design principle resulting from many calculations by these authors
is that polarisation coupling (for a dual polarised array) is reduced as the element
size is increased. Thus when multi-mode elements are used they should be as large
as the lattice allows. For a single polarisation, bandwidth of greater than 50% with
VSWR < 4 and scan over nearly \ hemisphere has been obtained (Tsandoulas160'162).
waveguide
steps
An iris can be used in each waveguide radiator, below the aperture, in addition
to dielectric plugs, as an aid to matching. Excellent results have been obtained by
Lee and Jones and by Lee.
Another example of the excelllent single-plane scan compensation results obtain-
able is in Fig. 9.92 (Tsandoulas162). This array uses open-ended waveguides in a rec-
tangular lattice for i?-plane scan. All dimensions have been optimised using a multi-
mode element code described earlier.
A newer technique can be used on single-mode elements. It uses a spatial filter
consisting of one or more dielectric slabs over the array face where the transmission
Linear arrays 125
10-
0.5.
\
\
N
x
(I \
-e-
c (I 1 sin9 cos<(>(H-plane)
/
/
\ /
y
0-5- / grating lobe
scan volume
10-
15
1 U-
i13
CO
1-2-
1-1-
10
10 20 30
scan angle 8 >
Fig. 9.92 E-plane scan at high and low frequencies (courtesy: Tsandoulas)
bandwidth = 12%
a/\0 = 0.670
dy/\Q = 0.500
ib/\0 =0.420
\ 0
lb/\0 =0.410
126 Linear arrays
coefficient of the filter varies with angle and with polarisation in such a manner as
to modify the active element pattern toward cos B. See Chapter 10.
Fig. 9.93 Boxed stripline array and centre strip (courtesy: Park)
Another stripline slot configuration uses boxed stripline (stripline with shorting
edges parallel to the strip conductor). Each array stick looks like waveguide (see
Fig. 9.93), except that the slots are all on the centreline. The strip conductor is
symmetrically angled below each slot, with the coupling proportional to the strip
inclination angle. Snaking the strip conductor with the slots along the centreline
eliminates unbalance and loading problems experienced in the earlier work of
Strumwasser* where the strip was straight and the slots angled. The slots here are
one wavelength apart (in the stripline medium), so no phase reversals are needed.
Mutual coupling is accommodated in the design by techniques described in Section
9.2. A 10-element array is shown in Fig. 9.93 (Park and Elliott118).
Resonant resistance values are predicted to typical accuracies of 10% and 3%, res-
pectively. Recently James86 has developed a variational calculation for the edge
admittance, and when this is incorporated into the patch modal solution the accur-
acy of the results should increase. Thus, one may conclude that microstrip antenna
design is now on a sound theoretical basis.
metal .
metaK
Because of the close spacing between patch and ground plane, bandwidth of
microstrip antennas is limited. An accurate calculation of bandwidth includes loss
terms due to radiation, copper loss, dielectric loss and a very small term represent-
ing surface waves excited along the dielectric surface. Using the modal theory, the
radiation Q can be obtained from the TMm mode (Carver26). Then the dielectric
and copper loss Q$ can be obtained from the mode fields. Finally the Qs are com-
bined in parallel to give bandwidth: A / / / = l/Q. This has been done using the
magnetic-wall assumption at the gaps, with the result shown in Fig. 9.95 (Derneryd
and Lind34). It can be seen that bandwidth is nearly linearly proportional to micro-
strip thickness in guide wavelengths; a thickness of 0.05 gives about 4% bandwidth.
For very small thicknesses, the copper losses become large, which increases the
bandwidth! For practical purposes, however, the variation is linear. There is a sort
of fundamental limitation here similar to that of Chu: a given substrate thickness
yields a fixed bandwidth, essentially dependent only on thickness.
Bandwidth can be increased by the simple expedient of extending the patch
above the substrate. It is not desirable to increase the entire substrate thickness due
Linear arrays 129
000 0 15
Fig. 9.95 Bandwidth of microstrip antenna {courtesy: Derneryd & Lind, 1979)
VSWR < 2
tan (6) = 10" 3
a = 10 7
er = 2.52
calculated
• measured
metal
dielectric
b metal
Fig. 9.96 Two-layer microstrip for increased bandwidth {courtesy: Hall et al.)
a Top view
b Side view
130 Linear arrays
to direct radiation from the feed conductors. Fig. 9.96 from Hall, Wood and
Garrett65 shows a way of coupling to the raised patches. Bandwidth can be
increased by a factor of 2 or 3, but the low-cost fabrication and smooth array sur-
face have both been lost.
Dual orthogonal polarisation can be achieved simply by feeding a patch with two
properly phased feeds. However, circular polarisation can be produced by a single
feed. Feeding a circular or square patch at the edge produces linear polarisation. At
one time it was thought that a more complex shape, such as a five-sided patch,
might be needed to radiate circular polarisation. CP can be produced by a nearly
square patch fed at the corner, where one pair of sides (magnetic walls) resonates
at a slightly higher frequency than the other pair. If the phase difference at the
centre frequency between the pairs of sides is n/2 circular polarisation results (see
Fig. 9.97). Another scheme uses a square patch fed at the centre of one edge, with
an inclined slot cut out in the centre of the patch (Kerr89). Figs. 9.98 and 9.99
show the patch and the pattern measured with a rotating linear dipole. Truncating
opposite patch corners has the same effect in producing CP.
a+A
to a set of patch conductances across the microstrip feed line. Using modal theory
each patch length (along the feeder) and width are chosen to give the required con-
ductance with the length then adjusted for match. Fig. 9.102 sketches a travelling
wave array built by Metzler; the actual array has 28 elements. The range of coupling
that can be achieved and the impedance properties of such series-fed patch arrays
need further investigation. Another travelling-wave configuration uses open-circuit
microstrip stubs as radiators, with stubs alternating left and right. The beam squint
can be removed without incurring a high resonant VSWR by using stubs in groups
of two with A/4 spacing. The 45° phase error increases sidelobes somewhat. Fig.
9.103 depicts such an array (James and Hall85). Stub width may be varied to
achieve an aperture taper.
Fig. 9.100 Strip line slot array corporate feed {courtesy: Waterman)
Fig. 9.101 26-eiement series-fed array for 1.415 GHz {courtesy: Metzler)
maximise gain, as in a rhombic. The antenna and the coupling are all produced by
successive etches on the substrate. A second configuration uses a dielectric-rod
surface-wave antenna, with the rod consisting of a tapered-height rectangular bar
made on the substrate, and excited by a vee antenna. Another uses a reflecting
corner with a wire feed several wavelengths long, with the corner made as part of
the substrate. An array configuration uses a grid of parallel conducting strips with
an array of microbolometers between the strips. An overview of this embryonic
field is given by Schwarz and Rutledge.137
fl
UU LT¥
Fig. 9.103 Travelling-wave array of microstrip stubs {courtesy of James and Hall)
1 ABRAMOWITZ, M., and STEGUN, L.: 'Handbook of mathematical functions' NBS, 1970
2 AGRAWAL, V. D., and LO, Y. T.: 'Mutual coupling in phased arrays of randomly spaced
antennas', IEEE Trans., AP-20, May 1972, pp. 288-295
3 ALEXOPOULOS N. G. et al: 'Radiation properties of microstrip dipoles', IEEE Trans.,
AP-27, Nov. 1979, pp. 853-858.
4 ALLEN J. L., et al: 'Phased array radar studies', Lincoln Lab. MIT. TR238, Aug. 1960
5 ALTMAN, J. L.: 'Microwave circuits', Van Nostrand Co., 1964
6 AMITAY, N. et al.: Theory and analysis of phased array antennas'. Wiley, 1972
7 AMITAY, N., and GALINDO, V.: 'The analysis of circular waveguide phased arrays, BSTJ,
47, 1968, pp. 1903-1931
8 ANDREASEN, M. G.: 'Linear arrays with variable interelement spacings', IEEE Trans.
AP-10, 1962, pp. 137-143
9 BACH H., and HANSEN, J. E.: 'Uniformly spaced arrays', in Collin and Zucker (eds.)
'Antenna theory Pt 1'. McGraw Hill, 1969, Chap. 5
10 BAILIN, L. L.: 'Fundamental limitations of long arrays'. Rept. TM33O Hughes Aircraft
Co., Culver City, California, 1956
11 BANDLER, J. W., and CHARALAMBOUS, C: 'Theory of generalized least Pth approxi-
mation', IEEE Trans., CT-19, 1972, pp. 287-289
12 BARBIERE, D.: 'A method for calculating the current distribution of Tschebyscheff
arrays', Proc. IRE, 40, 1952, pp. 78-82
13 BEGOVICH,N. A.: 'Frequency scanning', in Hansen, R. C. (Ed.): 'Microwave scanning
antennas: Vol. III'. Academic Press, 1966, Chap. 2
14 BENNETT, W. R.: 'Methods of solving noise problems', Proc. IRE, 44, 1956, pp. 609-638
15 BICKMORE, R. W., and SPELLMIRE, R. J.: 'A two-parameter family of line sources',
Rept. TM 595, Hughes Aircraft Co., Culver City, California, 1956
16 BLACKMAN, R. B., and TUKEY, J. W.: 'Measurement of power spectra', Dover Publ.,
1958.
17 BLOCH, A., MEDHURST, R. G., and POOL, S. D.: 'A new approach to the design of
super-directive aerial arrays,' Proc. IEE, 100, Part III, 1953, pp. 303-314
18 BLOCH, A., MEDHURST, R. G., and POOL, S. D.: 'Superdirectivity,' Proc. IRE, 48, I960,
p.1164
19 BOUWKAMP, C. J. and DE BRUIJN, N. G.: 'The problem of optimum antenna current
distribution', Philips Research Reports, 1, 1946, p. 135
20 BREITHAUPT, R. W.: 'Conductance data for offset series slots in stripline,' IEEE Trans.,
MTT-16, 1968, pp. 969-970
21 BRICOUT, P. A.: 'Pattern synthesis using weighted functions,' IRE Trans., AP-8, 1960,
pp.441-444
22 BROWN, J. L.: 'A simplified derivation of the Fourier coefficients for Chebyshev pat-
terns', Proc. IEE, 105C, 1957, pp. 167-168
23 BROWN, J. L.: 'On the determination of excitation coefficients for a Tchebycheff pat-
t e r n , ' / / ^ Trans., AP-10, 1962, pp. 215-216
24 BROWN, L. B., and SCHARP, G. A.: 'Tschebyscheff antenna distribution, beamwidth,
and gain tables'. NOLC Report 383, Feb. 1958
25 BUTLER, CM., et al.: 'Characteristics of a wire biconical antenna,' Microwave J., 22,
Sept. 1979, pp. 37-40
26 CARVER, K. R.: 'Practical analytical techniques for the microstrip antenna', Printed Cir-
cuit Antenna Technology Workshop, New Mexico State Univ., 1979
27 CHENG, D.K.: 'Optimization techniques for antenna arrays', Proc. IEEE, 59, 1971, pp.
1664-1674
28 CHESTON, T. C, and FRANK, J.: 'Array antennas', in Skolnik, M. I. (Ed.): 'Radar hand-
book'. McGraw-Hill, 1970
Linear arrays 135
29 CHIGNELL, R. J., and ROBERTS, J.: 'Compact resonant slot for waveguide arrays', Proc.
IEE, 125, 1978, pp. 1213-1216
30 CHU, L. J.: 'Physical limitations of omnidirectional antennas,'/ Appl. Phys., 19, 1948,
pp.1163-1175
31 COURANT, R.: 'Differential and integral calculus: Vol. II.' Interscience Publishers, Inc.,
New York, 1936
32 CRONEY, J., and FOSTER, D.: 'New techniques in the construction of frequency-
scanning arrays',Microwave J. 7, May 1964, pp. 72-74
33 DAS, B. N., and SANYAL, G. S.: 'Network parameters of a waveguide broad wall slot
radiator,' Proc. IEE, 117, 1970, pp. 41-44
34 DERNERYD, A. G., and LIND, A. G.: 'Cavity model of the rectangular microstrip
antenna', Proc. Printed Circuit Antenna Technology Workshop, October 1979, New
Mexico State University
35 DERNERYD, A. G., and LIND, A. G.: 'Extended analysis of rectangular microstrip reso-
nant antennas,' IEEE Trans., AP-27, 1979, pp. 846-849
36 DESCHAMPS, G., and SICHAK, W.: 'Microstrip microwave antenna'. Proc. 1953 AHerton
Antenna Symposium, University of Illinois
37 DIAMOND, B. L.: 'A generalized approach to the analysis of infinite planar array anten-
nas', Proc. IEEE, 56,1968, pp. 1837-1851
38 DIAMOND, B. L., and KNITTEL, G. H.: 'A new procedure for the design of a waveguide
element for a phased-array antenna', in Oliner and Knittel (Eds.): 'Phased array anten-
nas', Artech House, 1972, pp. 149-156
39 DOLPH, C. L.: 'A current distribution for broadside arrays which optimizes the relation-
ship between beam width and side-lobe level,' Proc. IRE, 34, 1946, pp. 335-348
40 DRANE, C.J.: 'Derivation of excitation coefficients for Chebyshev arrays', Proc. IEE,
110,1963, pp. 1755-1758
41 DRANE, C.J.: 'Dolph-Chebyshev excitation coefflceint approximation', IEEE Trans.,
AP-12, 1964, pp. 781-782
42 DRANE, C.J.: 'Useful approximations for the directivity and beamwidth of large scan-
ning Dolph-Chebyshev arrays', Proc. IEEE, 56, 1968, pp. 1779-1787
43 DUHAMEL, R. H.: 'Optimum patterns for endfire arrays', Proc. IRE, 41, 1953, pp. 652-
659
44 ELLIOTT, R. S.: 'Mechanical and electrical tolerances for two-dimensional scanning
antenna arrays', IRE Trans., AP-6, 1958, pp. 114-120
45 ELLIOTT, R. S.: 'Design of line-source antennas for sum patterns with sidelobes of indi-
vidually arbitrary heights', IEEE Trans., AP-24, 1976, pp. 76-83
46 ELLIOTT, R. S.: 'On discretizing continuous aperture distributions', IEEE Trans., AP-25,
1977,pp. 617-621
47 ELLIOTT, R. S.: 'On the design of traveling-wave-fed longitudinal shunt slot arrays',
IEEE Trans., AP-27, 1979, pp. 717-720
48 ELLIOTT, R. S., and JOHNSON, R. M.: 'Experimental results on a linear array designed
for asymmetric sidelobes', IEEE Trans., AP-26, 1978, pp. 351-352
49 ELLIOTT, R. S. and KURTZ, L. A.: 'The design of small slot arrays', IEEE Trans., AP-26,
1978,pp. 214-219
50 EVANS, J.E.: 'Synthesis of equiripple sector antenna patterns', IEEE Trans., AP-24,
1976,pp.347-353
51 EVANS, R. J. and FORTMANN, T. E.: 'Design of optimal line-source antennas', IEEE
Trans., AP-23, 1975, pp. 342-347
52 FLETCHER, R.: 'A new approach to variable metric algorithms', Comp. J., 13,1970, pp.
317-322
53 FLETCHER, R., and POWELL, M. J. D.: 'A rapidly convergent descent method for mini-
mization', Comp. J., 6, 1963, pp. 163-168
136 Linear arrays
54 FLETCHER, R., and REEVES, C. M.: 'Function minimization by conjugate gradients,'
Comp. /., 7, 1964, pp. 149-154
55 FORBES, G. R.: 'An endfire array continuously proximity-coupled to a two-wire line',
IEEE Trans., AP-8, 1960, pp. 518-519
56 FRANZ, K.: 'The gain and the (Rudenberg) 'absorption surfaces' of large directive arrays,'
Hochfrequenztech. u. Elektroakust., 54, 1939, p. 198
57 FRIEDMAN, B.: 'Principles and techniques of applied mathematics'. Wiley, 1956
58 FRIIS, H. T. and LEWIS, W. D.: 'Radar antennas', BSTJ, 26, 1947, pp. 219-317
59 FRY, D. W., and GOWARD, F.K.: 'Aerials for centimetre wavelengths'. Cambridge Uni-
versity Press, 1950
60 GALINDO, V., and WU, C. P.: 'Dielectric loaded and covered rectangular waveguide
phased arrays', BSTJ, 47, 1968, pp. 93-116
61 GANTMACHER, F. R.: 'The theory of matrices: vol. 1.: Chelsea Park, 1960
62 GILBERT, E. N., and MORGAN, S. P.: 'Optimum design of directive antenna arrays sub-
ject to random variations', BSTJ, 34,1955, pp. 637-663
63 GILL, P. E., and MURRAY, W.: 'Quasi-Newton methods for unconstrained optimization',
/. Inst Math. Appl.f 9, 1972, pp. 91-108
64 HADGE, E.: 'Compact top wall hybrid junction', IRE Trans., MTT-1, 1953, pp. 29-30
65 HALL, P. S., et ah'. 'Wide bandwidth microstrip antennas for circuit integration', Elec-
tron. Lett., 15, 1979, pp. 458-460
66 HAMMERSTAD, E. O.: 'Equations for microstrip circuit design'. Proc. European Micro-
wave Conf., 1975, pp. 268-272
67 HANSEN, R. C : 'Gain limitations of large antennas', IRE Trans., AP-8, 1970, pp. 490-
495; see also correction: AP-13, 1965, p. 997
68 HANSEN, R. C: 'Communications satellites using arrays', Proc. IRE, 49, 1961, pp. 1067-
1074
69 HANSEN, R. C : 'Microwave scanning antennas: Vol. 1.' Academic Press, 1964, Chap. 1
70 HANSEN, R. C. (Ed.): Spec. Issue on 'Electronic scanning', Proc. IEEE, 56, 1968, pp.
1761-2038
71 HANSEN, R. C : 'Formulation of echelon dipole mutual impedance for computer', IEEE
Trans., AP-20, 1972, pp. 780-781
72 HANSEN, R. C: 'Significant phased array papers'. Artech House, 1973
73 HANSEN, R. C. and BRUNNER, G.: 'Dipole mutual impedance for design of slot arrays',
Microwave J, 22, 1979, pp. 54-56
74 HANSEN, W. W., and WOODYARD, J. R.: 'A new principle in directional antenna design',
Proc. IRE, 26, 1938, pp. 333-345
75 HARRINGTON, R. F.: 'Effect of antenna size on gain, bandwidth and efficiency', /. Res.
NBS, 64D, 1960, pp. 1-12
76 HARRINGTON, R. F.: 'Sidelobe reduction by nonuniform element spacing', Trans. IRE,
AP-9, 1961, pp. 187-192
77 HARRINGTON, R. F.: 'Time-harmonic electromagnetic fields', McGraw-Hill, 1961
78 HILDEBRAND, F. B.: 'Introduction to numerical analysis', McGraw-Hill, 1956
79 HILL, D. R.: 'Circularly polarized radiation from narrow wall slots in rectangular wave-
guide', Electron. Lett., 16,1980, pp. 559-560
80 HOCKHAM, G. A., and WOLFSON, R. I.: 'Frequency scanning antenna using evanescent-
mode waveguide', Proc. IEE Conf. on Antennas, Nov. 1978, pp. 21-24
81 HYNEMAN, R. F.: 'Closely-spaced transverse slots in rectangular waveguide',//?£* Trans.,
AP-7, 1959, pp. 335-342
82 INCE, W. J., and TEMME, D. H.: 'Phasers and time delay elements', in Young, L., 'Advan-
ces in microwaves: Vol. 4'. Academic Press, 1969, pp. 1-189
83 ISHIMARU, A.: 'Theory of unequally-spaced arrays', IEEE Trans., AP-10, 1962, pp. 691-
702
Linear arrays 137
137 SCHWARZ, S. E., and RUTLEDGE, D. B.: 'Moving toward NMM wave integrated circuits',
Microwave /., 23, June 1980, pp. 47-52
138 SCHIMZU, J. K., and JONES, £. M. T.: 'Coupled-transmission line directional couplers',
IRE Trans., MTT-6, 1958, pp. 403-410
139 SKOLNOK, M. L: 'Introduction to radar systems'. McGraw-Hill, 1962
140 SLATER, J. C: 'Microwave electronics'. Van Nostrand, 1950
141 SOUTHWORTH, G. C: 'Arrays of linear elements', in Jasik, H. (Ed.): 'Antenna engineer-
ing handbook'. McGraw-Hill, 1961
142 SPELLMIRE, R. J.: 'Tables of Taylor aperture distributions', Rept. TM581, Hughes Air-
craft Co., Culver City, Cal., 1958
143 SPRADLEY, J. L., and ODLUM, W.J.: 'Systematic errors caused by the scanning of
antenna arrays: Phase shifters in the main feed line'. Rept. AFCRL-56-795, SR11/
1317, Hughes Aircraft Co., Culver City, Cal., 1956
144 STARK, L. et al.\ 'Multi mode phased array element for scan angle compensation of ele-
ment impedance', Alta Freq., XXXVIII, 1969, pp. 249-254
145 STARK, P. A.: 'Introduction to numerical methods', Macmillan PubL, 1970
146 STEGEN, R. J.: 'Excitation coefficients and beamwidths of Tschebyscheff arrays,' Proc.
IRE, 41, 1953, pp. 1671-1674
147 STEINBERG, B. D.: 'Comparison between the peak sidelobe of the random array and
algorithmically designed aperiodic arrays', IEEE Trans., A-21, 1973, pp. 366-370
148 STEINBERG, B. D.: 'Principles of aperture and array system design', Wiley, 1976
149 STEVENSON, A. F.: 'Theory of slots in rectangular waveguides',/. AppL Phys., 19,1948,
pp.24-38
150 STUTZMAN, W. L.: 'Synthesis of shaped-beam radiation patterns using the iterative
sampling method', IEEE Trans., AP-19, 1971, pp. 36-41
151 STUTZMAN, W. L.: 'Sidelobe control of antenna patterns', IEEE Trans,, AP-20,1972, pp.
102-104.
152 STUTZMAN, W. L.: 'Shaped-beam synthesis of nonuniformly spaced linear quays', IEEE
Trans., AP-20, 1972, pp. 499-501
153 TAI, C. T.: 'Characteristics of linear antenna elements' in Jasik, H. (Ed.): 'Antenna engin-
eering handbook', McGraw-Hill, 1961
154 TAI, C.T.: 'The optimum directivity of uniformly spaced broadside arrays of dipoles',
IEEE Trans., AP-12, 1964, pp. 447-454
155 TAMIR, T.: 'Leaky-wave antennas' in Collin, R. E., and Zucker, F. J. (Eds.): 'Antenna
theory: Part 2'. McGraw-Hill, 1969
156 TANG, R., and WONG, N. S.: 'Multimode phased array element for wide scan angle im-
pedance matching', Proc. IEEE, 56, 1968, pp. 1951-1959
157 TAYLOR, T. T.: 'One-parameter family of line sources producing modified su\itu/iru pat-
terns'. Rept. TM 324, Hughes Aircraft Co., Culver City, Cal., 1953
158 TAYLOR, T. T.: 'Design of line-source antennas for narrow beam width and low sidelobes',
IRE Trans., AP-3, 1955, pp. 16-28
159 TEMES, G. S., and ZAI, D. Y. F.: 'Least Pth approximation', IEEE Trans., CT-16, 1969,
pp.235-237
160 TSANDOULAS, G. N.: 'Wideband limitations of waveguide arrays', Microwave J., 15,
Sept. 1972, pp. 49-56
161 TSANDOULAS, G. N., and KNITTEL, G. H.: 'The analysis and design of dual-polarization
square-waveguide phased arrays', IEEE Trans., AP-21, 1973, pp. 796-808
162 TSANDOULAS, G. N.: 'Unidimensionally scanned phased arrays', IEEE Trans., AP-28,
1980, pp. 86-99
163 TSENG, F. I.: 'Design of array and line-source antennas for Taylor patterns with a null',
IEEE Trans., AP-27, 1979, pp. 474-479
164 TSENG, F. I., and CHENG, D. K.: 'Antenna pattern response to arbitrary time signals',
Can. J. Phys., 42, 1964, pp. 1358-1368
140 Linear arrays
165 TYRELL, W. A.: 'Hybrid circuits for microwaves', Proc. IRE, 35, 1947, pp. 1294-1306
166 UNZ, H.: 'Nonuniform arrays with spacings larger than one wavelength', IEEE Trans.,
AP-10, 1962, pp. 647-648
167 UZKOV, A. L: 'An approach to the problem of optimum directive antennae design', C.R.
Acad. Sci. URSS., 53, 1946, p. 35
168 UZSOKY, M., and SOLYMAR, L.: 'Theory of super-directive linear arays,' Acta Phys.
Hung., 6, 1956, pp. 185-205
169 VAN DER MAAS, G. J.: 'A simplified calculation for Dolph-Tchebycheff arrays', JAP,
25,1954,pp.121-124
170 VOGES, R. C, and BUTLER, J. K.: 'Phase optimization of antenna array gain with con-
strained amplitude excitation', IEEE Trans., AP-20, 1972, pp. 432-436
171 WASYLKIWSKYJ, W., and KAHN, W. J.: 'Element pattern bounds in uniform phased
arrays*, IEEE Trans., AP-25, 1977, pp. 597-604
172 WATERMAN, A.: 'Stripline antenna development', Proc. Printed Circuit Antenna Tech-
nology Workshop, October 1979, New Mexico State University
173 WATSON, W. H.: 'The physical principles of waveguide transmission and antenna systems',
Oxford University Press, 1947
174 WESTLAKE, J. R.: 'A handbook of numerical matrix inversion and solution of linear
equations'. Wiley, New York, 1968, p. 32
175 WHEELER, H. A.: 'Simple relations derived from a phased array antenna made of an
infinite current sheet', IEEE Trans., Ap-13, 1965, pp. 506-514
176 WHITTAKER, E. T., and WATSON, G. N.: 'Modern analysis'. Cambridge University Press,
1952,4th Edn.
177 WILLEY, R. E.: 'Space tapering of linear and planar arrays', IEEE Trans., AP-10, 1962,
pp. 369-376
178 WINKLER, L. P. and SCHWARTZ, M.: 'A fast numerical method for determining the
optimum SNR of an array subject to a Q factor constraint*, IEEE Trans., AP-20,
1972, pp. 503-505
179 WONG, N. S., et ai: 'Multimode phased array element for wide scan angle impedance
matching', in Oliner and Knittel (Eds.): 'Phased Array Antennas', Artech House, 1972,
pp. 178-186
180 WOODWARD, P. M.: 'A method of calculating the field over a plane aperture required to
produce a given polar diagram', Proc. IEE., 93 Pt. Ill, 1947, pp. 1554-1558
181 WOODWARD, P. M. and LAWSON, J. D.: 'The theoretical precision with which an arbit-
rary radiation pattern may be obtained from a source of finite size', Proc. IEE, 95 Pt.
Ill, 1948, pp. 363-370
182 WU, C. P. and GALINDO, V.: 'Surface-wave effects on dielectric sheathed phased arrays
of rectangular waveguides', BSTJ, 47, 1968, pp. 117-142
183 YARU, N,: 'A note on supergain antenna arrays,' Proc. IRE, 39, 1951, pp. 1081-1085
184 YEE, H. Y.: 'Impedance of a narrow longitudinal shunt slot in a slotted waveguide array',
IEEE Trans., AP-22, 1974, pp. 589-592
185 YOUNG, L. B.: in Montgomery, C. G. (Ed.): 'Technique of microwave measurements'.
MIT Radiation Lab. Series, Vol. 11, McGraw-Hill, 1947
186 YOUNG, L.: 'Parallel coupled lines and directional couplers'. Artech House, 1972
Chapter 10
Planar arrays
R. C. Hansen
Introduction
Planar arrays and linear arrays have much in common; thus many aspects of separ-
able arrays, i.e. those rectangular arrays with separable x and y excitations, have
been covered in Chapter 9. Included here are features that are specific to a planar
array such as change of beam shape with scan angles, grating and quantisation lobes,
excitations, directivity and, most important, feeding methods. Mutual coupling is
the key to understanding array behaviour as a function of element spacing, scan
angles etc., so it is the subject of a major part of the Chapter. Linear-array mutual
coupling effects are included here as they were omitted in Chapter 9.
Section 10.1 discusses array beam shape and beam width vs. scan, rectangular
and hexagonal lattices and their grating and quantisation lobes. The latter arise
from use of digital phasers. Directivity formulations are followed by discussion of
two circular (non-separable) aperture distributions that are used directly for large
arrays, and as a starting function for small arrays: the Taylor h and the Hansen one-
parameter. Feeds are the subject of Section 10.2, with fixed beam, single plane of
scan, and full-scan arrangements discussed. Finally mutual coupling, which is both a
difficulty to be reckoned with, and a means for understanding the subtleties of
arrays, occupies Section 10.3. An attempt is made to relate all recent concepts and
theories, including active impedance and active element pattern, isolated and
minimum scattering elements, the grating lobe series, active element efficiency, the
periodic cell, blind angles, matching networks in frequency, and matching networks
in spatial frequency. A unified understanding is now emerging.
Again it is a pleasure to thank many colleagues for help and counsel.
usually utilise a rectangular or triangular lattice; Fig. 10.1 shows the former. For
example, a rectangular array of waveguide slots may use half-guide wavelength
along one axis and outside-guide-width spacing along the other axis. Sometimes
alternate rows of slots are staggered to form an isosceles triangular lattice. A circular
disc array may use a rectangular lattice or a staggered triangular lattice. Sometimes
an equilateral triangular (hexagonal) lattice is used (see Fig. 10.2). Hexagonal lattices
are discussed below under grating lobes.
O C)
o o o
e e e
o o
Oy JO O 0 O O O
V
elements
10.1.2 Beamwidth
To calculate beamwidth of an array exactly, a numerical procedure must be used.
Since most tapered excitations derive array coefficients by sampling a continuous
distribution, it is appropriate to calculate beamwidth from continuous distributions.
Take a rectangular array with separable distribution along x and y, each of which is
a Taylor one-parameter (see Chapter 9). Then the space factor is:
F(u,v) = sine TI\JU2 — B2 sine n\Jv2 — B2 (10.4)
Here the Taylor w, v are used, and L and W are array length and width:
L ,
u = — (sin$ cos0— sin 6o cos0 o )
X
w
v - — (sin 6 cos 0 — sin 60 cos 0O) (10.5)
X
Only scan in the x plane (0 = 0) will be considered, as principal-plane scan is of
most interest. Taking first the beamwidth in the scan plane where v = 0:
sinh nB r-~ = L, ^ , ^^
F(u, v) = — sine iry/u2~B2y u = - ( s i n 0 - sin 0O) (10.6)
TtJj A
Since the value on axis is (sinh KB/KB)2, the beamwidth equation is:
sinh nB
(10.7)
This corresponds to a sidelobe ratio of 16.27 dB. Thus for lower SLR (higher side-
lobes) including uniform excitation, eqn. 10.7 is correct. For higher SLR the proper
equation is:
(10.9)
\phiB
These equations can be solved numerically given B, and the results are exactly those
given in Section 9.1.1 for linear arrays. Of course when u3 is determined, the actual
3 dB beamwidths are found from the scan angle 0O using:
63x = arc sin I — - sin 0O) 4-arc sin I — + sin 0O) (10.10)
As the beam moves towards endfire, the inside 3 dB point will occur at 90°; this
happens when sin0 o + u3X/L = 1. Beyond this the definition of beamwidth must
be changed, and to be consistent at endfire it should be the angle between the outer
3 dB points of the conical pattern. Thus at the transition angle, the beamwidth
abruptly doubles. For sin 0O > 1 — u3\jL:
Fig. 10.3 shows scan-plane beamwidth for square apertures of various lengths.
Cross-plane beamwidth is more difficult as both L and W affect it. Let 0O = 0
and 0 = 0O. Then:
L
u = — sin 0 o (cos 0O— 1)
A
W
v = —sin 0O sin 0 (10.12)
A
For large sidelobe ratios the half power u3,v3 are solutions of:
sinh TI\JB2 — U\ sinh iwjB1 — v\ sinh2KB
= (10.13)
" V2B2
Given L, W and B, a value of 0 3 can be found that satisfies the equation. If u3 or
v3 is greater than B, the hyperbolic form changes to trigonometric. The cross-plane
beamwidth is found from:
0 3 y = 2 arc sin (sin 0O sin 0 3 ) (10.14)
where the sin 0O factor represents a projected aperture. When the array is large, an
excellent approximation is:
0 3 y ~ 2 sin 0O sin 0 3 (10.15)
It is convenient to use the array aspect ratio W/L. For a given W/L andi?, the root
Planar arrays 145
0 3 can be determined as a function of (W/X) sin 60. Then the cross-plane beam width
can also be written in terms of v3:
W
— (10.16)
A
100 r
10
0-1
10 20 30 40 50 60 70 80 90
scan angle, deg
60 r
50
g 30
0 5
20
10
0-2 0-4 06 1 2 U 6 10
Fig. 10.4 Uniform rectangular aperture. Beamwidth per wavelength in plane normal to scan
70
60
50
"9 U0
E
20-
0-2 04 06
Fig. 10.5 25dB SLR rectangular aperture. Beamwidth per wavelength in plane normal to scan
Planar arrays 147
As the beam is scanned its shape changes as the projected aperture of the array
changes. For rectangular arrays the 3 dB beam contour is approximately elliptical.
However, for scans not in the principal planes, the combination of projected aperture
width and length and the scan angles results in an elliptical beam whose major dia-
meter is generally not oriented along the scan plane or principal planes (Elliott 25 ).
Since the area of an ellipse is proportional to the product of major and minor dia-
meters, the 3 dB beam area, called areal beamwidtb, is proportional to the product
of the major- and minor-axis beamwidths. The areal beamwidth is to first order
independent of azimuth angle 0, although the beam shape and orientation of major
axis may change with 0. Fig. 10.6 shows several beams as they change with scan.
Rectangular lattice: The array lattice controls the appearance of grating lobes.
Each point on the lattice contains an array element. A rectangular lattice with scan-
ning in either principal plane behaves exactly like a linear array (see Chapter 9), in
grating-lobe behaviour. At other scan angles the situation is not so simple, but can
be easily analysed by plotting the positions of the GL in the u, v plane* as developed
by Von Aulock.94 As shown in Fig. 10.7 the main beam and grating lobes occur at
the points of the inverse lattice; that is, the lattice spacing is \/dx and X/dy. All real
angles, sometimes called visible space, are inside or on the unit circle, and, of
course, when 6 = ir/2, u2 + v2 — 1. Angles outside the unit circle are then imaginary,
or in invisible space. When the main beam is scanned, the origin moves to the new
value of u, v, and all grating lobes move correspondingly. Thus, for scan in the
w-plane (0 = 0) the main beam point moves toward 4- 1 for 6 > 0 and all GL points
move the same amount parallel to the w-axis. When the GL just left of the unit
circle moves enough to intersect the unit circle, that GL becomes visible. Thus the
distance between the broadside GL and the unit circle must be no larger than unity
or the GL will not intersect the unit circle before the main beam stops at the right
side of the unit circle. The result is eqn. 9.10 of Chapter 9:
X dx
= 0 (10.17)
1 or
X sin
Unlike Chapter 9 these results have the inequality arranged to allow the grating
lobe to appear. Clearly scan along v(<j> = ir/2) is analogous:
0 = 7T/2 (10.18)
X "" 1 + sin B'
or cr T. S° </> a^
Of course scan to any pair of angles, or pair of u, v9 will produce one or more
GL if the array spacings are sufficiently large. The limiting cases are of more
interest, and these are of two types. First is diagonal plane scan sketched in Fig.
10.8. This is the GL produced when the diagonal point intersects the unit circle
normally. Values of u, v that allow this GL to appear are:
SQ-l SQ-l
(10.19)
SQdyl\
where SQ = \J\2/d% + X2/d^. This diagonal plane GL only occurs for SQ>2. In
Planar arrays 149
general, a grating lobe appears whenever the propagation constant is real; at the
transition from imaginary to real, it is zero.
tangent
GL
vgl = V i - (10.22)
dj\
elements are all equidistance by 2d, and the x-axis goes through a row of elements
spaced this distance apart, dx = d and dy = \[%d\2 where dx and dy are half the
distance to the next element, along the x or y axis. The hexagonal array is easily
analysed by separating it into two interlaced rectangular lattice arrays. The pattern
is then the sum of the two array patterns, and the grating-lobe locations for no
beam scan are simply:
nk mX mX
ugl = + • (10.24)
'Id'
The advantages of the hexagonal lattice will appear later. Fig. 10.9 shows the
inverse lattice in the w, v plane, where the GL points are all equidistant by \jd.
Scan in the w-plane is as before. Diagonal plane scan normal to the unit circle gives a
GL for:
\/3(\-d/X)
u8l = (10.25)
2d/X 2d/X
l-d/X
sin Uai = <t>gi = 6 0 ° . (10.26)
d/X
The minimum value of d/X for this lobe to appear is \. There are two tangent cases,
where the GL normally on the w-axis becomes tangent to the unit circle, and where
the normally diagonal GL becomes tangent. The first occurs for:
Planar arrays 151
l-d2jX2
ugl = ^ , vgl = V l ~d2i\2 (10.27)
Vi-rf2/x2A/3(i-<W)
• i
yj\-d2jtf n d
= — , <pgl = - - a r c s i n - (10.30)
uj k 3 A
In comparing the hexagonal lattice with the square lattice, both have grating-lobe
appearance when sin 6 = X/d — 1. However the element area in a square lattice is
d2, while that in the hexagonal lattice is 2d2\\Jl>. Thus the hexagonal lattice
requires only 0.866 as many elements to give the same grating-lobe-free area, or a
saving of 15% in number of elements (Sharp;86 Lo and Lee64).
10.1.5 Directivity
For simplicity a uniformly spaced array disposed on a rectangular lattice, with
broadside beam, is considered. The directivity is given by:
152 Planar arrays
D = (10.31)
F2 ineded(
Jo "Jo * t>
Now insert eqn. 10.1 with u0 = vQ = 0 into this expression. The numerator
immediately becomes:
/iV/2 Mil \2
NUM (10.32)
m m m
Z ZZZ
Z Z AnmApq cos (2n - 1) a cos (2m-1)$ cos (2p-l) a
n m p q
xcos(2<?- 1)0 (10.33)
with a = (ndj\)u and 0 = (irdy/\)v.
Through liberal and frequent use of trigonometric identities the four-cosine pro-
duct becomes a sum of eight terms where the integer factors A and B occur in all
possible combinations. The general term is:
i cos (Aa ± £0)
and the values A and B are:
m +q—
j2j cos [Vtf2 + b2 sin (0 4- arc tan a/b)] dcp = 2nJ0(\/a2 4 6 2 ) (10.34)
/JV/2M/2 \2
4 Z I Anm
D = Mil Nil MJ2 \ (10.36)
I
where
?! = sine 2n\J(n 4 p - l)2d2/X2 + (m + q-
The function sine x = sinx/x. Unfortunately this result for directivity does not
simplify appreciably either for uniform excitation or for half-wave spacing in the
lattice. For the half-wave case S2 simplifies to:
D~2DxDy
where Dx and Dy are the linear array directivities along x and along y. The factor
of 2 gives a result that compares well with exact square-array calculations
(Hansen44); the factor of 2 accounts for a pencil beam radiated on one side only.
In the previous derivation the element was assumed to be isotropic. Directivity
of an array of actual elements can be computed whenever the mutual impedance
between only two of the elements can be calculated. Following the method of
Section 9.1.1.2, the general directivity formula is:
120 I
n m
D = (10.1.39)
2- Ld Lu 2- AnmApqKnmpq
n m p q
Here the sums go over all the nm elements, and Rnmpq is the mutual resistance
between the nmth element and the pqth element. For slots and dipoles fast and
efficient algorithms exist for calculation of mutual resistance (Hansen and
Brunner47) so that calculation of directivity without approximations is facilitated.
A useful picture of the variation of planar array directivity with scan angle can
be obtained by considering a uniformly excited planar aperture. The usual spherical
coordinate system is used but the aperture is in the y-z plane. Thus the aperture
normal is d0 = n/2, <p0 = 0. The space factor is now:
where
G = — (10.41)
J_ W / J O * ' * » « « » < *
>7T/2 [IT
The 0 integral can be simplified by changing the variable to v. The necessary cos <j>
is inserted since it is essentially unity over the beam (for large apertures), with
larger error at angles where the sidelobe energy is low. Thus the <j> integral is
approximately:
cnl2 n X
E2d(j>~ (10.42)
G^4nLW^0o (1044)
X
This result, which is valid as long as the beam is narrow in both planes, states that
the effective aperture varies as sin 0Q. Near endfire the same (j> integral approxi-
mation is used, but the 6 integration must be approximated differently. There,
G = 37rx/2f f \ (10.46)
XV X
Fig. 10.12 shows how the effective aperture varies with principal plane scan, and
from this the range of usefulness of the sin 0o effective aperture may be observed.
The sin 60 effective aperture decays to the endfire value at:
sin 0O = i (10.47)
2y/2L/X
so it should be used only for scan angles well above this value.
156 Planar arrays
-2 -
—__L = 5
"- 10
\
- 6 --
\
COS8 0
-
\
-10 1 1 I 1 1
20 60 80 90
9 0 ,deg
nu n=i 1 — u 2ln%
(10.51)
Zeros of J\(jtu) are jun. Beamwidth is the ideal beamwidth (which is the same as for
a line source, see Section 9.1.1) times o. Table 10.1 gives pertinent parameters for
various SLR and h. Since the space factors are much like those for line sources they
are not graphed here. The aperture distribution becomes (Taylor90):
ft (>-4
'ft,(.-£;
= 1,F M = -
2 (jog(p)pdp\
7? = — — (10.52)
When eqn. 10.52 is inserted and the integration performed, the terms in the
numerator for m > 0 contain the factor J0(7rjLim), which is zero, leaving only the
m = 0 term. In the denominator the integration of the product of the two series
yields an integral of two Bessel functions which are orthogonal, thus reducing the
product to a single series. The final result is:
n= -J F2 (10.55)
n
I +y
!
20 0.9528 0.4465 1.2104 1.1692 1.1398 1.1186 1.1028 1.0906 1.0810 1.0732
25 1.1366 0.4890 1.1792 1.1525 1.1296 1.1118 1.0979 1.0870 1.0782 1.0708
30 1.3200 0.5284 1.1455 1.1338 1.1180 1.1039 1.0923 1.0827 1.0749 1.0683
35 1.5032 0.5653 1.1134 1.1050 1.0951 1.0859 1.0779 1.0711 1.0653
40 1.6865 0.6000 1.0916 1.0910 1.0854 1.0789 1.0726 1.0670 1.0620
Planar arrays 159
Rudduck et al.m have computed a limited range of values, shown in Table 10.2. As
is the case for the linear Taylor h distribution, there is a value of n that gives maxi-
mum directivity. From the Table it is seen to increase with sidelobe ratio.
F(u) = (10.56)
Similar to the Taylor one-parameter line source, this definition of the pattern is
valid for u>H. For u < H, the form is
F(u) = (10.57)
Jt and 11 are the usual Bessel and modified Bessel functions of the first kind and
order one. Analogously to the (sin TTU)ITTU case, the pattern form for u > H provides
a sidelobe structure much like that of 2Ji(nu)/TTUi and part of the main beam, while
the pattern form for u < / / provides a higher main beam peak than unity, thereby
raising the sidelobe ratio. The transition from Ix to Jx occurs at w = //, and from
this point to the first sidelobe is 17.57dB. From this transition point to the beam
peak is 2II(7TH)/ITH, SO the sidelobe ratio (see Chapter 9 for definition) is:
2/ X (7T//)
SLR = 17.57 dB 4- 20 log ; (10.58)
nH
From this the sidelobe ratio vs. parameter// can be obtained (see Table 10.3). In
this Table, values are given up to 50dB SLR, as the current antenna art allows
50 dB to be realised. Since both Jx and/ x are represented by absolutely convergent
series and are quickly generated by common computer subroutines, the one-
parameter pattern is easy to use.
The aperture distribution is:
g(p) = (10.59)
160 Planar arrays
Table 10.3 Characteristics of Hansen one-parameter distribution
SLR Edge taper
dB H rad V dB
17.57 0 0.5145 1 0.5145 0
20 0.4872 0.5393 0.9786 0.5278 4.5
25 0.8899 0.5869 0.8711 0.5113 12.4
30 1.1977 0.6304 0.7595 0.4788 19.3
35 1.4708 0.6701 0.6683 0.4478 25.8
40 1.7254 0.7070 0.5964 0.4216 32.0
45 1.9681 0.7413 0.5390 0.3996 38.0
50 2.2026 0.7737 0.4923 0.3809 43.9
Fig. 10.13 shows these radial distributions for sidelobe ratios of 20(5) 40 dB. The
3 dB value is a solution of the following equation and is also given in the Table:
(10.60)
y/2H H2
0 .1 .2 .5 .6 .9 1.0
Aperture Radius
Beamwidth is twice the u3 value. Excitation efficiency results from integration over
the aperture:
2(Jo g(p)pdp
V = (10.61)
Planar arrays 161
The Table also gives the product of beamwidth u3 and efficiency. In comparison
with the Taylor h circular distribution, the Hansen distribution is slightly less
efficient, but the required distribution is smoother and more robust.
Element pattern: When dipole or slots are used in planar arrays the spherical
coordinates as usually used do not allow the isolated element pattern to be written
simply, as in Chapter 9. In spherical coordinates, where the dipole is along the
jc-axis, the dipole pattern is:
IT
cos I — sin 9 cos 0
(10.62)
cos2
When 0 = 0 or 0 = TT/2, the result is unity. In the x-y plane (0 = TT/2) the pattern is:
h
cos I — cos <
F = ^ (10.63)
sin 0
while in thex-z plane (0 = 0) it becomes:
fa \
cos I — sin 8\
l ;
F = (10.64)
cos 6
10.2 Array feeds
Fig. 10.14 Quadrant of flat plane array {courtesy: Antenna Dept, Radar Systems Group,
Hughes, El Segundo)
Fig. 10.15 FPA corporate feed (courtesy: Antenna Dept, Radar Systems Group, Hughes, El
Segundo)
Fig. 10.16 4W/4CS /OM/ sidelobe array {courtesy: Westinghouse Defense & Electronic Systems
Center)
Fig. 10.17 Waveguide elements and slab line feed {courtesy: Tsandoufas)
Side view and front view of Z.-band radiating module designed for low-profile
high-precision airborne application. Integrated construction consists of ground-
plane flange, air slab-line power divider feeding six air-filled waveguide radiators
via a securely captivated probe exciter. Thin (0.020 in) dielectric membrane pro-
vides airtight environmental protection. Module measures 2.5 ft X 4.3 in X 3.4 in
and weighs 4.5 Ib. Loss is less than 0.35 dB from input connector to array face
164 Planar arrays
A power divider for the rows incorporating ring hybrids for sum and difference
patterns is shown in Fig. 10.18 (Tsandoulas92). Air slab line reduces loss compared
with conventional strip line. Polarisation diversity may be obtained by using non-
resonant slots cut into the outer conductor of large square coaxial line at 4- 45°, with
every other stick using — 45°. Coupling is controlled by slot length; feeding
alternate sticks properly allows the radiated polarizations to be varied (Wolfson and
Cho107). Perhaps the simplest type of flat plane array is the resonant grid. This
scheme is a modification of the Kraus61 end-fire array, and consists of connected
wire rectangles above a ground plane (Tiuri et aL91). Fig. 10.19 shows a wire-grid
F ig. 10.18 Azimu th power divider using ring hybrids {courtesy: Tsandoulas)
Fig. 10.20 Dual stripline slot arrays (courtesy: Physical Sciences Lab, New Mexico State
Univ.)
An unusual type of planar array is the image element array, originated by von
Trentini.95 In this scheme an element such as a slot in a ground plane is operated
with a thin high-dielectric-constant sheet parallel to and spaced about half wave (in
air) from the ground plane. The high e sheet is highly reflective, especially if it is
electrically quarter-wave thick, so that a set of images of the slot are set up. The
image element then acts as an end-fire array, with an effective area or gain that can
be as high as 15 dB above the slot alone. Thus these image elements can be used in a
widely spaced array with the narrowed element pattern suppressing grating lobes.
The ground plane and dielectric sheet act as a spatial filter, which is angle sensitive,
thereby producing low sidelobes from a sparse array. However at the Brewster angle
166 Planar arrays
(in the E-plane) the sheet is nearly transparent. The design must ensure that the
array has no grating or quantisation lobes near this angle, as they will not be sup-
pressed by the image element. When used in an array divided into quadrants for
monopulse tracking, the image element can provide improved difference pattern
sidelobes. Immell and Sasser49 found that two elements per quadrant gave
performance superior to a filled array of 19 slots per quadrant. In practice, the
elements can be waveguide or stripline slots or dipoles. The dielectric sheet may
have a relative permittivity of 30 or more; Ni-Al titanate is more machinable than
loaded resins. A honeycomb layer is used to support the dielectric sheet from the
ground plane.
Another unusual planar array is a flat plane array using a radial line feed.
Resonant slots are cut into the top plate of a radial transmission line; both radial
and circumferential slots (Goebels and Kelly32) and inclined slots (Kelly and
Goebels55), located in concentric rings, are used. However, the rectangular wave-
guide flat plane array allows more precise control of array distribution and
impedance.
end of the 'fingers', can be used to feed the branch arrays. However, such feeds are
not compact.
. * • » „ - ; . . . , : ; : - : ; !
Fig. 10.21 Ferrite scanned edge slot array {courtesy: Antenna Dept, Radar Systems Group,
Hughes, El Segundo)
Fig. 10.22 Frequency scanned edge slot array {courtesy: ITT Gilfillan)
allow this. There have been integrated phaser-elements developed where ferrite
posts or diodes are used to change the phase of the waveguide mode. Those schemes
do not in general allow adequate control of mode amplitude and input impedance.
The parallel-parallel feed is corporate in that each element is connected to a
phaser, and each phaser is connected to the output port by an equal length path
Planar arrays 169
(see Fig. 10.26). A printed-circuit spiral array is depicted in Fig. 10.27, where
successive layers contain the baluns and power dividers, PIN diode phasers, and
corporate feed. The parallel-parallel configuration is often implemented as a distri-
buted array where each element is part of a receive-transmit module. One of the
early larger distributed arrays was the FPS-85;(see Reed80). Fig. 10.28 shows an
X-band array of open waveguides fed by a post located on the end of the module.
radiator
phaser
ne
radiator phaser
s—v
29 ^
The RASSR array and a module (Collins and Hayes;16 Opp et al.ls) are shown in
Fig. 10.29, This alumina microstrip module contains TR, receiver preamplifier,
transmitter power amplifier and diode phasers. Fig. 10.30 shows the EAR array and
a module; the array contains 1818 modules, each of \ in diameter. Wrapped around
the circular waveguide element and ferrite phaser is the phaser driver circuit.
170 Planar arrays
In this Section mutual coupling for both linear and planar arrays is discussed, as
there are many similarities between the two. After some basic material is covered,
element-by-element approaches are taken up. Following this, the periodic structure
approach is explained. Finally compensation for mutual coupling effects is discussed.
radiator phaser
ront
Fig. 10.28 Distributed waveguide array {courtesy: Antenna Dept, Radar Systems Group,
Hughes, El Segundo)
172 Planar arrays
10.3.1 Fundamentals
Fig. 10.29A RASSR array {courtesy: Texas Instruments). B Element module {courtesy:
Texas Instruments)
axes, and is due to Wheeler.101'103 This current is phased to radiate at some desired
angle away from broadside. Such a current sheet can be used as a gedanken to
derive key scanning properties of phased arrays. The sheet may be either receiving
Planar arrays 173
or transmitting with the two situations giving related behaviour. An incident plane
wave (receiving) case will be considered first. Fig. 10.31 shows the current sheet
in side view, for both a wave incident in the 2^-plane, and in the //-plane. If the
Fig. 10.30A EAR array {courtesy: Westinghouse Defense & Electronic Systems Center)
B Element module {courtesy: Westinghouse Defense & Electronic Systems Centre)
current sheet is matched for normal incidence (6 — 0) to 77 ohms per square, then at
other scan angles there is a mismatch, as follows. For //-plane incidence (scan) the
incoming wave 'sees' a section of current sheet which is wider than the section of
174 Planar arrays
wavefront; thus the apparent resistance is lower by cos 6. The reflection coefficient
is now:
cos 0 - 1 - 0
V = —— = - t a n 2 - - (10.65)
cos 9 + 1 2
^-plane incidence gives the opposite result; the incoming wave sees a section of
current sheet which is longer than the section of wavefront, and thus the apparent
resistance is higher by 1/cos d. And the reflection coefficient is:
sec B - 1
r
Q
= taniz — (10.66)
sec 6 + 1
1/ H
\T /
(
/
—•H /
/ OH /
^E
H-plane
R
Ro
E-plane
90
When the current sheet transmits instead of receives, the reflection coefficient signs
will be reversed. Fig. 10.32 shows the variation of current-sheet resistance forE- and
//-plane scans, for transmitting, with Ro = 17. Of course, this simple current sheet
model gives no information about reactance. The electric current sheet requires an
open-circuit reflector behind the sheet, where open circuit means c = 0, 77 = °°.
Magnetic current sheets may also be considered, with E and H interchanged. Now
the reflector behind the sheet is a short circuit a = °°. Thus the magnetic current
sheet, unlike the electric current sheet, may be approximated physically. An array
of short slots in a metal sheet, for example, provides such an approximation. From
Planar arrays 175
the foregoing discussion and Fig. 10.32 is indicated the basic behaviour of phased
arrays, where active resistance increases in one scan plane, but decreases in the
other. This trend will be observed in later Sections where precise active impedances
or admittances are calculated.
Further insight can be obtained from another concept due to Wheeler;103
namely, that of an ideal element pattern. Consider again a transmitting current
sheet, but let the current sheet be composed of short dipoles with patterns:
i/-plane F(d) = 1
£-plane F(0) = cos B
The patterns are affected by the effective (projected) aperture which varies as cos 0,
where 6 is measured from the normal to the array. Since the effective aperture
broadens the beam, the radiation resistance is proportional to sec #, so that the net
effect is:
R F2
(ia67)
F, = ^
For the two planes this gives:
JD
//-plane — = sec 6
Ro
R
£-plane — = cos B (10.68)
Ro
in agreement with the current-sheet values. If the element power pattern were
cos B, the resistive part of the array impedance would be matched. Thus the 'ideal'
element pattern is F(0) — cos1/2B. This is a conical pattern, symmetric about the
axis. The ideal element pattern has been approximated by a Huygen's source
which is a crossed electric dipole and magnetic dipole, sometimes realised by a
dipole and loop. However, the pattern of the Huygen's source is:
and this is only a fair approximation to cos 1/2 #. In the Section on scan compen-
sation it will appear that both electric and magnetic modes are needed.* The
Huygen's source is then a crude approximation, giving a single electric and a single
magnetic mode.
10.3.1.2 Free and forced excitation: Arrays may be analysed from two concep-
tual element-by-element viewpoints. In the forced excitation model a driving voltage
* An alternative technique uses a medium over the array face with appropriately different
parallel and perpendicular transmission coefficients.
7 76 Planar arrays
(current) is applied to each element with the excitations phased from element-to-
element to produce the desired beam-scan angle. Each element has an active reflec-
tion coefficient which varies with scan angles, and which affects the actual gain
realised. The active reflection coefficient, of course, is associated with an active
element impedance (admittance). In the forced excitation model the active element
impedance (admittance) is calculated, from which the currents (voltages) at each
element may be obtained. For a dipole array the element currents are a solution of
the following equation, where the impedance matrix contains all of the inter-
element mutual impedancesZ^. The matrix is symmetric Z^ = Z^:
[V] = [Z][I] (10.70)
The overall array pattern is given by the sum over the array elements with the
currents as coefficients, all multiplied by the isolated element pattern. From this
the active element pattern can be obtained by factoring out the array factor. It
appears that this type of analysis is relatively easy to carry out, and this is true.
However, implementation of such an array is difficult as each element must be
fed by a constant-voltage (current) source. Simple feed networks in contrast are of
the constant-available-power type where an element impedance mismatch reduces
the applied voltage.
Free excitation assumes that each area of the feed network is equivalent to a
voltage (current) source in series with Ro. These are thus constant available power
sources. Such sources, with constant incident power, are suited to a scattering
analysis:
IK] = [S][Vt] (10.71)
Here Vt and Vr are the incident and reflected voltage (current) vectors and S is the
scattering matrix of coupling coefficients. The active reflection coefficient is given
by:
Ta = I I O ^ T ^ (10.72)
P Q
where C is the coupling coefficient between the 'oo' element and the 'pq* element.
The excitation coefficients differ by the progressive scan phase between the two
elements, and may also differ in amplitude if a finite array with a tapered excit-
ation is used. Although the scattering approach accurately represents most arrays,
and is conceptually simple, there is no direct way of calculating the coupling
coefficients. For isolated elements in a ground plane, the coupling in the //-plane
decreases as 1/r2 while that in theiS'-plane decreases only as 1/r (Wheeler102). In an
array, however, the coupling decay becomes asymptotic to 1/r2 as shown by
Hannan,39 Galindo and Wu29 and Steyskal.89 This is borne out by measurements on
large arrays (Amitay et a/.;3 Debski and Hannan18). Phase measurements show that
the coupled energy has the phase velocity of free space provided there is no external
loading.
Planar arrays 177
In some arrays there is a beam position (other than 6 — 90°) where the active
element pattern has a zero, or in different terms where the reflection coefficient
has magnitude of unity. Such an angle is called a 'blind spot'. It will be shown
later that the appearance of these can be precluded by proper choice of array
parameters.
V V V V V
V V V V V
Kj \QA Kj m K)
isolated element gain pattern, and a factor displaying the impedance effects
explicitly. The active element pattern is measured with that element excited and all
other array elements present and terminated in Zg. In contrast, the isolated element
pattern is measured with all other elements open-circuited. This is not quite the
same as with all other elements absent, except for canonical minimum scattering
antennas (see next Section), see Fig. 10.33.
178 Planar arrays
For the active element pattern approach, call the array factorNh(6), where th<
peak value of h is unity. Only one space variable is shown for simplicity and N ii
the number of elements. By definition the field of the array is produced by the
active element times the array factor. Similarly the array power is N times the
active element power. Using the definition of gain, the array gain is:
= Nga{0)h\6) (10.73)
MR
where ga is the active element pattern. The formula and concept are simple, but the
measurement, with many terminated elements, is difficult.
Exactly equivalent is the open-circuited element pattern approach. The array
field is written as:
E(6) = NIaEoc(d)h(d) (10.74)
where Ia and IQC are the active and open-circuited driving currents. The open-
circuited element gain is immediately:
Unit voltage has been used for convenience. The open-circuited element resistance
is Roc. Now the array gain, with a matched load at angle 0, is:
ocv K J
G(9) = ' ° " (10.76)
Ia is again the active current and Rg is the generator resistance, which allows the
array gain to be written in terms of the open-circuited element gain:
4NRiRggoc(d)h\6) „„„
_j_ 7 |2 ^ '
(10.78)
Note that this is the normal reflection coefficient with the Za in the numerator
conjugated. The active array gain then becomes (Allen2):
MR-
G(6) = —±gi(d)h2(d)\i-r\2 (10.79)
Now the two expressions for array gain can be equated, resulting in a relationship
between active and open-circuited element gain:
Note that Ra and T are calculated at the angle 0; the equation relates active element
pattern to element active impedance (Oliner and Malech76).
It is interesting to examine the formula in detail at broadside with a match there.
Then the product of element directivity and input resistance for a half-wave dipole
above a ground plane is (Kraus60):
gtRi = 480sin2£/z (10.81)
76
and the active resistance at broadside is (Oliner and Malech ):
Then the active element gain is ga = 4itDxDy. For half-wave spacing the array gain
is:
where A is the array area. Since this 'array' is a part of an infinite array (uniform
excitation) the result is as expected.
00 84)
-
This Af-port antenna possesses N orthogonal radiation patterns, which are 5,-y. In
the absence of non-reciprocal components, e.g. ferrites, the antenna is reciprocal,
which makes all patterns real, and symmetric about the origin; i.e. reversal of a
!
Note that + denotes the conjugate transpose
180 Planar arrays
line to any point on the three-dimensional pattern gives an equal point on the
pattern. Scattered and radiated patterns are equal. No impedance or admittance
matrix exists, and the N eigenvalues of the scattering matrix are all — 1. Unlike
most scatterers (or antennas) the scattered field pattern of a CMS antenna is
independent of the incident-field type and direction, although the amplitude of
the pattern will depend upon the incident wave. If, then, the pattern of an element
is taken in the array environment with all other elements open-circuited, this
pattern is the same as the isolated array pattern only if the element is a CMS
antenna. For example, when dipoles in an array are open-circuited, there is a
small effect due to the half-length conducting rods that remain when each
dipole is open-circuited. As expected, half-wave dipoles are not CMS antennas.
Similarly when slots in a ground plane are shorted across the feed terminals
the remaining half-length slots affect the current distributions on the ground
plane.
Wasylkiwskyj and Kahn96 showed that the mutual impedance (admittance)
between two identical MSA can be written as an integral of the power pattern
of an isolated element over certain real and complex angles:
where w = u +jv, and the power pattern is defined as P(w) =f(w)f(w + n) and
the integration path in the complex w plane is from (— 7r/2, — °°) to (— 7r/2, 0) to
(3/2, 0) to (3/2, + °°). The power pattern is normalised so that its integration over
all real angles is unity. This integral can be expressed in wavenumber coordinates.
For example, with two antennas in thexy plane, the mutual impedance is given by:
Mutual coupling, then, is specified by the element pattern and the lattice, and is
completely independent of the means utilised to produce that pattern. This
development for MSA is a generalisation of that of Borgiotti which is discussed in
Section 10.3.2.
Most multi-mode elements do not have orthogonal patterns, and hence are not
MSAs. However, many single-mode antennas approximate to an MSA. Short dipoles,
where the current distribution is essentially linear, are closely MSA. A resonant
(near half-wave) dipole is approximately MSA if the radius is very thin; this thinness
forces the current to be nearly sinusoidal. Mutual impedance between thin collinear
half-wave dipoles is shown to agree with the Carter results (Wasylkiwskyj and
Kahn96). Andersen, Lessow and Schjaer-Jacobsen5 show that crossed dipoles are
approximately MSA in one pattern plane, and that the mutual impedance calcu-
lated from MSA theory is good if the centre of one cross lies on a line bisecting the
arms of the other cross. Small helices are also approximately MSA.
Planar arrays 181
10.3.2.1 Mutual coupling between two elements: There are several intuitive
rules that concern mutual coupling between two antenna elements. First, the mag-
nitude of mutual impedance decreases with distance. Secondly, if one antenna is in
the pattern maximum of the other, coupling will be strong compared to that which
the pattern null produces. If the antenna is in the radiating near field of the other,
then the approximate near-field pattern is used. Thirdly, if the field polarisations
are parallel, coupling will be stronger than if they are collinear. For wire antennas
this can be restated in terms of shadowing: large shadowing will correlate with large
coupling. Fourthly, larger antennas have smaller coupling. For example, large horns
have lower mutual impedance than small horns. These rules, although useful, are no
substitute for actual values of mutual impedance or admittance. Since most arrays
use dipoles, slots, patches or horns, the discussion will be limited to them.
In element-by-element approaches it is necessary to start with mutual impedance
(or admittance) between elements. Since these are often slots or dipoles, and since
slots can be replaced by Babinet flat-strip dipoles, which if narrow in turn can be
replaced by cylindrical dipoles, the mutual impedance between dipoles is a basic
building block. For mutual impedance the zero-order (sinusoidal current distri-
bution) theory is usually adequate. For this case, the formulas of Howard King56
based on the original work of Carter14 can be used. They give mutual impedance
between parallel coplanar dipoles of unequal length in echelon. However, these
formulas contain 24 sine and cosine integral terms, of various complicated and
diverse arguments. Although the results of King could be rearranged and grouped
into a computer algorithm, it is simpler to directly "derive the formula, using the
exact near electric-field formulations of Schelkunoff and Friis.85 They show that
the exact field from a sinusoidal current is written as three spherical waves, although
the field components are in cylindrical coordinates. Mutual-impedance is then the
integral of this electric field (from each half dipole) times the current distribution
at the other dipole. With the geometry of Fig. 10.34 the mutual-impedance
expression is:
— ;30
Z = I [to - 2^ 2 cos kd2 + to + to - 2to cos kd2 + i//6]
x sin A : ^ — x)dx
where
ex
k = 2TT/X, Sx = sin kdu S2 = sin kd2, to = P( # * * ) (10.87)
182 Planar arrays
(10.88)
The integral could be evaluated numerically, e.g. by Romberg integration, and for
large slot separations this method is preferable. However, the adjacent and nearby
slots are most important, and for these an exact evaluation is necessary.
" 2 d !T
The exact solution can be written as two sums (Hansen and Brunner47), where
n steps by 2:
1 5
" ~ I I Cm[ejku[E(kA-kU)-E(kB
m= -l n = -
d = Cx = 1, Co = -2coskd2
When the two dipoles are the same length the formula simplifies (Hansen45) to:
Z = M mt= - 2 n I= - l Am
«J
where
j3 = \/(x0 + mdf +yl-n(xo + md) (10.91)
and the coefficients are:
A_2 = A2 = 1, v4.j = Ax = — 4cosfoi
2
,4 0 = 2(1 4- 2 cos kd) (10.92)
When j o = 0, i.e. the dipoles are collinear, y0 is replaced by the dipole radius a.
When the dipoles are half wave, the m = 2 and 4 terms disappear and the exponen-
tial simplifies. This formulation is similar to that of Richmond83 for two equal-length
thin dipoles with axes at an angle. Computer subroutines for Ci, Si are readily
available in most computer libraries, with computation time comparable to that of
trig functions. These subroutines use the economised series developed by Wimp and
Luke.106
Fig. 10.35 shows Zy between two parallel half-wave dipoles o n a i ? +jX plot.
The curve, with spacing/wavelength as a parameter, is similar to a Cornu spiral. It
can be seen that the magnitude of impedance decreases as spacing increases. Data
are normalised by the self resistance, which, for a zero-order (zero-thickness)
dipole, is 73.13 O. Fig. 10.36 is a similar plot for collinear dipoles. The spacing here
is between dipole tips; for centre-to-centre spacing add 0.5. As expected, the
collinear coupling is less owing both to lower shadowing and a pattern null. Recall,
however, that the dipole near field has both axial and radial electric fields and that
the latter does not have a pattern null along the axis.
When slots are located in a stripline surface, or in contiguous waveguides, the
mutual admittance between two slots behaves as though the slots are located in an
infinite ground plane, to a very good approximation. The slots can then be replaced
by strip dipoles by Babinef s principle developed by Booker10 for self impedance,
and also for mutual impedance (Begovich9). The slot mutual impedance Z 12 and
strip dipole mutual admittance. Yn are related:
2Z 12 = r?2r12 (10.93)
where 17 = 12O7T. The factor of 2 has been reduced from 4 as the slots radiate one
side only.
When the slots are narrow, which is usually the case, the equivalent strip dipoles
are also narrow; they can be accurately replaced by cylindrical dipoles of radius
equal to one-fourth the slot or strip dipole width (Lo 65 ). Thus the extensive
development of dipole mutual impedance can be used. As discussed in Section 9.2,
the Carter dipole mutual impedance is satisfactory for the design of linear slot
arrays.
184 Planar arrays
Mutual impedance between microstrip patch elements has not received much
attention, probably because arrays of these most often employ a fixed beam. In
principle, an approximation to mutual impedance could be obtained for rectangular
0.05
-0.4 -0.2
R/R
01r
005
R/Rn
patch elements by assuming that two opposite gaps act as virtual radiators for linear
polarisation (see Chapter 9). Then a Carter type of zero-order theory could be used
Planar arrays 185
to find the mutual impedance to the TEM mode inside the patch, and this
impedance could then be transformed to the feed terminal. More sophisticated
techniques such as moment method are probably not necessary as the field distribu-
tion across the patch gaps is approximately the canonical value. A recent approach
by Krowne and Sindoris62 calculates //-plane coupling between rectangular patch
elements by replacing the two radiating gaps by two coupled microstrip lines, where
the first is excited and the third acts as receiver. Some measured data are given by
Jedlicka and Carver.51
-AOr
-50 -
-60 -
-70 -
- 80
-180 -90
deg
Coupling between horns has been investigated by Lyon et al.66, who find that
the coupling is sensitive to orientation, as the polarisation rule suggests. Fig. 10.37
shows coupling between identical £-plane sectoral horns with 8 dB gain, at a separ-
ation of 3.73X. As the polarisation rule predicts, the//-plane and/T-plane couplings
alternate peaks and valleys as the orientation angle is changed. For large spacing the
parallel polarisation decays as 1/r2, or 12dB for each doubling of distance r, while
186 Planar arrays
the collinear coupling decays as 1/r. Coupling data for pyramidal horns of a separ-
ation of 5.93X are shown in Fig. 10.38. Hamid36 used GTD to calculate mutual
coupling between sectoral horns with 3X apertures; the coupling showed a typical
peak and null behaviour with centre-to-centre separations, with the envelope vary-
ing roughly as coupling ~ (— 49 — s/X) dB.
-AOr
-50
-60
CO
en
c
"Q.
8 -70
-80
-180 -90 + 90 + 180
(10.95)
and from the current vector the active impedance of each element in the array is
determined. When the solution is desired for several scan angles, so that only the
drive vector changes, but not the Zzy matrix, a simultaneous equation solver can be
used that stores a diagonalised matrix, allowing a rapid solution for new vectors
after the first. From the current vector (active-array excitations) of course array
patterns can be calculated. The impedance-matrix solution has an intrinsic advant-
age: array excitation tapers can be incorporated directly into the drive vector, unlike
the infinite array methods of Section 10.3.3 where the excitation must be uniform.
Figs. 10.39 and 10.40 show active resistance for the centre element of a 7 x 9 array
of half-wave dipoles, in free space, and quarter-wave over a ground plane. Also
shown in Fig. 10.40 is active resistance for comparison for a very large array. It can
be seen that small arrays add an oscillatory behaviour to the infinite array curves.
Active reactance for the centre dipole of an array over a ground plane is given in
Fig. 10.41, where again the oscillations can be observed. Active impedance of
dipoles near the edges behaves differently, but almost no data has been published.
200 r
80
Fig. 10.39 Active resistance of 7 X 9 array of half-wave dipoles with 0.5\ spacing (courtesy:
B. L. Diamond)
200 r
^--^\ H-plane
0 20 40 60 80
6,deg
Fig. 10.40 Active resistance of dipole array over ground plane [courtesy: Diamond)
infinite array
Fig. 10.41 Active reactance of arrays of half-wave dipoles with 0.5 X spacing [courtesy:
Diamond)
65X149
9 X11
Planar arrays 189
determined from an infinite array. An infinite array with the proper spacing, scan
range, and bandwidth is then designed as described elsewhere in this Section, with
the result of mode voltages and reflection coefficient vs. u. A Fourier inversion is
used to obtain the coupling coefficients, which are then summed to give a set of
mode voltages and reflection coefficient for each element in the array. Only the
centre elements in a large array experience essentially all the coupling; elements at
or near the edge employ fewer strong coupling coefficients as they decay rapidly
with element separation except near grating lobe incidence. Effects of amplitude
taper can be determined simply by including the amplitude coefficients in each
sum. From the reflection coefficient of each element the active impedance can
be written, and the pattern of the array is found by adding the modal array patterns
which use element modal patterns and mode voltages. This multi-mode finite
array solution has been extended to two-dimensional slot arrays by Woody and
Hansen.109
Active element patterns are of interest for small arrays although it is no longer
valid to multiply the array factor by a single active element pattern in order to get
the overall active array pattern. Fig. 10.42 shows calculated and measured active
element patterns of the centre dipole in a 5 x 5 array of half-wave dipoles in a half-
wave lattice, and quarter-wave over a ground plane. The broadening and flattening
: / *
Vw
/calculated - • measured \
A
10-
dB i
• E-plane
1 I 20-"
1 I
i i i I i
s 5
90* 60 3 0 * 0 * 30* 60 90°
Fig. 10.42 Active element patterns of 5 X 5 array of dipoles over ground plane {courtesy:
Rupp)
effect on the E-plane pattern is quite noticeable. Figs. 10.43 and 10.44 are active
element patterns for the centre element of a 7 x 9 array of dipoles for various
lattice spacings. From these curves it is apparent that the active-element pattern
limits the scan-angle range. As more elements are added to the array the number
of oscillations in the pattern increases and the curves become smoother. Also
190 Planar arrays
the fall off becomes steeper. Figs. 10.45 and 10.46 compare centre and corner
active element patterns for the 9 x 1 1 array of dipoles previously discussed. The
corner elements do not show the symmetric mutual coupling effects that tend to
flatten the active-element pattern. Dummy elements can be added around the edges
of an array to be scanned if space is available. The general design procedure is to
design using infinite array parameters to obtain the basic array parameters, then
to refine this design to accommodate edge effects.
D x «D y -o.8X
Dx=Dy»0.7X
Fig. 10.43 H-plane active element pattern of centre element of 7 X 9 array of dipoles
{courtesy: Diamond)
isolated
dipole
Fig. 10.44 H-plane active element pattern of centre element of 1 X 9 array of dipoles over a
ground plane (courtesy: Diamond)
Fig. 10.45 E-plane active element pattern of 9 X 11 dipole array over ground plane {courtesy:
Diamond)
Corner
element
Centre
element
-80-60-40-20 0 20 40 60 80
0,deg
Fig. 10.46 H-plane active element pattern of 9 X 11 dipole array overground plane {courtesy:
Diamond)
especially for the array without ground plane. Figs. 10.47 and 10.48 show element
impedance by number with and without ground plane for scan angles of 0°, 40°,
45° for parallel dipoles. It appears that the edge elements in the scan direction are
more affected than the edge elements in the reverse direction, but the VSWR of the
reverse edge elements is higher owing to the grating lobe pointing in that direction.
A rough tabulation of the number of edge elements where the edge element
impedance is outside the centre cluster is shown in Table 10.4. It must not be
supposed that these results apply to planar arrays, but the trends should be useful.
Another method of analysing finite arrays utilises correlation functions (Zaghloul
192 Planar arrays
and MacPhie,111). First, we have the spatial cross-correlation function, which is the
integral of the aperture field of the nth elemenl times the total aperture field. The
Fourier transform of this is the cross-power-density spectrum between the nth
element and the array. Integration of this over the u-v domain gives total power,
from which active element impedance is expressed as a sum of terms, one for each
element. These terms contain unknown element aperture fields times a scan
independent integral. The unknown element fields can be expressed in terms of
reflection coefficient or impedance, so that a set of simultaneous equations results.
45°
112
cj
X
scan
angle 0°
3,49
2,50
16
40 56 72 88 104 120
Fig. 10.47 Linear array element impedance, ground plane (after Gallegro)
obtained from the spherical waves that comprise the exact near field. In terms of
computation and accuracy the correlation method does not appear advantageous.
The Carter method has an interpretive advantage in that all terms have a direct
physical meaning.
216 r
180
108
72
36
51
51
Fig. 10.48 Linear array element impedance. No ground plane (after Gallegro)
G«(«, v) _
(10.96)
Ga(p,o) ksjk2-klo-kl yo
Ba(u, v) __ _ k2—
(10.97)
Ga(p,o) , + ««-i
Table 10.4 Number of 'edge elements': grating lobe for 45° scan
Z- crater
U] = A/Ux V] = A/Uy
Here the expressions for H(oo and Hnm are identical with eqns. 10.118 and 10.123.
Wheeler correlated these terms with the u, v grating lobe points, and identified each
term as an impedance crater, a concept developed by Rhodes.82 The normalised
admittance becomes:
Planar arrays 195
n m
Ya(u,v)
(10.98)
Ga(o,o) Ga(o, o)
Here Ya/Ga is the admittance (impedance) crater which is a plot of the normalised
admittance (impedance) of an element in u, v space. Fig. 10.50 shows the crater for
a short dipole and Fig. 10.51 is the crater for a half-wave slot (or dipole). Wheeler
discovered that mutual-coupling effects with scan, and the effects associated with
GL onset, could be visualised by placing an impedance crater at each (inverse)
lattice point in the u, v plane. The term 'crater' arises because the resistance pre-
sents the topological form of a crater. Thus the grating lobe series is a link between
the current-sheet concepts discussed earlier and the periodic structure approach.
Fig. 10.50 Impedance crater for a small electric dipole, described by a contour map
{courtesy: Wheeler)
Resistance
Reactance
The grating lobe series can be written using open-circuited element patterns
instead of impedance craters. From the periodic cell calculations of impedance/
admittance in Section 10.3.3, the real part of active impedance is (Oliner and
Malech,;76 Wasylkiwskyj and Kahn96):
+Z 2
goc{6)\{ , • COS flgocffnm)]
(10.99)
dxdy g cos e [ cos0 nm ^ oc (0)J
196 Planar arrays
where the prime signifies omission of n = m = 0. This expresses the active resistance
as a sum of open-circuited element patterns evaluated at the grating-lobe lattice
points, for the visible grating lobes only. By using the relationship between the
active and open-circuited element pattern previously developed, the active element
pattern is written in terms of the sum of visible grating-lobe open-circuited element
patterns (Wasylkiwskyj and Kahn98):
(10.100)
cosdnmgoc(6)
ky
2k 3k 4k
/? / / / / / / / / / / / / / / / / /
3k
/ / / / / / /
////////////// rz
mmmmmm^^^^^^^^^^ kx
Comparing this result with that of Hannan,37 which is the same except that the
ratio of element patterns in the summation uses active patterns, thus the inference
is that an element pattern ratio as used is independent of the impedance termin-
ation of the other elements. That is, the ratio is the same for match-terminated and
for open-circuited element patterns. A further inference follows: that a matching
network may be connected between the elements and the generators without
changing these ratios. If the network allows T -> 0, then the active element pattern
will approach the ideal element pattern. For the case of no visible grating lobes, the
formula reduces as expected to:
Planar arrays 197
(10.101)
A"
where w = —jyu2 + v2—k2 and the fields are evaluated at the u, v grating-lobe
points. Mutual impedance between two elements can similarly be written as an
integral in wavenumber space of aperture distribution FTs, which can be reduced
to the aperture complex power equations developed by Borgiotti11 and Rhodes.81
T? = l-^j^\r\2dudv (10.103)
T? = 1 - - C\T\2du (10.104)
7T J O
Thus element efficiency is a measure of the effect of mutual coupling. But it is use-
ful in relation to the concept of 'ideal element'. For a square array lattice with
spacing of A/2 or less, no grating lobes exist. The ideal element power pattern, since
the ideal element has no reflected power for scan angles in visible space, is just cos 6.
However, there are array phasings that provide a beam outside the visible unit circle
inu,v space, and for these | F | = 1 since no radiation occurs. Thus the ideal element
in the non-grating-lobe case has mutual coupling, which changes F from zero to
unity as the scan vector cross the unit circle. If the lattice spacing is increased to
Vo3X so that radiation always occurs, as the scan vector crosses the unit circle, a
grating-lobe scan vector enters the unit circle. The ideal element pattern for the case
of grating lobes was first defined to be cos 6 (in power) out to the angle at which
the first grating lobe appears, and zero from that angle to 90° (Hannan37). A more
useful definition has been developed, based on the grating lobe series of the pre-
198 Planar arrays
vious Section. Now the ideal element pattern is the limit of the active element pat-
tern as 0 -> 0. That is, if a matching network were applied to the array terminals, as
the network allowed the match to approach a perfect match, the active element
pattern would approach the ideal element pattern. Recapitulating, with F = 0
(Wasylkiwskyj and Kahn99):
DxDy cos 0
iidedl(0) =
'cos0£ o c (0 n m )
1+1 (10.105)
where the prime indicates that the sum is over all visible grating lobes, but not
including the main beam (n = 0 = ra). For a linear array the ideal element pattern is:
Dx cos 0 (10.106)
8idedl(P) =
cos0g QC (0 n )
1
cosdngoc(d)
For example, consider a linear array of thin slots with spacing between X/2 and X.
The ideal pattern is:
gidedi(0) = Dx cos 0 for 0 < sin < l/Dx - 1
Dx cos 0
for 1/DX- 1 <sin0< 1 (10.107)
cos0
1 +•
COS0!
where cos2 0x = 1 — (sin 0 — l/Dx)2. Taking a specific example, let the spacing be
0.75X. Fig. 10.52 shows the ideal element pattern. Such a pattern would maximise
element efficiency.
0-8
0 7
06
^ 05
CD
~ 0u
Q.
0 3
0 2
0-1
0
0 0 1 0 2 0 - 3 0 - 4 0 - 5 0 6 0 - 7 0 - 8 0 - 9 1-0
sin 9
Applying eqn. 10.106 to half-wave dipoles yields the following results, where
angle 0 is zero at broadside.
Planar arrays 199
collinear, g ideal = 1 / 2
COS2 0
parallel, in plane of dipoles,
2 - sin2 0
1
parallel, in normal plane, gideal = (10.108)
2 - sin2 0
The collinear ideal pattern is isotropic, while the parallel in-plane pattern is roughly
like that of a half-wave dipole. Somewhat unusual is the behaviour of the parallel
normal plane pattern, which has twice the value at endfire. This is perhaps related
to the fact that apertures at endfire have twice the directivity as at broadside. Of
course, array designers attempt to avoid grating lobes; the utility of these ideal
element pattern concepts is in providing insight and understanding of the inter-
actions among element patterns, lattice and performance.
Returning to ideal element efficiency, eqn. 10.109 relates 17 to reflection
coefficient. In a space, where for a rectangular lattice of size a^ a2
(*! = 2nai sin 6 cos 0
<x2 = 2na2 sin 0 sin 0 (10.109)
1.0r
.2 0.5
0.5
element spacing,
Visible space is bounded by an ellipse in a space; the integral in eqn 10.103 is the
common area between the ellipse and the IT XTT cell in ex space. Fig. 10.53 shows
ideal element efficiency for an infinite array with rectangular lattice (Kahn52). Note
that for a square lattice with A/2 spacing, the ideal element efficiency is TT/4 as dis-
200 Planar arrays
1-0r
0-5
0-5 10
element spacing t a-j
Fig. 10.54 Ideal element efficiency for hexagonal lattice (courtesy: Kahn)
third network is a 2N-port which is the inverse of the first; it restores the array
excitation vectors, which have now been matched. An excitation in an infinite
array producing a beam in invisible space is represented by a singular (zero or
infinite) eigen excitation. For finite arrays such excitations produce very small or
very large eigen excitations. Attempts to match such eigenvectors produce tolerance
problems similar to those encountered in superdirectivity. Kahn53 has calculated
active element patterns for a linear array of uniform line sources. Fig. 10.56a and
10.56Z? are for an element spacing of A/2, for the centre element of 15- and 25-
element arrays. The solid line is the ideal cos 6 and the dashed line is the actual
element pattern. Fig. 10.56c is for the edge element in the 25-element array. The
edge-element pattern is broadened and reduced slightly in gain. Fig. 10.57 is for
a 25-element array with 0.6366X spacing. Again the solid curves are the ideal
element pattern discussed elsewhere and the dashed curves are the actual element
patterns. Centre, edge, and next-to-edge element patterns are in the three graphs.
M2 = 729
M2 = 121
M2 = 49
M2 = 9
M2 = 1
As expected the finite array size fills in the grating-lobe null somewhat. The
edge pattern is flatter, with reduced gain. In contrast, the next-to-edge pattern
has full gain but a rapid falloff. From these data there appear to be two to three
'edge' elements, which corroborates the experimental measurements of Gallegro
202 Planar arrays
(see Section 10.3.2.2). It was mentioned that invisible space excitations cause
extreme eigenvalues. For a 25-element array with spacing of 0.3183 X there are 17
eigenvalues of the order of unity, and eight that are very small. When the two
largest and all the very small ones are discarded, a usable match occurs, as evidenced
in the active element pattern of Fig. 10.58^. Use of all eigenvalues results in the
unstable pattern of Fig. 10.58&.
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 90
pattern angle.9 deg pattern angle.0 deg pattern angle 9. deg
b
Fig. 10.56 Active element pattern of linear array, d = 0.5 \ (courtesy: Kahn)
a N = 15 centre element
b N = 25 centre element
c N = 25 end element
cos 6 pattern
V
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90 ' 0 10 20 30 40 50 60 70 80 90
pattern angle. 9 deg pattern angle. 6 deg pattern angle. 8 deg
b
Fig. 10.57 Active element pattern of linear array, d = 0.637 \, N = 25 {courtesy: Kahn)
a centre element
b next-to-end element
c end element
ideal element pattern
Descattering the array via a network requires that the element transmit and
receive patterns be equal, i.e. that the elements are minimum scattering (see
Section 10.3.1). In practice, these remarks are useful for single-mode elements such
as slots and dipoles. A further requirement is that all mutual impedances be reactive
Planar arrays 203
where / m / 2 is the spherical Bessel function of order m/2. Pursuing the example
further for isotropic elements (m = 0) the root gives d = A/2, so that a half-wave
spaced array may be completely decoupled and descattered.
X>
j/> 1 0 10
o
Inor
5 0.5 S 05
Q. Q.
power
nn | ! 1 1 1 1 1 1 \(
w
0 10 20 30 40 50 60 70 80 90 0 10 20 30 40 50 60 70 80 90
pattern angle.Sdeg pattern angle, 8 deg
b
When a grating lobe just enters visible space, the active element pattern has a null
(at that angle), or in active-impedance terms the reflection-coefficient magnitude is
unity. Such a blind spot cannot be removed by a matching network (Wasylkiwskyj
and Kahn"). However, for those blind spots that occur at scan ranges inside the
grating-lobe scan contour (in u-v space), a suitable network can improve or even
remove the null, provided that the open-circuited element pattern does not have a
null at that angle.
10.3.3.1 Slots: The edge effects that make the analysis and understanding of a
finite array difficult are removed when the array is made infinite: the finite array is
204 Planar arrays
cell
feed
wavegi
where the subscript o is for the dominant unit cell mode. Assuming a cosine slot
field with slot dimensions as, bs and feed waveguide dimensions a, b, the mode
voltages become:
kxoas . ky yobs
cose -££-=• sine (10.114)
where the unit cell dimensions are Dx = dx/X and Dy = dyj\\ the uniform excit-
ation space factor is:
sin A
sine A = (10.115)
206 Planar arrays
These symbols simplify the formulas while allowing the various factors to be inter-
preted physically by inspection.
The active conductance, normalised to the feed guide characteristic admittance
is then:
(10.117)
2
2DxDyX> cose *f
where the scanning factor is:
1 — sin2 0O cos2Y
Hoo = ' "" T ° (10-118)
COS0 (
This will later be seen to be the dominant form of a more general scanning factor
Hnm-
For //-plane scan (scan parallel to the slots) it reduces to cos0 o while for
Z?-plane scan it becomes 1/cos 6O. This is in accord with the transmitting current-
sheet concepts presented earlier. The factors involving kxoas and kyobs are distribu-
tion factors representing the effects of the field distribution along and across the
slot, respectively. And the remaining factors can be separated into a factor for a
single waveguide fed slot and a mutual coupling factor:
3X3 cose 2 — -
2a
J
active _ ^ u
G
single slot
If the array lattice spacing is X/2 each axis, the active conductance is 3/TT times that
of a single slot. This, of course, is valid only at broadside where the scan factor is
unity.
When the combination of scan angles and lattice spacings allows one or more
grating lobes to appear, the conductance will have an added term for each GL. A
grating lobe appearing due to //-plane scanning sets up a higher TE mode in the
unit cell, and the slot conductance changes smoothly at grating lobe onset. This is
because the characteristic admittance of the mode is zero at GL onset. In contrast
the TM mode required by jE'-plane scanning has a characteristic admittance that is
infinite at GL onset, with the result that the conductance and VSWR have an
Planar arrays 207
(10.121)
J
The sums include all positive and negative integers except the term for n = m = o.
The generalised wave numbers are:
r n
kxn = A: sin 6O cos $ o 4- —
Dx
L
kym = k sin<9osin0o + ^ - (10.122)
[ Vy
and the generalized scan factor is:
»f
sin y o cos 0 O 4- —X/ I — 1
/ -2 ° 2 (10.123)
/(sin 0O cos 0 O H 1 4- j sin 0O sin <j)o H 1— 1
"V \ Ac/ \ ^y/
Instead of displaying active admittance data for various scan planes and spacings,
the active impedance of a dipole array will be discussed. The active slot admittance
and active dipole impedance will be seen to be simply related. Since most of the
available data have been computed for dipole arrays, it will be shown later. A dipole
array over a ground plane will be discussed as this includes the special case of no
ground plane.
10.3.3.2 Dipoles: The slot equivalent circuit is simple because the susceptance
is in parallel with both the feed waveguide and the unit-cell waveguide. Assuming
dipole feed terminals at the centre of the dipole, the reactance is in parallel with the
unit-cell waveguide, but in series with the feed terminals. Consideration of the
ground plane displays one of the advantages of the unit-cell approach. Whereas the
dipole array in free space is represented by a unit cell on each side of the dipole
208 Planar arrays
with both unit cells semi-infinite in length, the ground plane is simply represented
by a short across one of the unit cells at height h from the dipole. Since the ground
plane to dipole spacing is usually of the order of A/4, the higher-order unit-cell
evanescent modes are sufficiently damped out in travelling from the dipole to the
ground plane and back that they can be neglected. However, the ground plane
affects both resistance and reactance through the dominant mode. The dipoles are
assumed to be strip (flat) dipoles, so that the dipole array is the Babinet equivalent
of the slot array previously discussed. Fig. 10.61 shows the dimensions. The dipole
current distribution is assumed to flow only in the x-direction and to be constant
over the dipole in the y-direction. Along x a cosine distribution is assumed. A
match is then made between the electric field in the dipole plane produced by the
dipole, and that produced by a set of unit cell waveguide modes. From this, the
series for active reactance is obtained. For the active resistance the ground plane
is replaced by an image dipole at distance 2h. The results obtained originally by
Stark87 have been cast into the unit-cell format and symbology. Wave numbers
are the same as for the slot case. The active resistance is:
480a 2 kxoas kyobs
Rn = cose2 sine2 (sin2 kzoh)Ho (10.124)
J L
Dy = dy/X
When a ground plane is not present, the sin2 kzoh factor is replaced by \. The
expression for reactance is:
240a2 2 JE £ 2 2
2 I cose - r- sine - ^ (sin kzoh)Ho
D
7tDxyJJyvA I I I
(10.125)
Planar arrays 209
Symbols are the same as before. Absence of the ground plane is handled by putting
h = 0.
Next, dipole active impedance is examined. Keep in mind that slot active admit-
tance can be obtained from the active impedance of a dipole without ground plane
simply by multiplying by a scale factor. Most of the data which follow were calcu-
lated by Diamond and are shown by courtesy of Oliner and Malech.76 Figs. 10.62
and 10.63 show active resistance normalised to unity at broadside for half-wave
3 r
E-piane
20 40 60 80
8 , deg
H-plane
0 20 40 60 80
8,deg
These two Figures are for half wave spacing, so no grating lobes occur. The next
two Figures, Figs. 10.64 and 10.65 show active resistance for dipoles over a ground
plane for lattice spacing of 0.6 and 0.7 wavelength. The solid curve is from periodic-
structure calculation, while the circles use the Carter element-by-element approach
on a 65 x 149 element array. The abrupt discontinuity in//-plane active resistance
is noteworthy; in the 2^-plane only modest change of direction occurs, while in the
220 -
200 -
180 -
1
I \ \
160 - / , H Plane \
140
-e-120
~ 100 /D Plane
- v
a
Q:
80
E Plane / \ \
60 Scant 0-0') \
40
Grating Lobe
Boundary—**
20
\
n i i i i
20 40 60 80
8,deg
diagonal plane the values are little affected. Agreement between the element-by-
element and periodic-structure approaches is excellent except for angles beyond
85° where the edge effects become significant. Since the dipole element pattern in
the //-plane does not go to zero, the edge effects are expected to be larger than for
/^-plane scan. Since curves for dipoles without ground screen, with element spacings
above 0.5X, have not been presented it should be recalled that they can be obtained
from the curves of dipoles with a ground screen by dividing out the 2 sin2 (kh cos 6)
factor. Note that the behaviour of active impedance with scan plane is a function of
Planar arrays 211
element pattern rather than element polarisation. Thus the singular behaviour is for
scanning normal to the radiating elements, which is //-plane for dipoles but is
/T-plane for slots.
Turning now to the active reactance, the computation involves a double infinite
series, and one must always be concerned with convergence questions. The periodic-
structure series converges slowly. When no ground plane is present, the element-by-
element calculation converges very slowly (a large array is required to approximate
the periodic structure results), but with a ground plane it converges more quickly.
180
160
UO
120
: 100
3
80'
60
D-plane scan
W
20
0,
20 40 60 80
e.deg
Fig. 10.65 Dipole array over screen active resistance
o Carter
periodic structure
Dx = Dy = 0.7
trol the //-plane reactance singularity at grating-lobe incidence (90° for this case).
Figs. 10.70 and 10.71 show active reactance for lattice spacings of 0.6 and 0.7
wavelength; the singularity at GL onset is apparent.
From all these data it is apparent that arrays with large scan ranges will experi-
ence large changes in active-element impedance. Methods of reducing these changes
will be discussed later.
vO-005
005
-10
-20
Fig. 10.66 Dipole array reactance, H-piane scan showing dipole-width effects
250
200
150
100
50
0'
20 40 60 80
8,deg
Fig. 10.67 Dipole array reactance; E-plane scan showing dipole width effects
30
c j 20
X
10
80
0
20 40
0 , deg
-10
Fig. 10.68 Dipole array reactance; diagonal scan showing dipole width effects
180 /
/ ' '
160
III 1
140
!1 h11
H Plane Scan
1 Term
1« 1 9 Te r m s
III 1 101 TV} r rn T
N ' 441 To rm r
ill
a
j
100 a
X
o ill1 '
in i
80 n i
11 i
I i
60
/" '
Jr / /
E Plane Scan
1 Torm
40 '^y' / / / / — 121 Terms
— 441 Terms
0
20 40 60 80 100
e deg
Fig. 10.69 Dipole array with ground screen active reactance, showing effect of number of
terms
element-by-element method
periodic structure method
. = D v = 0.5
214 Planar arrays
200
180
160
H-plane scan
U0 grating lobe
(0=90°) boundary
120 D-plane
scan
180
160
D plane scan(<J>=45°)
UO
120
t
x 80
/ H plane scan
(4>=90°)
60 /
40 E plane scan (c^zO^ 0 ^
20* -
^ g r a t i n g lobe appears
1 1 i I i I i i
°c) 20 40 50 80
0. deg
active-element pattern null, or blind spot. The leaky-wave pole gives an approxi-
mate behaviour of cancellation around the blind angle, where F is expanded into a
Taylor series about the pole. Fig. 10.73 shows the active reflection coefficient for
an array of narrow slots fed by parallel-plate waveguides and covered by a dielectric
slab, for iT-plane scan. This curve was calculated using 21 modes. Slot spacing was
X/2, waveguide width was 0.3X, while the slab had e = 2.56 and thickness 0.2X.
Using the leaky-wave pole Knittel et al.59 calculated the dots shown in the Figure,
where the leaky-wave Q is given by:
u2 Re (kjk)
2 (10.126)
l-u 2lm(kx/k)
i-o
0-8
0-6
0-4
0-2
Fig. 10.73 Blind spot of array of slits (courtesy: Knittel, Hessel and Oliner)
kx is the wavenumber along the array. For this array the grating lobe occurs at
90°, and the active reflection coefficient is also unity there. Another example
used a brick array of rectangular guides 0.905 by 0.4X, on a square brick lattice
of 1.008X spacing. An array of only 7 x 7 elements exhibits a blind spot over
26 dB deep in the //-plane at 27° (Farrell and Kuhn26 ). A grating lobe occurs on
the 60° circle in u-v space; the blind spot is usually just inside the grating lobe,
but in this example it is well inside (Oliner75). Other discussions of blind spots are
in Section 10.3.2.
The understanding of how blind spots arise also gives indications of corrective
measures to take (Knittel;58 Oliner75). These include reducing dimensions to
Planar arrays 217
preclude/eliminate higher propagating modes, alternating external structure or
external impedance, and changing various parameters to obtain a suitably low
VSWR over a wide scan-angle range.
* The dominant TE mode in a rectangular waveguide can be decomposed into two plane waves
that reflect alternately from the narrow walls
218 Planar arrays
103 A Scan compensation
H plane
(unaffected)
Fig. 10.75 Active resistance of dipole array over ground plane with baffles {courtesy:
Edelberg and Oliner)
without baffles
— with baffles
Fig. 10.76 Slot with parasitic elements {courtesy: Clavin, Huebner and Kilburg)
It is also possible to effect the required changes through use of an external spatial
filter (Mailloux69). Here, of course, the spatial filter not only has a transfer function
220 Planar arrays
that varies with 0, but also with polarisation. Such filters have been developed by
Munk and colleagues.71'72 The filter consists of several contiguous flat dielectric
sheets, placed over the array face. Since the plane-wave transmission coefficient of
a dielectric sheet varies with both angle and polarisation the potential is obvious.
To analyse this configuration, and to design a compensation unit, the field from an
infinite 2-dimensional array of elements is expanded into a 2-dimensional spectrum
of plane waves, weighted by the open-circuited element pattern. A single plane-wave
component has a transmission coefficient through the sheets that is obtained from
the cascading of impedances. Each sheet has its own thickness and dielectric con-
stant, and low loss; the transmission coefficient is r, found from the impedance
cascade. The sum of the weighted plane waves over the array elements times the
transmission coefficient gives the radiated field (active element pattern); if the
sheet parameters are optimally chosen the result might approach cos 6. The field
of a uniformly excited array of dipoles, for example, is:
JZ
Here the dipoles are along x in the x-y plane; h is the isolated dipole pattern; for
details of the wavenumbers see Section 10.3.3. The scan vector i$axu + ayv, where
the aX9 ay are unit vectors. This bundle of inhomogeneous plane waves is propagat-
ing when kz is real; for imaginary kz the waves are evanescent. / is the realised
dipole current.
The transmission coefficient r of an exfoliated dielectric as mentioned is found
from the cascaded impedances; a simple way of calculating this cascade uses an
A-B-C-D matrix or T matrix for each layer, since the matrix of the cascade is the
product of the matrices of each layer. The T matrix is loosely related to trans-
mission coefficients, so it is used here. With slabs 1-rc, with slab 1 at the array face,
the overall T matrix is:
T = 7V T V , . . . TV TV (10.128)
However, more than one T matrix is required per slab, so the impedance cascade
contains more factors. The interface from free space to a slab, or from one slab to
a contiguous slab, requires an impedance transformation matrix. A transmission
line also requires a matrix. Thus n slabs utilise a cascade with n line matrices inter-
spersed among n 4- 1 transformation matrices. When the transmission line is a
dielectric slab, the incidence angle and polarisation must be included. The con-
stituent matrices are (Montgomery eta/.;70 Ramo et a/. 79 ):
-i(H KH.
Planar arrays 221
e cos 6
r = parallel polarisation (10.130)
— sin 2
COS0
= , ==r normal polarisation
Ve-sin 2 0
lossless
dielectric T = (10.131)
slab
Obviously the slab thickness is / and the angle of propagation from normal is $.
Given the parameters of the sheets, the overall T matrix is easily computed. The
transmission coefficient r = S21 a n d f° r reciprocal slabs S2i = Si2« Since T22 =
I/S12 the transmission coefficient is immediately:
r = (10.132)
T = (10.133)
1 - r2 exp (-/2/c/Ve-sin 2 0)
Here r is the air-dielectric interface Fresnel reflection coefficient. Note that the
insertion-phase delay term which allows a virtual straight-line path is not included
inr.
Returning to the electric-field expression, r is inserted into the double sum.
However, for general scan, the polarisation of the elements must be resolved into
parallel and normal components, with each multiplied by an appropriate r and
element pattern. If now the dielectric slabs can be chosen to make the sum approxi-
mately cos# over a scan range, the array is compensated. Munk, Kornbau and
Fulton have matched an array of short waveguide slots by a single dielectric slab of
e = 1.33 and r/X 0 = 0.367, with a VSWR < 1.5 for 6 between 0 and 80° in both
principal planes. This assumes that the self admittance of the slot has been matched
separately at broadside; the slab produces a match vs. scan angle. Work is in progress
on slab matching of resonant slots and dipoles. At the stage of writing, the band-
width behaviour has not been determined. This technique is significant and promis-
ing in that it utilises single-mode elements, which are simple and inexpensive.
222 Planar arrays
10.4 References
Conformal arrays
Giorgio V. Borgiotti
11.1 Introduction
Conformal arrays can be defined in the most general terms as arrays that are non-
planar. However, often a more restrictive definition is used that postulates an array
whose elements are flush mounted on a non-planar surface.12 It is implied in this
definition, adopted here, that the presence of the surface on which the elements are
located plays an important role, affecting the radiative and impedance properties of
the array. Accordingly, the study of non-planar radiating arrays in free space, such
as the circular arrays of dipoles, a subject of thorough investigations in the past,22
is not considered to be within the scope of this Chapter.
Conformal arrays are, in general, more difficult to analyse than their planar
counterparts. Notwithstanding, conformal arrays are considered attractive
alternatives for certain applications for which planar arrays have definite draw-
backs. Specifically, great potential exists in the following two areas:
(a) Ground-based systems, requiring a wide angle or omniazimuthal coverage
(b) Arrays mounted on missiles or high-speed aircraft.
The maximum scan angle from broadside for a planar array cannot be made much
greater than 60°, because of the gain losses due to both the reduction of the pro-
jected aperture in the scan direction and the aperture mismatch. Consequently,
obtaining a wide-angle coverage by using planar arrays requires the use of a multi-
face system. This solution has been adopted in ground radar systems (Fig. 11.1).
However, the potential offered by conformal arrays in this area is tremendous, a
conformal array on a spherical surface (Fig. 11.2) or on a wide aperture cone being
in principle a solution better matched to the geometry of the problem.
For missile borne and airborne applications, conformally flush-mounted arrays
have the advantage of not introducing any additional aerodynamic drag and also of
providing wide-angle coverage, if so required. Other advantages are:
• Less space is required in the missile body, because the radiating elements are
pushed far forward in the missile nose, leaving space to package the phase
shifters and control circuitry.
228 Conformal arrays
Fig. 11.1 Planar array multiface system for wide-angle coverage (courtesy: Raytheon Co.)
11.2.1 General
The tremendous progress of microwave component technology in recent years,
particularly in the area of phase shifters, has made phased arrays a practical solution
for scanning antenna systems. For planar arrays effective analytical tools have been
developed for performance prediction, especially for the evaluation of the gain
230 Con forma/ arrays
vis-a-vis scan angle. On the other hand, methods of analysis and synthesis of con-
formal arrays have not been developed to the same extent as their counterparts for
planar structures.
The analysis of the electromagnetic properties of a conformal array is a problem
whose mathematical structure consists of two essentially separate parts. One is the
classical 'external' electromagnetic problem, that is the determination of the
electromagnetic field in the region exterior to a closed surface on which known
electric or magnetic currents exist. The solution of this problem is known for
simple geometries. The other problem consists of determining the system of
currents on the radiators mounted on the conformal surface when the radiators are
excited via a certain feed network. They key quantities for these calculations are
the scattering coefficients between elements (Section 11.2.2) whose evaluation
requires the inversion of a matrix of a large dimensionality with complex elements.
This is a difficult numerical problem and can be circumvented in certain cases by
using appropriate approximate methods (Section 11.2.4 and 11.4).
In general, the methods of conformal array electromagnetic analysis can be
broadly categorized into two classes:
(a) modal methods
(b) high-frequency asymptotic (ray) methods.
Method (a) is rigorous. However, its area of applicability is restricted to certain
canonical structures having a high degree of symmetry. Nevertheless, the method
is very valuable because the structures that can be analysed exactly are by no
means trivially simple and may in fact constitute accurate analytical models for
practical antennas. For example, the cylindrical periodic array of apertures is a
complex structure that can be studied rigorously (Section 11.4). Moreover, in other
cases, a structure that may be too complex to be modelled 'globally' by using a
canonical geometry can, however, be adequately represented by it in a limited
region of interest.
Approach (b) is a high-frequency method. The radiation of each array element
is represented through a ray system. The method is useful to solve complex prob-
lems not amenable to modal analysis, or for which modal analysis leads to difficult
convergence problems, as, for example, with conical arrays.
22 A
o
- T_ ~- B
Without any loss of generality, each radiator is assumed to have only one input
port. In fact, if the elements have two excited ports (for polarisation diversity or
polarisation control, for example) the situation will be accommodated simply by
232 Conformal arrays
F(6,d>) = £ ^ ( 0 , 0 ) (11.3)
The array RGP is expressed in eqn. 11.3 via the element RGPs and the array free
excitation.
The element RGP is a particularly useful concept for large cylindrical (and, of
course, planar) arrays with elements on a regular grid.4 In fact, in this case, the
RGPs are identical - a coordinate rotation apart - except for those atypical
elements located near the array edge. In more general cases, the evaluation of the
element RGPs may not be a convenient intermediate step for the calculation of
the RGP of the array. An alternative characterisation of the array properties
employs, along with (b) and (c) above:
(al) the magnetic or electric currents on the array elements when all the elements
are excited by the array free excitation. This characterization is completely
equivalent to that using (a) since, in both (a) and (al), the internal and external
properties of the structure have been taken into account. Once the element currents
have been calculated, the problem is reduced to the evaluation of the radiation of a
known current in the presence of the array surface.
The procedure for the determination of the magnetic currents on aperture
elements will be developed in its general conceptual lines in Sections 11.2.3, 11.2.4
and 11.2.5, the details pertaining to specific geometries being discussed in Sections
11.3 to 11.6.
r, = ^ (n.6)
at
and called 'active reflection coefficient' at the /th element.
Suppose that 'looking into the network', the elements ports are not matched
and/or the elements are not uncoupled in the sense that the energy fed into one
element is not routed only to the array input terminal but is coupled also to the
other elements and reflected back to the excited element itself. These 'internal'
couplings due to the structure of the FN are graphically indicated in Fig. 11.3 and
and denoted by S-^. It is apparent that if the internal couplings are different from
zero, the free excitation a depends not only upon the transmission set of
coefficients Tt between the array beam ports and the element ports but also upon
the reflected waves b. In other words, if the aperture is not perfectly matched, part
of the energy associated with a is reflected back into the feed network, and, if
there exists internal coupling, back into the aperture again. The process iterates and
multiple reflections occur, affecting the amplitude and phase distributions, that
become very frequency sensitive. Denote by A the excitation at one of the beam
ports of the FN, the other beam ports being match terminated. Simply because of
the linearity of the network the following relationship holds:
a = SNb + TA (11.7)
where T is an N component vector and SN is the iVth-order square matrix represent-
ing the internal couplings between the elements. The physical meaning of T and SJV
is the following. The /th component Tf of T is the normalised voltage at the
terminal of the /th element, when the elements are removed and replaced by
matched loads. An element S^ of SJV is similarly the complex amplitude of a wave
incident from within the network to the input terminal of the kth element when
the elements have been removed and replaced by matched loads and the beam ports
have been match terminated, whereas the /th element has been replaced by a power
generator of unitary power. By using eqn. 11.5, b is eliminated from eqn. 11.7,
obtaining thus:
a = (I^-SJVS0)-1^ (11.8)
where lN is the unit matrix of order N. If the network has no internal coupling, i.e.
if
$N=0 (11.9)
then
a = TA (11.10)
and, in such a case, the free excitation a does not depend upon S°, i.e. it is not
affected by the element mutual couplings. If instead SN is different from zero, the
antenna illumination becomes very frequency sensitive, because so is SN in practical
cases. A physical insight of the phenomena is obtained by resorting to the following
expansion:
2
+ -.- (11.11)
Conform aI arrays 235
Ground
TEn
plane
nominal pol.
Assume that the first M waveguide modes (ordered with nondecreasing cutoff
frequencies) are sufficient to represent with good accuracy the transverse electric-
field distribution on the element apertures. The choice of the number of modes is
made on the basis of physical considerations and can be validated through con-
vergence tests; i.e. by repeating the calculations of the quantities of interest with an
increasing number of modes and stopping when the results become insensitive to
any further increases.
The superscript k will be used to identify the various aperture modes. The index
k runs from 0 to M— 1, the value 0 being used for the fundamental excited mode.
The modal voltages for the fundamental excited modes are the sum of a set of
incident and reflected waves:
1/2
°7
F°7 = a
= (11.13)
YQ being the characteristic impedance for the waveguide fundamental mode.
With a parallel notation the voltages V* for the passively excited modes can be
represented by the set of vectors, having a dimensionality equal to the number of
elements N:
Vk = {¥/*}, (k = O , . . . , M - 1 , i = 1,...,AO (11.14)
When the array is free excited by the set of waves represented by the vector a, the
following relationships hold, because of linearity:
Con formal arrays 237
1/2 k
Y0- V = 1) (11.15)
fe
6Ofe being a quantity equal to unity for k = 0 and zero for k ^ 0 and S being square
matrices of Nth order. The modal voltages at each element aperture, i.e. the mag-
netic current distribution on the array surface, can be evaluated, (under the Mmode
approximation), for an arbitrary free excitation if the elements of the matrices
Sk, called generalised scattering coefficients, are known. The generalised scattering
coefficients, in turn, can be evaluated from the mutual admittances between the
modal components of the aperture magnetic currents as discussed hereafter.
radiation network
mode '0
element
P mode *1
mode *2
mode' 0'
element
mode ' 1'
q
With reference to Fig. 11.5 for k ^0 the meaning of an element Spq of the
N xN matrix Sfe is the following: If the element labelled here by the subscript V is
excited by a free excitation ap - 1, then the modal voltage for the kth mode for the
element labelled 'p' is SpqY0~1/2ap. An intuitive appreciation of the nature of the
mutual coupling phenomena can be possibly achieved by thinking of the radiation
network in Fig. 11.3 as a network having N xM ports, rather than only N, withiV
ports connected to the feed network and (M — 1) x N ports terminated by either
modal admittances (for higher-order modes), or by an admittance determined by
the structure of the transition from the feed network to the element waveguide
(for the cross-polarised fundamental mode), Fig. 11.5.
238 Conformal arrays
JJ (11.16)
the integral being extended over the pth aperture. Let hlq(s) be the transverse mag-
netic field on the array surface generated by a magnetic-current distribution on the
qth aperture. The magnetic current is assumed to have the functional form corre-
sponding to the i th mode with unit modal voltage, that is, having the form elq(s) x ft
with h the normal to the array at the location of the qth aperture. In evaluating the
magnetic field, the apertures are considered absent; i.e. the metallic continuity of
the array surface is imagined to be reconstituted. The admittance between the
modes k and i (possibly equal) on the apertures p and q (possibly coincident) is
defined by the expression:
(11.17)
For a fixed pair of modal indexes 4fc' and H\ the quantities YpQ form a matrix of
order N denoted by Ykl. The following matrix of M x N rows by M x TV columns
can then be formed:
Y oo Y01 ... Y0^"1
(11.18)
yM-1,0 rM-\,M-\
- 0 0 Yh J
w
The diagonal elements of D are all equal to the modal admittance viewed by the
&th mode 'looking into' the element (Appendix 11.8.1). Form the matrix
Con forma/ arrays 239
r D (o) 0
0 0
D = (11.20)
- 0
s1
s = s2 (11.22)
where the tilde (~) indicates transpose and the broken lines stand for a matrix of
order ( M - l ) x J V .
Direct inversion of the matrix in eqn. 11.21 is difficult except for very small
arrays. Assuming, for example, a number of elements N = 300 and a number of
modes M = 3, a relatively simple characterisation of the element magnetic currents,
then in order to identify the submatrices Sfe of S, the inversion of a complex matrix
of order N xM= 900 is necessary. This is a difficult and time-consuming task even
with present high-speed computers.
The problem can, however, be circumvented by using a series-expansion method
for the evaluation of eqn. 11.21. Consider the identity:
where 1 is the unit matrix of order M x N. Assume that the norm of the matrix in
the inner parentheses in eqn. 11.23 is 'small' in the following sense:
-D"1Y)|i< 1 (11.24)
It can be shown that eqn. 11.24 holds in any practical case, except for large
element mismatch. Then the expansion
(11.25)
is valid and when inserted in eqn. 11.21, taking into account eqn. 11.23, gives:
S = (11.26)
240 Con forma I arra ys
not requiring any matrix inversion. For cylindrical arrays having a large number of
elements arranged in a regular grid, it is possible to avoid large matrix inversions by
resorting to a different approach that uses the high degree of symmetry of the
structure and neglects array edge effects. The philosophy is similar to that widely
used for planar array analysis. A discussion and numerical examples are given in
Section 11.4.
In Section 11.2.5 the formal expressions for the generalised mutual admittances for
aperture elements were given. A procedure was discussed to obtain the element
mutual couplings and the magnetic current distribution on the array surface under
the assumption that M aperture modes represent adequately the radiator transverse
electric-field distributions. This formalism will be applied in this Section to the
study of an array of apertures on a conducting circular cylinder. To provide the
necessary background, certain well known results concerning the field of a magnetic
current flowing on a cylindrical conducting surface will be briefly recalled. The
cylinder is assumed infinitely long, that is no edge diffraction effects are here
considered.
\\ez{hz)ed$dz (11.31)
J J
2TT
Conformal arrays 241
the integrals being extended to the aperture, and u and w being the variables conju-
gate to £, z in the FT. Eqns. 11.30 and 11.31 can be lumped together and written
concisely as follows:
l
'*dA (11.32)
aperture p
where
u=(u,w) (11.33)
is a position vector in the u, w plane and
E(W) = e^{u)4>^ez{u)z (11.34)
The procedure for the determination of the field in the region p>R is
242 Conformal arrays
and
(11.36)
where & is the free space propagation constant and
£ = y?k2-w2 (11.37)
with the branch of the square root chosen to satisfy radiation conditions, i.e.
\\/k2-~w2\ for \w\<k
, o (1138)
~j\\/k2 -w2\ for |iv|>fc
) and /5/^2)'( ) are the Hankel function of order n and second kind and its
derivative, respectively.
Expressions for the tangential magnetic field on the cylinder surface p = R are
similarly established. By using the coordinates £, z,
(1L40)
jk H^(R^)}
Y^ In \\
which identify the currents on the conducting surface of the cylinder, T? denotes the
characteristic impedance of free space.
the integration being over the aperture of the reference element whose centre is at
z = (j) = 0. More explicitly by introducing the circumferential and axial components
of the field:
Introduce for the FT of the transverse component of the fcth mode the following
notation, in accordance with eqn. 11.32:
Explicit expressions of the FTs of the mode functions for rectangular and circular
waveguides are given in Appendix 11.8.2. Introduce in eqn. 11.43 the expressions
11.39 and 11.40 for the transverse components of the magnetic field. The
expression for the generalised mutual admittance becomes;
\ )\~e ^'
Rt;Hpm)6'[R'W|
using eqn. 11.45 for the generalised self admittance only (sPQ = 0), and using
instead in eqn. 11.43 high-frequency asymptotic expressions for the magnetic field
as discussed in later Sections. For the self admittance a further simplification con-
sists of replacing the ratio of the Hankel function and its derivative with its asymp-
totic value for large arguments:
(1L47)
(1L4S)
" = i^7i
The unit vector in the direction normal to the surface at the points is denoted by
h(s). The binormal to the geodetic ray at s is defined by
i(s-s')xh(s) = b(s-s) (11.50)
whose components are
b^s-s) = tM(s-s) (11.51)
bz(s-s) = -rt(s-s'). (11.52)
Introduce two functions Hb(s) and# t (s), where
Con forma/ arrays 245
(11.53)
The expressions for Hb(s) and Ht(s) will be given later. Their physical meaning is
that of the components of the GTD dyadic Green's function for an element of mag-
netic current tangential to the cylinder.14 The tangential magnetic field on the
cylinder surface excited by the /th mode on the reference aperture is then given by:
Fig. 11.7 Field at s is the sum of two ray contributions propagating in the directions t and tl
I' and z being the coordinates of a generic point of the reference aperture. The
radius of curvature is constant on the ray geodetic path and is equal to
R
Rt = (11.55)
cos2 5
where 6 is the angle formed by i and the % axis. The radius of curvature in the direc-
tion of 5 is also constant and equal to
R
(11.56)
sin2 6
246 Conforms/arrays
/ * \1/3
7 = s (1L57)
(ai?)
where k is the free-space propagation constant. Introduce the function
k2 e~iks
G(s) = — (11.58)
and the Fock functions9 u{ ), p( ), w'( ), v'{ ) whose expressions are given in
Appendix 11.8.3. Then the longitudinal and transverse GTD Green's functions for
8 ¥=n/2 are given by:
( |) J (11.59)
- \I-J- (41
[ ks \ksf J
(11.62)
which are identical to the exact expressions of the b and i components of the mag-
netic field of an elementary magnetic dipole on a ground plane having the orient-
ation of b and i, respectively. For 6 -> TT/2
•is) (11-63)
^9 i -64)
It can be shown easily from eqn. 11.64 that in the axial direction the field on the
cylindrical surface is stronger than on a plane. For very large r, i.e. when the
observation point is in the deep shadow,
Con formal arrays 247
Ht(s)~-j-Hb(s) (11.66)
In the deep shadow the field is a slow wave decaying exponentially along the sur-
face ray.
in eqn. 11.54. Since there is clearly no reason to use the double subscript 'p, q\ it
will be replaced from now on by the single subscript ' p \ By using the fact that
f, h9 b form a normal trihedron, from eqns. 11.43 and eqns. 11.51-11.54 one
obtains for the GTD equivalent of eqn. 11.45:
Yk%) =- j | A d%dzek{s)'\\A,d%tdzti{s-sp-st)Hh{\s-sp-sf\)
d%'dV''b{s-sp-s')
01-67)
which is the generalised mutual admittance for the ith and fcth modes for two aper-
tures whose relative position is identified by sp.
z
Is —Sp —sV is — sp — s\
where
and the components of b(s — spq — s) are given via eqns. 11.51-11.52. For the
fundamental mode the direction of the electric field on the aperture is constant.
Thus, for axially polarised modes,
xcos-- (11.72)
In a parallel way, the mutual admittance between two axial slots can be established.
The straightforward details are here omitted. In Figs. 11.8-11.10 numerical
examples are given,* and also a comparison is made between the results obtained via
160
Fig. Mutual admittance Y00 between the fundamental TEl0 modes for two circum-
ferential slots as a function of the azimuthal separation <f>0. Vertical separation
mag.
GTD solution
phase
o exact modal
GTD and modal expansions. Notice that the phase in Fig. 11.8-11.11 is normalised
by multiplying Y00 by the geometric phase term exp (jksp).
The explicit expression of eqn. 11.67 is not as simple as eqn. 11.72 for modes
having the polarisation of the electric-field variable over the aperture. This occurs
for all modes of circular waveguides and for some higher-order modes of rectangular
waveguides. However, the numerical evaluation is straightforward.
-60 85°
05
o
-70 80° £
-80 75° E
-90 70°
-100 JJ65°
z o /A
m
Fig. 11.9 Mutual admittance Y between the fundamental TEl0 modes for two circum-
ferential slots as a function of the vertical separation z0. Azimuthal separation
Yki(sp) - -
z\\bp-jj (11.78)
-70! -100°
-90 -120°
•o
-110
-130 -160°
-150 -I ~Q -180°
Fig. 11.10 Mutual admittance Y00 between the two fundamental TElQ modes for two axial
slots as a function of the vertical separation z 0 . Azimuthal separation 0O = 0
— mag.
GTD solution
phase
Exact model o
By introducing the FT of the mode functions, eqn. 11.44, expr. 11.78 becomes:
polarised) and the mode TMOi on two apertures whose relative location is indicated
again by sp. In the expressions for the FT of the mode functions (Appendix 11.8.2),
according to eqn. 11.79, the following replacements are made:
t<-k (11.80)
and
H<-80 (11.81)
for the second aperture, and
(11.82)
for the first aperture. Also, introducing the polar components of the FT of the
vector mode functions, the following identity is recognised:
= e,(*,8o) (11.83)
and
t{kip)'b = (11.84)
aperture q
aperture p
(11.85)
252 Conformal arrays
valid for large separation sp. In the above, the roots xnp and xnp of the Bessel func-
tions are as defined in eqns. 11.221 and 11.223.
where it is recalled that N and M are the number of array elements and aperture
modes, respectively, and sp = 0 identifies the reference element located at
z = (p = 0. The FT of eqn. 11.86 is (see eqn. 11.32):
where
Insert the z and <p components of eqns. 11.35 and 11.36 into eqn. 11.87. Introduce
the asymptotic expression for the Hankel function for large arguments. Thus for
large p the pattern for the array radiated field is:
dw (11.89)
and
EJp,<p, z ) ~
(11.90)
Expressing f in terms of w by using eqn. 11.38 the integrals in eqns. 11.89 and
11.90 can be evaluated asymptotically (for large z) through a standard application
of the saddle-point method 4 . The following expressions are thereby obtained for
the 0 and $ components:
Conformal arrays 253
e'lkr 1 -
(1L92)
F)
The numerical evaluation of eqns. 11.91 and 11.92 is straightforward. The number
of terms required is approximately equal to and greater than the argument of the
Hankel function
\n\>~kRsin6 (11.93)
The series are therefore slowly convergent and of impractical use for large cylinder
radii. A procedure for transforming eqns. 11.91 and 11.92 in computationally
simpler expressions is presented below. The key idea is the transformation into a
rapidly convergent series of which only one or a few terms are retained. The
procedure is different but equivalent to the Watson transformation used in scatter-
ing problems23 (for an introductory discussion see Reference 6).
To introduce the technique consider a continuous function C(v) of the argument
p. The value taken by C(y) for v equal to an integer n must be equal to C(n). By
using the sifting property of the Dirac delta functions, the following identity is
established:
x £ b{v~ri)dv (11.94)
+
f 8(v-n) = £ ei2ww (11.95)
f C(n)exp\-jnL-Z-
The identity (eqn. 11.96) will be used for the evaluation of the pattern com-
ponents (eqns. 11.91 and 11.92). Different mathematical expressions will be given
for apertures in the lit and shadow regions with respect to the directions of
observation. The patterns of the whole array will be calculated here as the super-
position of the patterns of the single apertures which can be either in the lit or in
the shadow region. Use the superscript *p' for the pattern of the generic pth aper-
ture. By using the identity eqn. 11.96 in eqns. 11.91 and 11.92 (here applied to a
single element) the following expressions are obtained, after allowing for the
appropriate phase terms due to the element location:
irR n
1
v \ . v cos 6
P ,k^8j+jejB,,ifcco,0|| dv (H.97)
I \
and
£ . \l (v , J ^
2"
(11.98)
which is the starting point for the derivation of relatively simple expressions valid
for large kR sin 6.
..O—J o O 1 O 1 O 1 O O
-3-rr 4>_) -2tT -TT 0O 4> rr 0, 2TT 3TT
2Tr 2TT
I*. 2TT ^r •*«
(a) Lit region: The minimum azimuthal separation between the aperture
angular location and the direction of observation is denoted by ± |^ 0 |, with the
positive sign if the direction of observation is reached from the aperture moving
counterclockwise along the shortest arc. In the lit region \\po\ <TT/2. An approxi-
mation for the far-field expression is
x sin $o> k cos 8) + sin i|/0 c o s 0€ zp(k sin 0 sin ^O, k cos 8)] (11.100)
These expressions are recognised to be the same as the ones that would hold for an
aperture — having the same magnetic-current distribution — on a ground plane
tangent to the cylinder at the element centre.
e'!kr 1 ( f / n
E$(r, 0, 0) ~jk — exp (jkzp cos 0) exp — jkR sin 0 I \jj+ — ~
r~*oo r 2 II \
+
QpQc sin 0, k cos 0) sin 0 4- ezp(k sin 0, k cos 0) cos 6]g j ^ 1
2sin 2 0j
x
[ e 0p(~ ^ sin 0, k cos 0) sin 0 — ezp(— k sin 0, k cos 0) cos 0]
and
is the curvature radius for an helical path on the cylinder whose tangent forms an
angle S with the cylinder generator. Introduce the path lengths
01.104)
and according to eqn. 11.57, the distance parameters
1/3
7* = s± (11.105)
2R2t(9)
It is evident that the arguments of the Fock functions in eqns. 11.101 and 11.102
are equal to eqn. 11.105, i.e.
:
--\(\kR sin$)1/3 = T± (11.106)
Expressions 11.101 and 11.102, with the help of eqns. 11.104,11.105 and 11.106,
lead to an interesting and illuminating physical interpretation of the radiation
mechanism. The portion of the electromagnetic energy that is not radiated in the lit
region flows around the cylinder and is continuously radiated tangentially. Neglect-
ing the contributions of surface rays completely encircling the cylinder, to each
direction of observation are associated two diffraction points, whose distances from
the source along the geodetic rays are s+ and s~, respectively. The two surface rays
form an angle 6 with the cylinder generator and their radii of curvature are given by
eqn. 11.103, At the diffraction point the rays leave the surface tangentially and
propagate towards the observation point.
For large cylinders, unless \\}/~ \ ~ \p* ~TT, the contributions pertaining to i//0,
the smallest of the two angles, will be dominant. Thus, only one of the two terms in
eqns. 11.101 and 11.102, representing the two contributions, needs to be
considered.
11.4 Cylindrical arrays with a large number of elements: The periodic structure
approach
cylindrical arrays having the elements arranged on a regular grid. The actual finite
array is considered embedded in an infinite periodic structure. Neglecting the array
edge effects makes all the element RGPs identical, except for a rotation and a geo-
metrical phase term.
The method of analysis is based on the systematic use of the rotational and trans-
lational symmetry of the double periodic structure. Every arbitrary free excitation
of a finite portion of the infinite array is considered as a weighted superposition of
a set of elementary excitations of the infinite array matching the symmetry of the
structure, in the sense of generating fields equal for each array elementary cell,
except for a uniform phase progression. These elementary excitations are eigen-
vectors of the symmetry operators representing the congruences of the structure.16
Since the generalised scattering matrices Sfe defined in Section 11.2.4 commute
with the symmetry operators, the eigenexcitations are also eigenvectors of Sk. The
analysis of the structures for these eigenexcitations is much simpler than for a
generic excitation. The general case, on the other hand, can be analysed by applying
superposition.4 These concepts lead to a relatively simple method of analysis, and
also to an effective technique for element pattern design.
The basic ideas of the method are described in the next Section. Details of
analytical or numerical nature are presented in Sections 11.4.3 and 11.4.4.
0 o
o olo o 0
., - [I (H.10T)
s2 = (11.108)
A point of the cylinder surface can be identified by the position vectors, given by
eqn. 11.28. Again the element whose centre is located at £ = z = 0 is called
'reference element'. The position of a generic element is identified by the vector
smn = mSl+m2 (11.109)
where the index n designates a particular ring and m the element location within it.
The matrices Ykl defined in Section 11.2.4 have in the present case an infinite
number of rows and columns. The translational andL-fold rotational symmetry of
the structure is reflected in the fact that the mutual admittances between the two
apertures identified by the pairs of indexes m, n and p, q depends only upon the
differences of the indexes (m—p mod L)
Yik[(m-p)s1+(n-q)s2] = Y*p.n>q (11.110)
Of course, the same property holds for the elements of the generalised scattering
coefficients:
sk[(m-p)sl+(n-q)s2]=s^p,niq (11.111)
On the basis of the special structure of eqns. 11.110 and 11.111, it is easy to show
(Appendix 11.8.5) that the eigenvectors of the matrices whose elements are given by
eqns. 11.110 and 11.111 are the infinite dimensional vectors (suitably normalised)
ex
e(u0, wo) = e(uo) = J^£ P [-/Ho("»i + ns2)] (11.112)
«o = ^ (11.114)
K
260 Conformal arrays
which indicates that the phase progression between two adjacent elements in a ring
is 2mjN. The 'azimuthal number' / can take only integers values ranging from 0 to
N — 1, or from — N to N — 1, for elements in a rectangular or in an equilateral tri-
angular grid, respectively. The longitudinal number vv0 takes all real values from
— rr/h to n/h, and from — 7r/2/z to ir/2h for the rectangular and the equilaterla tri-
angular grid, respectively. The justification of these statements lies in the concept
of a reciprocal cell in the wavenumber plane. In the planar array context a dis-
cussion is found in reference 3.
Because of the very definition of eigenvectors of an operator, the following
equations hold:
Yhie(u0) = Yki(uo)e(uo) (11.115)
and
Ske(u0) = rk(u0)e(u0) (11.116)
with Y (u0) and rk(u0) eigenvalues associated with the eigenvector e(u0). From
ki
eqns. 11.21, 11.115 and 11.116 it follows, via some straightforward algebraic
manipulations, that the eigenvalues rk(u0) must satisfy the set of M linear
equations in M unknowns (M being the number of aperture modes approximating
the element magnetic currents)
M-l
I = -Ys0(u0)+Y08s0,
s0
fe = O
(s = 0 . . -Af—1) (H.117)
which allows one to determine the Tk(u0) if the set of numbers Ykt(u0) is known.
The expressions for the Ykl(u0) are given in the next Section.
In Appendix 11.8.5, it is shown that the system of vectors (eqn. 11.112) form
an orthonormal complete set for u0 belonging to the appropriate limited set of
points in the uOi w0 plane. This means that any arbitrary excitation a of a finite
portion of the infinite array can be represented as a linear superposition of the
vectors e(w0).
To make the analytical development less abstract, from now on the discussion
will focus on the triangular arrangement of elements shown in Figs. 11.14 and
11.16. The minor modifications of the various equations for other element grids
such as the rectangular are easily worked out and are not reported here.
Any free excitation of the finite array embedded in the infinite periodic struc-
ture can be expressed as a weighted superposition of the vectors (eqn. 11.112),
because of their completeness property (Appendix 11.8.5). The following expan-
sion holds:
a
= 1 } nJ2hcii>Wo)e[R>Wo)dWo (11.118)
The coefficients c(i9 w 0 ), because of the orthonormality of the basis vectors (eqn.
11.112) (see Appendix 11.8.5), are obtained by projecting a on eache(/, w0):
Con formal arrays 261
1/2 k
Y001/2VVrrt t
Y = art80k+ £*
j\ (11.122)
L-l +<*>
Z L hl(s — msi — ns2) exp [— juo(msi + ns2)] d%dz (11.124)
(1U25)
]
. IU »\ ' [«'"• " S e n
I 6 \u~l\dudw, (11.128)
2TTR
A'(,-mSl-«s2) = ^- {_
2n
1
(11.129)
which, inserted in eqn. 11.124, and interchanging the orders of the integration and,
recalling eqn. 11.115, gives, through simple manipulations,
1 4- OO + °° •*" ° °
xexp[/(«— «o)( ms i + j
(11.130)
Con forma/ arrays 263
Notice that now both the summations are extended to infinite limits. Using eqn.
11.44, one obtains
f •-;]• ' - [ ° - T ]
It will prove convenient in view of further developments to associate to each
bidimensional vector u0 (eqn. 11.113) a regular lattice of points in the u, w plane
identified by the set of position vectors, forming a 'reciprocal lattice'
= u0 4- ptx + qt2 (11.134)
with p and q arbitrary integers and t\ and t2 defined in eqn. 11.132. For elements
in the equilateral triangular arrangement of Fig. 11.1, the coordinates of the points
(eqn. 11.134) are given explicitly by:
Uopq =
R*~fP* W
OPQ = (2^-p)~ (11.135)
(11.137)
which is the result sought.
For the numerical evaluation of eqn. 11.37 a truncation radius in the u, w plane
264 Conformal arrays
is chosen. The summation will be extended to the points (eqn. 11.34) internal to
the circular region so defined. The truncation radius has to be chosen large enough
to ensure that, outside the circle, the modulus of the transform of each of the M
selected mode functions | efe(w) | is negligible with respect to its peak (say, — 20 dB).
11.4.4 Array RGP for an arbitrary excitation. Single excited element RGP
The evaluation of the RGP via the eigenexcitation method does not require the
inversion of large matrices. Also the RGP of the single excited element is the same
for every array element, neglecting edge effects. The computational procedure is
straightforward. The derivation of the expressions of the d and (j> components of the
RGPs is fairly complicated, however, and will be reported only in its general lines.
Deriving an expression for the RGP requires performing the following steps:
(a) Establish an expression for the tangential electric field on the cylinder pertain-
ing to each eigenexcitation.
(b) Determine the associated field for large p.
(c) Use superposition according to eqn. 11.121.
Step (a): The aperture distribution on the generic element 'm, «' for the eigen-
excitation e(u0) is clearly the following:
I [6oh +
Z1TL fess0
Step (b): From eqns. 11.89, 11.90 and the periodicity of the structure, it follows
that the far field for eigenexcitation e(u0) is given by
vl/2
/ h
g(i,wo;r,6,0) =
x
[^0 (wopa \ 0) + 4>S<f){uQPq; ^)] exp | — /i ($ — — j | (11.140)
M-X
XI [5Os+rs(«o)]6*(«op«) (11.141)
s=0
[ 2n( n\ |
/ / Q-mn sxp \jiL — I m lH— I + ; « M cos ^ (11.143)
m n [ \ ) J
The RGP of the single excited element in the reference position is obtained by
having in eqn. 11.43 all amn = 0 except fortfoo= 1.
In connection with the evaluation of eqn. 11.143 notice that the second sum-
mation is extended only to the wavenumbers in the z direction corresponding to
propagating harmonics.
11.4.5 Aperture matching and element pattern design
We want to show that the element RGP can be modified via an appropriate design
of the element matching network. A technique will be introduced, and its effective-
ness exemplified by several illustrative cases. Although more complicated schemes
can be developed, only the simplest case of matching networks identical for all
elements will be here discussed.
Consider the plane tangent to the array surface through the centre of the
reference element and, on that plane, the direction of the cylinder axis and the
circumferential direction, perpendicular to the axial one. Suppose that one seeks
to increase the element RGP in a direction forming the cosines 7 and a with the
axial and circumferential directions, respectively. To accomplish this purpose a
266 Conforma/ arrays
matching network, the same for each element, will be selected to match the array
when excited by the eigenexcitation identified by
Go = Gii, w0)
Bo = B(i,w0)
The element RGP will be matched also in the directions — y, — a, symmetric with
respect to broadside. The approach and the ensuing procedure is easily justified by
observing that the realised gain pattern of an element in an array on a cylinder
having a large radius is obviously 'similar' to that of the same element in a planar
array with the same element grid. For the latter case, the element pattern is maxi-
mised in a direction a, y if the array is made transparent, i.e. matched, for a plane
wave incident from such a direction, or, equivalently, if it is matched for the array
beam steered in the direction a, y. The extension of the procedure to a cylindrical
array will therefore consist of matching the array to an eigenexcitation have the
same relative phase progression among the elements. All our considerations will
apply to a narrow band, ideally monochromatic case.
To introduce the technique consider the terminal admittance for the funda-
mental excited mode pertaining to the eigenexcitation identified by eqn. 11.113.
This 'active' admittance is, of course, the result of the mutual couplings among
the array elements for the particular eigenexcitation considered. Its expression is,
clearly,
= Yo l l o ( " ° \ (11.145)
111 (w J
Suppose now one introduces in each element a matching network, consisting of
shunt susceptance — Bo located at half transmission-line wavelength from the
aperture — a waveguide iris — and a perfect transformer having a turns ratio equal
to n2:n1 = (Go^o) 172 (Fig. 11.15). Clearly this network matches the array for the
eigenexcitation e(u0). Assuming that higher-mode interaction with the matching
network is negligible, a reasonable assumption, the set of aperture voltages for the
various modes (eqn. 11.120) are multiplied by the factor y/G0/Y0.
Conformal arrays 267
Following this procedure the gain can be enhanced in prescribed directions off-
broadside, at the expense, of course, of broadside gain. The total effect is a flatten-
ing of the element pattern with respect to the case of array matched for equiphase
excitation (7 = w0 = 0).
Fig. 11.16 Aperture geometry and element grid for the numerical examples
Lattice 1 Lattice 2
h = 0.3\0 h = 0.28867 \ 0
d=1.039\0 d=1\0
a = 0.22\ 0 a = 0.22\ 0
By using a smaller size lattice (lattice 2 of Fig. 11.14) the patterns do not have
resonance notches but only a slight dip before 90°. For an array of approximately
50 wavelengths diameter, Figs. 11.19 and 11.20 show the patterns and also
illustrate the effect of 'off-broadside' match. For circumferential polarisation the
effect on the element RGP of a match for i/R = 2n sin 80°/X and w0 = 0 is an
increase of the gain off-broadside. At 80° the gain reduction with respect to broad-
side is approximately 6dB, instead of the 12dB for the more conventional equi-
phase match. For the RGP of an axially polarised element the match off-broadside
produces similar effects (Fig. 11.20).
100 J
30 60 90 120 150 180
4> t degrees
Fig. 11.17 Element realised gain pattern: Circumferential plane; circumferential polaris-
ation; equi-phase match (R = 99.23X)
In Fig. 11.21 the realised gain patterns in the circumferential plane for three dif-
ferent match conditions are shown for a cylinder of 1IX radius; the pattern for
frequency variations of ± 5% with respect to centre frequency are shown in Figs.
11.22 and 11.23. Broadside and 80° off-broadside match conditions have again
been considered. A notch begins to appear at high frequency. The modification of
the realised gain pattern with frequency depends substantially upon match con-
ditions, and thus is not due simply to geometrical reasons. In calculating the curves
of Figs. 11.22 and 11.23, the matching susceptance shunting the aperture has been
assumed constant with frequency (a reasonable approximation in a ± 5% band).
The element RGP in a conformal array is only an indirect indication of the scan
coverage capability (unlike the planar case). The influence of the array geometry,
Con formal arrays 269
lattice and match condition is illustrated in Fig. 11.24 to 11.28. Two arrays have
been considered, having 37 and 313 elements, respectively. The cylinder radius is
equal to approximately 11 wavelengths and the element lattice is the one denoted
30 60 90 120 150
^ , degrees
Fig. 11.18 Element realised gain pattern: Circumferential plane; axial polarisation; equiphase
match {R = 99.23X) (Lattice 1).
by '2' in Fig. 11.16. The projected apertures of the arrays, in the direction of the
normal to the plane tangent to the array at its centre, are circular. The projected
illuminations are Gaussian with a lOdB taper. The azimuthal sectors occupied by
the arrays on the cylinder are equal to 25° and 56° for the small and the large
array, respectively. The array pattern is calculated by using expression 11.143. The
free array excitation is chosen always in such a way to have the radiative contribu-
tions of all elements adding in phase at the peak of the beam. Since the phase
patterns of the element realised gain are essentially flat up to 80° from broadside,
the phases of the excitations turn out to be practically coincident with those
obtained from simple geometrical optics considerations. In Fig. 11.24 the realised
gain pattern of the array with 313 elements is shown, for broadside scan. The maxi-
mum gain is approximately 30 dB for equiphase match. Two different matching
conditions have been considered: equiphase and i/R = 2n sin 80°/X (with w0 = 0).
The two patterns are very similar, as expected, the differences lying in different
peak gains with the far-out sidelobes slightly higher for the case of match off-
broadside, because of the less efficient filtering effect of the element pattern.
270 Con forma I arrays
Figs. 11.25 and 11.26 show the envelopes of the maximum array gain vs. scan
angle for linear polarisation in the plane of scan and different match condition for
the array with 37 elements. Figs. 11.27 and 11.28 show similar curves for the large
array on the same cylinder. For match off-broadside the gain drop-off at 80° is
approximately 6 dB for the circumferential plane and 7 dB for the axial plane. The
absolute gain at 80° is 2 dB greater than that which would be obtained with the
same elements broadside matched.
Fig. 11.19 Element realised gain pattern: Circumferential plane; circumferential polarisation
(R = 49.33X) {Lattice 2)
match at HR = 2*7X sin 80°
equiphase match
The methodology developed in Section 11.3.4 for the calculation of the generalised
mutual admittances can be formally extended to arbitrary convex surfaces. The
following GTD formulation is based on the work of Lee et al.15
A point of the surface is again identified by a position vector s. The large
parameter in the asymptotic expression of the Green's function is:
m(s) = tikRt(s)}1/3 (11.146)
Con formaI arrays 271
in the cylindrical case this parameter being a constant along a helical path. In eqn.
11.146, Rt(s) is the radius of curvature at the point s in the direction of the surface
ray; that is in the direction of the tangent t. The radius of curvature in the direction
CD
10
•I"
5 30
O
30 60 90
TT/ 2 - © , deg
Fig. 11.20 Element realised gain pattern: Axial plane; axial polarisation (R = 49.33X)
(Lattice 2).
match for w = 2n/X sin 80°
equiphase match
c 10-
g. 20
c
'§• 30 match at i/R = —sin 80°
•o
w 40
. equiphase match
* 50
0 30 60 90 120 150 180
$ t degrees
Fig. 11.21 Element realised gain pattern: Circumferential plane; circumferential polarisation
(R= 11.6U)
ds (11.147)
272 Conformal arrays
the integral being extended to the ray geodetic path. Another parameter of interest,
entering in the expression of the surface magnetic field, is
1/2
ks
(11.148)
2m(s')m(s")T
m \
10
E
to
Q. 20 \
c
d
01
30
~o
to
V
in
0 40 \\
\
f ' '
•
(3 30 60 90 120 150
<f> t degrees
depending upon the curvature at the source and observation point. The parameter
T\ is unity for the special case of constant ray curvature. Finally, the mean radii
of curvatures will be introduced:
RtV,*") = [Rt(s')Rt(s")]1/2 (11.149)
Rb(s',s") = [Rb(s')Rb(s")]l/2 (11.150)
f
Denote by s(s',s") the shortest distance between the points s and s" along a geo-
detic path. Redefine for the general convex surface the functions Ht(s) and Hb(s)
as follows:
Hb[s(s',s")] = G(s) ( l - f A
(11.151)
Conformal arrays 273
Ht[s{s',s")} =
(11.152)
90 120
t degrees
with G(s) given by eqn. 11.58. When the surface is cylindrical eqns. 11.151 and
11.152 reduce to the appropriate expressions given in Section 11.3.4.
Consider a small pencil of surface rays originating ats' and propagating towards
s". The divergence factor/^ of the pencil is defined by (Fig. 11.29):
sdxjj'
D(s,s ) = (11.153)
where pc is the caustic distance of the wavefront at s" and is always positive.
Denote by i (s',s") the unit vector tangent to the ray path at s' from a generic
point of aperture '/?' to a generic points" of aperture 'q\ In a parallel way i (s",s')
is defined. Expression 11.67 for the mutual admittance between two apertures
for the ith and kth modes is generalised for an arbitrary convex surface as follows:
d£dV'i(s'\sf)Hb[s(st,sff)]D(s\sff)
xD(s',s")[b(s',s"ye\s')] (11.154)
274 Con formal arrays
where p and q identify the two apertures considered and %', rf and if, 17" are two
planar rectangular systems on the planes tangent to the surface at the centres of the
aperture *p' and 'q\ respectively.
£D
T3
- 50.00
-60.00
0.00 20.00 40.00 60.00 80.00 100.00 120.00
<p t degrees
(11.157)
s(s\s")
with — 7r < r2(s', s") < 7T. Similarly,
0 10 20 30 40 50 60 70 80 90
4> t degrees
Fig. 11.25 Array scan coverage. Circumferential plane; circumferential polarisation match for
HR = k sin 80°. Number of array elements = 37; R = 11.61 \ (Lattice 2)
/ 2n
Match for — = — sin 80
R \
- Equiphase match
If, at each point of the geodetic path, i is expressed via its orthogonal components
along two orthogonal axes on the tangent plane in the circumferential direction 0
and in the direction — f (towards the cone tip)
(11.160)
then
b = (f r ,*0) (11.161)
Also, see eqns. 11.149 and 11.150,
\fr\r~i tan 60
R+ = (11.162)
I sin £2(s', s") sin £l(s", s')\
276 Conformal arrays
tan 60
Rh = (11.163)
cos Q(s , s") cos O(s", sr)\
f
Also
T(S,S") = [hkrl\smQ,{s,sft)\smd0]l/3\A\cos2/3e0 (11.164)
0 10 20 30 AO 50 60 70 80 90
— - 9 , degrees from broadside
Fig. 11.26 Array scan coverage. Axial plane; axial plane match for w = k sin 80°. Number of
array elements = 37; R = 11.61X (Lattice 2)
Equiphase match
Axial plane match
(11.168)
In applying eqn. 11.154, only the shortest geodetic ray needs to be considered.
However, if | A| ~7r, two surface ray contributions should be taken into account.
Higher-order geodetic ray contributions are negligible for structures large in terms
of wavelengths, i.e. those for which the formulation applies.
Conformal arrays 277
0
2
4
6
8
10
12
14
16
10 20 30 40 50 60 70 80 90
f t degrees
Fig. 11.27 Array scan coverage. Circumferential plane; circumferential polarisation match for
HR = k sin 80°. Number of array elements = 313;/? = 11.61X {Lattice 2)
Circumferential polarisation match
Equiphase match
where T is
Fig. 11.28 Array scan coverage. Axial plane; axial polarisation match for w — ks\nS(f.
Number of array elements = 3 1 3 ; / ? = 11.61X (Lattice 2)
Axial polarisation match
— Equiphase match
Fig. 11.29 Divergence factor for a pencil of rays originating at s and propagating towards s"
(a) Lit region: As for the cylindrical case the smallest azimuthal separation
between the aperture angular location and the direction of observation is ± |i^ol>
with the positive sign if the direction of observation is reached from the aperture
moving counterclockwise along the shortest possible arc. A rectangular system of
coordinates has the origin at the cone tip, with the z axis coincident with the cone
axis. In this reference the vector sp identifying the location of thepth aperture has
the expression
sp = (11.175)
Because of the structure rotational symmetry only the difference between the azi-
muthal angular coordinate of the direction of observation and the azimuthal
angular coordinate of the aperture appears in the radiation pattern. Thus one can
assume that in eqn. 11.175 yp — 0, that is, consequently, <j> = ^ 0 . Then the propa-
gation vector k in the direction of observation has the following components:
kx = k sin 6 cos i//0 (11.176)
ky = k sin 0 sin i//0 (11.177)
280 Conform a I arrays
kz = k cos 6 (11.178)
Let J, 7? be two orthogonal coordinates on the plane tangent to the cone at the
aperture centre, (Fig. 11.31). The unit vectors | , T) in the direction of the | and 17
axes point in the circumferential direction and toward the tip of the cone, respect-
ively. The unit vectors x, y, z are related to | , 7? and to the unit vector h in the
direction of the normal to the aperture by the transformation
i = y (11.179)
7? = — x sin 60 4- z cos 60 (11.180)
h = x cos 60 -\- z sin 0O (11.181)
Extending the result established for the cylinder in the lit region, the radiation
pattern is approximated by that of an aperture in a plane tangent to the cone at
the element location, and is given by
e~jkr
p
r->oo' r
where tp(u, w) is the FT of the aperture transverse electric field distribution with
respect to £, 17 (see eqn. 11.88). Explicitly the arguments of e p in eqn. 11.182 are:
fc'l = k sin 0O sin \jj0 (11.183)
Con formal arrays 281
direction of
observation
geodetic paths
to diffraction
points
6)-[«i>(*-i*-T?)x/2] (11.187)
or
(11.188)
e~Jkr
Eg(r, 6,0) ~ - / * e1k-*P(- fj cos \p0 + | sin 60 sin i//0) *P(k• i * '*?)
(11.189)
or finally
e-jkr
(b) Shadow and penumbra region: The electromagnetic field radiated from an
aperture propagates along geodetic rays on the surface with a decay due to the loss
of energy continuously radiated in the direction of the ray tangent (Section
11.3.5). Thus to each direction of observation certain geodetic paths are associated,
having at certain 'diffraction points' their tangents pointing towards the direction
of observation. As for the cylindrical case two geodetic rays only will be assumed to
contribute significantly to the pattern. They are associated with the azimuthal
separation \jj+ and $' = — |^~| between the aperture and the direction of obser-
vation, the positive angle corresponding to a motion in counterclockwise sense. The
angular separations between the aperture centre and the two diffraction points are:
A+ = ¥~~: (11.192)
and
A" = i//~-f| (11.193)
The angle y formed by the direction of observation and the generator of the cone,
the same at both the diffraction points, is given by (Fig. 11.31)
cos 7 = cos 6 cos 0O (11.194)
The directions of the geodetic rays from the aperture centre to the diffraction
points are identified by the angles £2+ and O" between the geodetic rays and the
direction of the cone generator. Their expressions are
£1+ = 7 - A + s i n 0O (11.195)
£T = - 7 - A " sin 0O (11.196)
Conforms/ arrays 283
The distances of the two diffraction points from the tip of the cone are found to be
It is convenient to consider the contributions of the two rays from each diffraction
point separately, and using eventually superposition. For each contribution the
components in the directions of the normal and binormal to the geodetic rays,
given by eqns. 11.200 and 11.203, will be considered. It will be apparent that this
reference is the most 'natural' to express the field. The 0 and0 components of the
284 Conformal arrays
E^(rJP)^}'xp{
11.7 Acknowledgment
The GTD formulation for the surface magnetic field, discussed in Sections 11.3.3
and 11.3.5 has been contributed by S. W. Lee. Sections 11.3.4 and 11.6.1 also are
based in part on his material.
11.8 Appendices
„-.
Lw
f>U (11 .214)
where
k = coy/ql (11.216)
is the wavenumber in an unbounded medium having the same permittivity e and
permeability JLI of the material filling the waveguide. The modal admittances are
then given by the following expression for the TE and TM modes, respectively:
TE modes:
rTE = M_ /I (11.217)
k
TM modes:
For a rectangular waveguide with sides having the lengths 2a and 2b, the cutoff
frequencies are given by the following expressions, valid for both TE m n and TM mn
modes:
(11.219)
Consider a circular waveguide having a cross-section with the radius 'a\ In the
following, as customary, the first of the two indexes 'identifying a particular mode,
refers to the azimuthal periodicity. The cutoff frequencies are given for TE modes
by:
(/ c )™ = -$*= (11.220)
2iray/€fJ.
where x'np is the pth root (ordered with increasing magnitude) of the equation:
dJn(x) = 0 (11.221)
dx
Jn being the Bessel function of the first kind and order n. For TM modes, the cut-
off frequency is given by:
(f\TM = (11.222)
11.8.2 Fourier transforms of the vector mode functions for rectangular and circular
waveguide radiators
Consider the 'reference' element whose centre is located at £ —z = 0 (Fig. 11.6).
For a rectangular aperture with % and z dimensions equal to 2a and 2b, respectively,
the vector mode functions can be written as follows. For TE nm modes,
,2-1-1/2, x l/2
(n.224)
where rn is the Newman factor equal to unity for n — 0 and equal to 2 otherwise,
and
[ t \ [muz mir\
( ) (11-225)
(11.226)
From eqns. 11.224 and 11.227 it is apparent that in order to establish the expression
for the FT of the vector mode functions, the FT of eqns. 11.225 and 11.226 have
to be established. This is quite elementary, the results for the transform of eqn.
11.225 being:
1
nm, ^ \ ,
FZ (u, w) = — exp
P — }(m + n 4- 1) —
2* [ >2
Inn
mix
(11.228)
Conformal arrays 287
The expression for the FT F" m (w, w) of eqn. 11.226 is obtained from eqn. 11.228
by using the following replacements:
u <- w
Then from eqns. 11.224-11.228 the FT of the % and z components of the various
vector mode functions are immediately obtained.
Consider now a circular waveguide aperture element with the radius of the cross-
section denoted by a. Introduce a planar polar system of coordinates p, i// on the
aperture of the waveguide element, associated with the rectangular system | , z via
the standard transformation
sin
(11.230)
and for the TMnp modes
^npp\ fcOSMi//)
01-232)
TI pxnpJn_x{xnp) \ a
In the plane of the wavenumbers M, W introduce a planar polar system t, /i with
] (H.235)
(n.236)
e?{t,ii) = 0 (11.237)
the latter equality holding for all pairs of indexes n and p.
The | and z components of the FTs are obtained from eqns. 11.234-11.237 via
the coordinate rotation
M (11.238)
e2np(w, w) = et(t9 fi) sin M + eM(f, ju) cos /x (11.239)
In the text the alternative notation e^ has been also used for e^.
W
2(t) = — (11.241)
where Ai(t) and Bi(f) are the Airy functions whose definition and discussion is in
Abramovitz and Stegun1. The general definition of Fock functions is not given
here, but only those functions used in the text are here listed:
(11.243)
1 r e-Jxt
r \ d (11.245)
7T ' r W ( 0
and the expression for # ( / ) is obtained by taking the derivatives of eqn. 11.251.
Numerical values of an and an are found in Abramovitz and Stegun,1 page 478.
For small value of the argument the functions (eqns. 11.242, 11.243) can be
evaluated by resorting to the following small argument expansion, reported by
Lee:13
1/2
+ n i 9STI
I 12 64
It has been indicated that the smoothest crossover between the residue series and
the small-argument expansions occurs for x ~ 0.6 (Chang et alP). The functions
/(*)> z(x) a n d g^'\x) are tabulated and graphs for g(x) and f(x) are in Bowman
et al.6, among others. Interpolation from the tables or from the graphs in the region
I x | < 1 (where the functions are smooth) is a procedure simpler than attempting a
Taylor expansion leading to rather involved expressions.
Tables of the Airy functions for moderate values of the argument are in
Abramovitz and Stegun.1
11.8.4 Evaluation of the radiation pattern of an aperture on a large conducting
cylinder
^o = 0-f? (11.254)
K
whose absolute value is less than n/2 in the lit region.
Consider first a positive \p0. From eqns. 11.97 and 11.98 the components of the
pattern of thepth aperture, located at %p, zp, are:
e-jkr l + oo r
££(r, 0, <t>) ~ exp (jkzp cos 0) exp \-jp UJ0 - ^ I
. P COS0 .. , ^., ,
-fccos0 + _ . 2 e z p \-,kcosS\\dv
(11.255)
Eg(r,d,(P)~-jy- J
r_>oo r nRsind -°° H^2)(kR sin 0)
where only the term n = 0 has been retained in 11.97 and 11.98. To perform an
asymptotic evaluation of eqns. 11.255 and 11.256 for large kR sin 6, the Hankel
functions and their derivatives will be replaced by their Deb ye asymptotic
expressions. The approximation is not valid in the immediate neighbourhood of
v~kR sin 0 (see for example, Sommerfeld19). It is argued, however, that the rapid
oscillatory behaviour of the phase of the integrand makes only a small region of the
v axis, where the phase is stationary, contribute significantly to the integral. This
region is far from v~kR sin 0. (The situation is, of course, totally different for an
observation point in the transition or shadow region.) On this basis the Hankel
functions in eqns. 11.255 and 11.256 and their derivatives will be replaced with
their approximations (Hasserjian and Ishimaru11)
HJ?\kR sin 0) ~ /-
' nkR sin 0 sin 0
x exp [-jkR sin 0(sin 0 - 0 cos 0)] (11.257)
Sin0)
" • ' — sin0
x exp [-jkR sin 0(sin 0 - 0 cos 0)] (11.258)
where
V
cos0 = yo n (11.259)
kR sin 0 '
Consider eqn. 11.256 first. By using eqn. 11.257 and changing the integration vari-
able from v to 0
e~ikr , ., ^ f /fffcR sin 0 sin 0
e
ex ULl+^1 ex -'L\-?~
exp 7y +o exp
+oo [ \ ° 2/J /v \ f ° \ ° 2;
C S
J-~ H^2)(kR sin 6) 6zp\Ry ° / ^ J-~ H?\kR sin Q)
^ . f c c o s f \l ) ^ (11.263)
R
I
where the relationship
e*v*Hil\z) = /# 2) (z) (11.264)
has been used. The procedure for the derivation of eqn. 11.261 is repeated for this
case and leads again to the expression 11.99.
For the <f> component eqn. 11.255 by using eqn. 11.258, one obtains
e~jkr 1 f f [
%r, 0, 0) — exp ( jkzp cos 0) exp fkR sin 0 sin j3
r * Jc | |
sin 0
cos |3 cos 0 1
sin 0 cos 0, — k cos 0) H — — ezp(k sin 0 cos ]3, /: cos 0)
sin0
sin 0 1
fcsin0sin/3d/3 (11.265)
By applying again saddle-point integration
ke~*kr
%r, 0, 0) ~ e-JOr/2) s i n 0 e x p ( / f a ^ c o s fl)
cos
x exp —fkR sin 0 sin I \p0 — l \\e<pp \ksind
" f— I ^Ao — — i, k cos 0
/ ^X
4- cos h//0 cot 0e zp I A: sin 0 cos (i//0 - ^ | , k cos 01 [ (11.266)
e ! ( r+ ~exp
, 0, 0) ~ exp [ jkzp cos 6} f _
r nR
\J-°° H^2)\kR sind)
v
x |e0p
-exP
+ — H™\kR sin 0) r0pl/T*COS7 +
fcRsin20
)--/-
vRsind [ J-°° ^ 2 ) ( / c ^ sin
(11.272)
In this case the asymptotically important contributions to the integrals in eqns.
11.271 and 11.272 come from the region of the v axis where v ~kR sin 6. In that
neighbourhood the Hankel function and its derivative can be approximated by a
combination of Airy functions and their derivatives as follows. Introduce the large
parameter
m(6) = (\kR sin 6)1/3 (11.273)
and make a change of variables
1
v-kR sin 6
= —Z7^ (11.274)
294 Conformal arrays
sin 0 ) ^ ^ £ (11.276)
VW (0)
where the functions w2(t) and w ^ O have been defined in Appendix 11.8.3. By
using eqns. 11.275, 11.276 and 11.263 (with \\p_\ replacing i//0) eqns. 11.271 and
11.272 become
exp
: [e / r/w(0) , ^T \ / tm(0) \
</v>D k sin 0 , k cos d\— \ l -\
[ \ /? \ kR sin9J
(11.277)
with F a suitable integration path, see Appendix 11.8.3. Similarly fromeqn. 11.272,
exp (
e
/^ c °; ^M cJ-m sin e (r ^ |
w 2 (r)
X
^ P l - 1 1 1 ^ - A : sin (9,^ cos ( 9 ) ^ ) (11.278)
Conformal arrays 295
Since e^p{u, w) and ezp(u, w) are slowly varying functions of their arguments and
the main contribution to the integrals 11.277 and 11.278 comes from a neighbour-
hood of t - 0, no great error is committed if the argument of the FT of the aper-
ture field transverse field distribution is replaced by its value at t — 0. Also to form-
ally simplify the expressions, the Fock functions are introduced (Appendix 11.8.3).
By using the definitions, eqns. 11.258-11.260, the expressions 11.271 and 11.272
take the forms of eqns. 11.101 and 11.102.
(11.279)
Putting
m-p =s (11.280)
n-q = t (11.281)
eqn. 11.279 becomes, because of eqn. 11.110.
In the same way it is shown that the set of vectors e(w0) satisfy eqn. 11.116.
If the two wavenumber vectors u0, ux identify two eigenexcitations, to show
that the system eqn. 11.113 is orthonormal the following equality should be
verified:
h L-i +<*>
7TT Z Z exp-/(no-Hi)*(msi +ns2) =
27TL
(11.284)
where /, k and w0, w t are the azimuthal and longitudinal numbers for the eigen-
excitations. By recalling the expressions of the two periodic vectors, Si,s2, for the
equilateral triangular arrangement given by eqns. 11.108 and 11.284 becomes
296 Conformal arrays
+o L 1
° f } "
^ exp\f(i~k)jn-j(wo-W2)nh\ £
Y L J m=o
in the interval ix/h < w 0 — Wi <7r//z. This equality, however, is proved by expand-
ing in a Fourier series of period 2TT//I the delta function in the right side of eqn.
11.286.
The formal statement of completeness for the system of vectors (eqn. 11.112) is
written as follows:
2TT ,
2
r exp [-jwo(n -p)h] dw0 = 8np8mQ (11.288)
J-irl2h
sin ( / i - p ) ^ L_!
J
I exp/i
and this shows that the eigenexcitations form a complete set and every array
excitation can be represented as a linear superposition of them.
11.9 References
3 BORGIOTTI, G. V.: 'A novel expression for the mutual admittance of planar radiating
elements', IEEE Trans., AP-16, 1968, (3). pp. 329-333.
4 BORGIOTTI, G. V., and BALZANO, Q.: 'Analysis and element pattern design of periodic
arrays of circular apertures on conducting cylinders', IEEE Trans., AP-20, 1972, pp.
547-555
5 BORGIOTTI, G. V.: 'Modal analysis of periodic planar arrays of apertures', Proc. IEEE,
56,1968,pp.1881-1892
6 BOWMAN, J. J., SENIOR, T. B., and USLENGHI, P. L. E. (Eds.): 'Electromagnetic and
acoustic scattering by simple shapes'. North-Holland, 1969
7 CHANG, Z. W., FELSEN, L. B., and HESSEL, A.: 'Surface ray methods for mutual coup-
ling in conformal arrays on cylinder and conical surfaces'. Polytechnic Institute of New
York, Limited Circulation Report, 1976
8 EDELBERG, S., and OLINER, A. A.: 'Mutual coupling effects in large antenna arrays',
IRE Trans., AP-8, 1960, pp. 286-297
9 FOCK, V. A.: 'Electromagnetic diffraction and propagation problems'. Pergamon Press,
1965
10 HARRINGTON, R. F.: 'Time-harmonic electromagnetic fields'. McGraw-Hill, 1961, Chap.
5, pp. 245-250
11 HASSERJIAN, G., and ISHIMARU, A.: 'Currents induced on the surface of a conducting
cylinder by a slot', /. Res. NBS, 66D, May-June 1962, pp. 335-365.
12 KUMMER, W. H.: 'Preface, special issue on conformal arrays', IEEE Trans., AP-22, 1974,
p.l
13 LEE, S. W.: 'GTD calculation of mutual admittance and element pattern of slot on con-
formal arrays', Dept. of Electrical Engineers, University of Illinois, Urbana, Illinois,
1979.
14 LEE, S. W., and SAFAVI-NAINI, S.: 'Approximate asymptotic solution of surface field
due to a magnetic dipole on a cylinder', IEEE Trans., AP-26, 1978, pp. 593-598
15 LEE, S. W., and MITTRA, R.: 'GTD solution of slot admittance on a cone or cylinder',
Proc. IEE, 126,1979, pp. 487-491
16 MONTGOMERY, C G., DICKE, R. H., and PURCELL, E. M.: 'Principles of microwave
circuits', Dover Publications, New York, 1965
17 PRIMDORE BROWN, D. C, and STEWART, G.: 'Radiation from slot antennas on a
cone', IEEE Trans., AP-20, 1972, pp. 36-39
18 PRIMDORE BROWN, D. C: 'Diffraction coefficients for a slot-excited conical antenna',
IEEE Trans., AP-20, 1972, pp. 40-49
19 SOMMERFELD, A.: 'Partial differential equations in physics', Lectures on Theoretical
Physics, Vol. 6, Chap. IV
20 'Special issue on phased arrays', Proc. IEEE, Nov. 1968, (11)
21 STEISKAL, H.: 'Analysis of circular waveguide arrays on cylinders', IEEE Trans., AP-25,
1977,(5), pp. 610-616
22 TILLMAN, J. D.: The theory and design of circular antenna arrays', University of
Tennessee Experiment Station, 1966
23 WATSON, G. N.: 'The diffraction of electric waves by the earth', Proc. R. Soc, A-95, pp.
83-99
Chapter 12
Circular arrays
D. E. N. Davies
12.1 Introduction
The field of circular (or cylindrical) arrays has received far less attention than that
devoted to linear and planar arrays. At first sight this seems surprising since one of
their principal advantages is the ability to deflect beams electronically through 360°
with little change of either beamwidth or sidelobe level. Circular arrays have been
applied, to a limited extent, in both communication systems and navigational aids
since the late 1930s.1 There has, however, been surprisingly little change in their
scale of application since that date, though there are now signs of a significant
increase of interest in such arrays for applications to DF (Direction Finding) and
ESM (Electronic Support Measures) systems.
The analysis and general understanding of the properties of circular arrays has
progressed steadily for many years and the most significant developments were
probably made by several groups of workers in the 1960s when the concept of
phase mode excitation was developed. The lack of applications in the past is prob-
ably due to the fact that the basic problems of exciting circular arrays with the
correct values of amplitude and phase are, in general, more complex than for linear
arrays. Furthermore, electronic scanning of directional patterns for circular arrays
may be difficult to implement and can require both the amplitude and phase of
each element of the array to be changed.
The basic symmetry of circular arrays offers a number of advantages, which in
addition to the 360° scan angle already mentioned can include an ability to
compensate for the effects of mutual coupling by breaking down the array exci-
tation into a series of symmetrical spatial components. This can also give rise to
directional patterns which remain constant in shape over broad bandwidths. These
advantages, together with new and more convenient methods of array phasing,
are beginning to increase interest in such arrays for a number of potential appli-
cations.
Early work on circular arrays related to their application for direction finding1
and their ability to produce omnidirectional coverage2 from a cylindrical con-
figuration of elements wrapped around a central mast or support structure. This
Circular arrays 299
latter configuration was applied in broadcast antennas to avoid shadowing from the
basic mast structure which supported the antenna. In this case controlled excitation
of the vertical aperture of the cylindrical aperture is used to achieve the desired
pattern in the vertical plane. This is generally achieved by stacking multiple rings
of circular arrays.
Other circular array applications have included Wullenweber arrays for direction
finding,3 wide bandwidth HF communication arrays,4 wrap-around antennas for
shipborne communications,5 navigational aids,6 spacecraft antennas,7 null steering
antennas for mobile communication applications8 and wide-bandwidth microwave
direction-finders.9
In the field of radar systems there have been several experimental circular arrays
aimed at achieving 360° electronic scanning10""14. These have been aimed at
evaluating different scanning concepts applied to fairly small apertures containing
perhaps 16 or 32 elements. The problems of high-power feeding and phasing for
circular arrays still represent a significant restriction on this type of application,
particularly for wideband operation. For the radar field there is, nevertheless, the
possibility of designing circular arrays for receiving only, where the problems of
phasing can more easily be achieved by array signal processing techniques.
There has also been a limited application of circular array configurations to
sonar systems15 although the published material available in this field is far more
limited.
Probably the most obvious way of exciting a circular antenna array is to arrange the
signals from all the elements to add together in phase for a single (main beam)
direction. Assuming this direction to lie within the diameter plane of the array and
taking 6 = 0 as the main beam direction for the array shown in Fig. 12.1 requires
that the excitation at angle $ round the array of radius r is:
For omnidirectional elements and if the number of elements in the array is very
large (i.e. small interelement spacing), the normalised directional pattern in the
horizontal plane can be expressed as:
1 f27r \lnr 1
D(6) = — Jo F(0)exp; — cos (0-0)1 </0 (12.2)
giving
corresponding linear array of length equal to the diameter of the circle. In practice,
the interelement spacing needs to be about A/2 or less in order to approximate
to this function.
incoming wavefront
in 9/2)
-10-
dB
-20-
-30
80 120
9, deg
Fig. 12.2 Directional pattern of phase-compensated circular array for a continuous aperture
distribution with no amplitude tapering and fora diameter of 5\
The reduced beamwidth and increased sidelobes are due to the increased con-
centration of elements at the sides of the circular aperture (compared with a linear
array). It has been shown by Fenby16 and by James17 that low sidelobes can only
be achieved by means of a symmetrical amplitude taper on the front half of the
circular array with very low or zero excitation on the back half of the array. These
results assume omnidirectional elements. Electronic beam rotation of such patterns
therefore requires electronic control of both amplitude and phase of each element
The vertical pattern (through the plane of the main beam) is also a J0(x) func-
tion of the form
where a is the vertical angle measured from the plane of the circle. Fig. 12.3 shows
such a vertical pattern for a 5X diameter array. Again this result assumes a con-
tinuous array.
The vertical beamwidth is rather large since this is analogous to the equivalent
Circular arrays 301
'end fire' position for a linear array. A single ring circular array therefore provides
a convenient-shape fan beam for rotation through 360°, although it is important
to realise that it is not possible to exercise independent control of the horizontal
and vertical beamwidths for a single ring array of omnidirectional elements. The
amplitude tapers designed to reduce the sidelobes in the horizontal plane will also
increase the beamwidth in the vertical plane.
J o [ 5 T T ( 1 - C O S CX)]
160
Fig. 12.3 Directional pattern in the vertical plane of a phase-compensated circular array of
diameter 5X
The most well-known use of circular arrays is probably for the Wullenweber
array which employs an arc of about 120° and excites it with beam cophasal
excitation.3 It therefore represents a phase compensated curved linear array. The
process of electronic beam rotation involves commutating the required aperture
distribution around the array to the appropriate grouping of elements. Wullenweber
arrays generally use mechanically rotating goniometers or corresponding electronic
switches. An organ pipe scanner could perform the same function for waveguide
feeds. Such arrays have been used for HF receiving antennas and for direction
finding.
The effect of directional elements on beam cophasal patterns has received only
limited study and the evidence suggests that directional elements would not be
particularly helpful, with the notable exception of the case of the restricted arc
excitation. Most Wullenweber arrays employ directional elements of a simple
cardioid or half-cosine shape. An advantage of operating untapered beam cophasal
arrays is that since all the elements have the same amplitude, electronic beam
rotation may be achieved by control of phase alone. However, the sidelobe levels
are too high for most applications. The rotation of tapered excitations usually
requires control of both amplitude and phase and this represents one of the limi-
tations of beam cophasal systems due to the cost and complexity of scanning.
A circular array of omnidirectional elements is very prone to mutual coupling
effects. On transmission it is even possible for one element to pick up more power
by mutual coupling than it is trying to radiate (i.e. negative radiation impedance).
The effect of mutual coupling in beam cophasal circular arrays can be minimised
by using directional elements and restricting the active arc.
302 Circular arrays
Recent work has also shown that by departing slightly from the beam cophasal
excitation for phase but avoiding any amplitude taper, it is possible to achieve side-
lobe levels below — 20 dB. This offers pattern rotation by phase variation alone.18
Although there have been several experimental arrays, the practical utilisation
of beam cophasal circular arrays seems to have been mainly restricted to direction
finding applications and HF communication arrays where the availability of a
compact circular array antenna can replace a complete field of rhombic antennas.4
In such cases, wide bandwidth performance of the arrays is clearly essential and
scanning is performed by either a rotating coupler or set of switches for fixed
beams.
\ circular
array
elements
feed into
lens
circulator
A different approach for electronic scanning for beam cophasal arrays employs
an array signal processing technique. This achieves a continuous fixed-rate beam
scanning by means of synthesising the correct form of phase variation (with time
and space) that needs to be applied to each element of a circular array. For the case
of beam cophasal patterns this corresponds to sinusoidal frequency modulation
and the feeds to each antenna element can be obtained by applying this frequency
modulated signal to a suitable tapped delay line. The above scheme has been
demonstrated for a receiving array by applying the same set of delayed FM signals
to act as local oscillators for each element of the array.11 The experimental system
employed 16 elements working at 400 MHz and a fixed scanning rate of 400 kHz.
A schematic diagram of this concept is shown in Fig. 12.5. The project was aimed
at the receiving system for a within-pulse scanning radar. The scheme can also be
used to rotate a restricted range of amplitude tapers on the array and the technique
can be extended for non-beam cophasal excitations provided that the desired phase
variation around the array is not discontinuous.
A form of continuous modulation scanning has also been adopted for an elec-
tronically scanned cylindrical array applied to a TACAN navigational aid.22*23 This
application requires a continuous rotation of a specific directional pattern.
Several of the array signal processing schemes discussed in Chapter 13 may also
304 Circular arrays
Fig. 12.5 Schematic of system for continuous electronic scanning of the directional pattern
of a circular array
Since the beam cophasal pattern of a circular array has only one main beam,
the beam pointing direction may be chosen to lie anywhere over the surface of a
hemisphere. This requires suitable phase (or time delay) excitation appropriate to
both the azimuth and elevation angle required. There is, however, no control of
the corresponding beamwidths in the two dimensions, which will be determined
solely by the angle of elevation of the beam. At first sight this seems attractive in
terms of generating one or more beams from a single ring array and then using
phasing or commutation to rotate the pattern. The main problem is that the vertical
beamwidth from a single ring is, in general, far too large and the need to apply
amplitude tapering makes it even larger.
A planar aperture, such as a circle, can, in principle, deflect a pencil beam over a
complete hemisphere (subject to mutual coupling and element pattern constraints).
A circular array can be regarded as the outer ring of such a planar aperture. (This
explains the high sidelobe performance of the beam cophasal pattern). The com-
putation of array phasing for 2-dimensional scanning of a circular planar array is
substantial though it is possible to adapt some of the array signal processing
schemes to this end.
Circular arrays 305
In the early 1960s workers in both the UK and USA started examining the direc-
tional properties of circular arrays in terms of the Fourier analysis of both the array
excitation and the corresponding far-field pattern24"26. This approach has proved
extremely attractive for synthesising directional patterns and has led to more
convenient methods of exciting arrays and of achieving electronic beam rotation.
The technique also offers the capability of compensating for the effects of mutual
coupling. It is, perhaps, surprising that the applications of this concept have been
slow to start but are now being applied in DF and null steering and this will be
discussed later. Before covering the details of such schemes, the basic theory of
phase modes and amplitude modes for circular arrays will be reviewed briefly.
12.3.1 Concept of phase and amplitude modes for continuous circular arrays
Phase modes: First consider the case of phase modes for continuous circular
arrays, where the effect of interelement spacing can be neglected (i.e. for very
small interelement spacing). The excitation function for any continuous circular
array is a periodic function of angle with period 2TT radians. The excitation function
may therefore be analysed in terms of a Fourier series. It is convenient to use the
complex form of Fourier series where the excitation of the array F(<t>) is expressed
as:
F{<t>) = Z C m exp(/m0) (12.5)
m=-N
Here Cm represents the complex Fourier coefficient of each spatial harmonic of the
array excitation and eqn. 12.5 represents the excitation function F(0) as the sum
of (27V-f 1) terms. Each term is called a phase mode excitation of the array. Any
arbitrary excitation function can be expressed in terms of these exponential phase
mode components. The mth phase mode consists of an excitation which is constant
in amplitude but with the phase varying linearly with angle so that there are m
complete cycles of phase change around 360° of angle. The corresponding phase
mode of order — m has the reverse direction of phase change with angle. The far-field
directional pattern arising from the mth phase mode turns out to be an identical
function having a similar phase variation with angle and a constant amplitude given
by a Bessel function coefficient. This mth far field phase mode may be represented
by:
2 7 r
f C ( / 0 ) [ ^ ( 0 ^ ) ] ^ 0 (12.6)
Z7
giving
A » W = C m / m / m (0r)exp(/m0) (12.7)
where j3 = 2TT/X.
Since each spatial Fourier harmonic of array excitation gives rise to a corres-
ponding spatial Fourier harmonic of the far-field directional pattern, the above
306 Circular arrays
argument for Fourier analysis can obviously also apply to the complete far-field
directional pattern which is also a repetitive function of angle. Therefore, it is
possible to take any desired far-field pattern and break it down into its complex
Fourier components and to separately excite each of these Fourier components
around the array, taking note of the appropriate Bessel function term which relates
the array excitation of that particular mode to the corresponding far-field signal
strength. Needless to say, the case of the zero-order mode is when all the elements
are excited in phase and this gives rise to an omnidirectional far-field pattern of
constant phase, as would be expected from symmetry. This corresponds to the DC
term in the time/frequency application of the Fourier series. The higher-order
modes are also omnidirectional but with higher rates of progressive phase change
with angle. This method of analysing and exciting such pattern from circular
arrays has many advantages for pattern syntheses.
Amplitude modes: If the cos/sin form of the Fourier series had been chosen, the
basic constituent pattern would have been m cycles of amplitude variation around
the array. This is termed an amplitude mode and gives rise to a similar amplitude
mode in the far-field pattern with the same Bessel function coefficient relating
these two functions. It can easily be seen that the summation of a pair of positive-
and negative-sequence phase modes of the same order will give rise to a corre-
sponding amplitude mode. By choosing the phase combination of each pair of
phase modes one can construct either a set of'cosine type' amplitude modes (used
for synthesising even-order patterns such as beam patterns) or 'sine type' amplitude
modes. The latter can be used to synthesise odd functions in directional patterns,
such as the monopulse-type split beam pattern. The far-field amplitude modes may
also be regarded as angular spatial harmonics.
This corresponds to exciting all the modes to the same level in the far field, resulting
in a directional pattern of the form:
uniform linear array, but it is important to realise that this function now extends
over 360° and relates to a 6 variable, rather than 180° and the sine 0 variable for
linear arrays.
The use of the Fourier technique for exciting circular arrays also means that
all the well known pattern synthesis techniques for linear arrays may be trans-
ferred to circular arrays. The taper function is then applied to the mode excitation
rather than to the aperture function. The resultant pattern is then identical to that
of the linear array theory, but corresponding to a 0 scale rather than a sin 6 scale.
In general, low sidelobe tapers will have higher excitations for the lower-order
modes. The types of network needed to excite such phase modes are discussed in
Section 12.3.4.
which equals the beamwidth of a linear array having the same dimensions as the
circular array diameter.
Even when the radius is not small, it is still possible for certain modes to be
difficult or impossible to excite if Jm(fir) is small, or zero. A physical explanation
of this effect can be obtained with the help of Fig. 12.6. For the direction 6 = 0,
the far-field phase of the signal received from a small arc of the continuous array
located at 0 and excited by mode m is
cos(pr-ir/4-mnl2)
J (12.16)
m\Pr) *
Clearly (3/* should be a multiple of n/2 to avoid zeros for any coefficient; i.e. the
diameter should be an integral number of half-wavelengths.
Fig. 12.6 Location of the 'phase centres' for the mth phase mode
The bandwidth of the array can be assessed, since |3r must not be allowed to
change by more than, say, 7r/8 to avoid excessive changes in the Bessel coefficients.
This condition gives
SX/A « ± 16X/r (12.17)
asymptotic expansions given above. If /3r cos a changes by 7r/8, there will be sub-
stantial changes in the Bessel functions. Thus the beam limits can be considered
to be given by
/3rcosa = /3r — TT/8 (12.19)
2
By approximating cos a as 1 — a /2, the vertical 'beamwidth' becomes
6 = 2a^^(XJ2r) (12.20)
which equals that of a linear endflre array of length equal to the diameter.
The above implies that the directional response with varying 6 but at angles out
of the plane of the array (a =£ 0) remains a phase mode but the amplitude of the
mode is a function of a. This represents a further problem for pattern synthesis,
since if the design produces a specified pattern in the plane of the circle, the shape
of the response at different vertical angles (different a) changes due to the different
relative levels of the phase mode components. A low sidelobe pattern at a = 0
could therefore deteriorate at higher angles. Fortunately, the above disadvantage
can be overcome by the choice of suitable directional elements as discussed later.
The aperture excitation for a phase mode of order m with discrete omnidirec-
tional elements is given by:
Fm(6) = CmeimeS(d) (12.22)
where Cm is the excitation level of the mode.
Fromeqns. 12.21 and 12.22:
Since this represents Fm(6) in terms of an infinite series of phase mode excitations,
the corresponding far field can be expressed as:
where g = (nq — m), h = («g + ra), « — total number of elements and Cm = exci-
tation of rath mode.
The first term in the above equation represents the required far-field phase mode
pattern but the remaining terms correspond to higher-order distortion modes.
As in the case of the continuous array, there can be little radiation for modes
higher than ± 0r. Now the first higher-order distortion modes are separated by n
basic modes from the desired mode ra, so that it can be anticipated that the higher-
order modes due to the finite number of elements will not radiate very strongly
provided that
(n-m)>(3r (12.25)
Since the maximum value of m is |3/% this condition becomes
n>2$r (12.26)
which confirms that the inter-element spacing should be
d = 2nr\n<\\1 (12.27)
The properties of a circular array of omnidirectional elements with the above
spacing are thus similar to those of a continuous array. It may still be difficult to
excite all the desired modes and the array is narrowband. These difficulties can be
attributed to the two 'phase centres' of each mode, which are separated by a large
distance in a large-diameter array. This suggests that improvements can be obtained
by employing directional elements, which effectively suppress the radiation from
one of these phase centres.
that they improved the performance of phase modes. These results showed that
the amplitude of the phase modes did not tend to zero in the horizontal plane
when/ m ($r) became zero.
The situation has been recently analysed by Rahim27'42 who has derived general
expressions for phase modes excited from directional elements and the following
represents extracts from his arguments.
Consider an ^-element array where each directional element has a pattern
G(\p) and the elements are disposed to retain circular symmetry. In most practical
applications this means directional elements pointing radially outwards from the
array. The pattern G(4J) can be represented in terms of its Fourier components:
P
G(i//) = £ Bhem (12.28)
Cm £ {\Anq+m\expj[(nq +m)6
i// m - nQ ]} (12.35)
Higher-order distortion terms again arise due to the finite inter-element spacing.
Eqn. 12.35 also implies that distortion terms can be more significant with direc-
tional elements. This means that it is not possible to use directional elements in
order to relax the inter-element spacing, and hence economise on numbers of
elements. As noted earlier, there is little radiation from these higher-order terms
provided that the inter-element spacing is low (d < X/2) or the pattern does not
attempt to excite mode orders approaching the limit of (|3r).
Most experimental circular arrays with directional elements have employed slot
radiators or dipoles in front of reflecting screens. Examples of such arrays are
shown in Figs 12.19, 12.21 and 12.22, referred to later in the Chapter. Rahim
looked at the particular case of D(\p) = (1 4- cos \p) since this could be a realistic
approximation to many practical antenna elements. In this case the function
relating the far-field mode to the array excitation reduces to:
Dm{6) = Cmjme>m6{Jm($r)-jJ'm(Vr)) (12.36)
Examination of Bessel-function graphs reveals that when / m ($r) -> 0, J'm(fir) is
approximately a maximum and vice versa (except for $r -> 0). Therefore the resul-
tant far-field values of the modes becomes approximately constant. This offers
wide-bandwidth performance and the ability to design low sidelobe patterns by
means of accurate control of relative mode amplitude. It also explains why previous
experimental results have been better than expected.
An indication of the dramatic improvement achieved by the use of (1 4- cos \p)
elements can be seen from Fig. 12.7. This shows the theoretical variation of relative
mode gain with frequency for the first-order mode for both omnidirectional and
directional elements. These results correspond to an array having a diameter of 2X
at 300 MHz. Some experimental results28 confirming this for the same diameter
array are shown in Fig. 12.8. These results indicate that modes 0, 1 and 2 remain
stable in amplitude over an octave.
Circular arrays 313
The above evidence confirms the need for directional elements for circular
arrays where wide bandwidth and/or low sidelobes are required. Element patterns
of the form (1 + cos \j/) seem near optimum for such applications and the element
spacing should be A/2 or less to minimise distortion terms.
Fig. 12.7 Theoretical results showing the stability in the amplitude of mode 1 vs. frequency
due to the use of directional elements of the form (7 + cos 0) for 2\ radius array at
300 MHz.
— '1 + cos 0' elements
omnidirectional elements
dB
mode 0
mode 1
°mode 2
The distortion terms associated with the element spacing give rise to departures
from true omnidirectional performance (amplitude ripples) and departure from the
linear variation of phase with angle.29 Fig. 12.9 shows a typical experimental result
for the polar diagram of a phase mode. These results relate to a 16-element array
314 Circular arrays
having element patterns of the from (1 + cos;//). The amplitude ripple is mainly
due to distortion terms corresponding to an element spacing of d = X. Fig. 12.10
shows a theoretical result for the variation of phase with angle for a first order
phase mode on a 4-element array. These cyclic phase errors are clearly important
for the design of DF systems and also represent phase errors for the constituent
components of any synthesised directional pattern.
Oth
Fig. 12.9 Experimental directional patterns showing the amplitude variation of the zero
and 1st order phase modes for a 2-wavelength circular array employing directional
elements of the form (/ + cos 0)
Rahim's analysis also showed that the use of directional elements of the form
(1 + cos \jj) also improved the vertical plane response of the phase modes27 by
removing the zeros of the pattern (which again corresponded to zeros of a Bessel
function.) Thus the use of a phase-mode synthesis technique in the plane of the
circle could retain the pattern shape out of this plane with such directional ele-
ments.
Several authors have shown that the well known Butler matrix network, used to
generate multiple beams for linear arrays, may also be used as a convenient way of
exciting the phase modes of a circular array. If an n x n port Butler matrix is
connected to an ft-element circular array, then each of its n input ports produces
the correct array excitation for the corresponding spatially-orthogonal set of n
possible phase modes that can be excited from such an array. This arrangement is
shown in Fig. 12.11. Excitation of each input port of the Butler matrix corresponds
to the excitation of a phase mode, so any (spatially band limited) pattern may be
excited from a circular array by applying the correct values of amplitude and phase
to these ports. This excitation function will depend upon the Fourier terms of the
180- far-field
phase shift
160 -
uo-
120-
100-
80-
60
40
20
0 —I . 1
180 210 240 270 300 330 360 9
-20
azimuth angle
-UO
-60-
-80
-100-
-120
-KO- §
oT
-160
-180
Fig. 12.10 Cyclic phase errors for first-order phase mode due to use of four omnidirectional elements at an interelement spacing of 0.4X
Calculated first order phase mode f o r four omnidirectional elements
Ideal first order phase mode (continuous array) Co
316 Circular arrays
desired directional pattern and the appropriate Bessel coefficient relating each
array phase mode to the far-field phase mode (which is also a function of the
directional elements used). This synthesises the desired directional pattern pointing
in one fixed direction.
T T T T T \
E E E E E E
i1 I2 i3 \U i5 in
Fig. 12.11 Butler matrix network with its output ports feeding the elements of a circular
array. Excitation of each input port to the network excites a corresponding
phase mode on the circular array
a = circular array with n elements
m = matrix network
A further important and useful property of this Butler matrix feed is that if a
linear progressive phase taper of value 0, <p, 2$, 3 0 , . . . is applied to the input ports
of the Butler matrix in addition to the above required excitation, then the resulting
signals to the array elements commutate around the circumference on the array.
This results in an angular rotation of the far-field pattern through the angle (j). This
makes the Butler matrix feed network a very attractive system for both excitation
and electronic rotation of patterns for circular arrays. A schematic representation
of this system is shown in Fig. 12.12. This result was shown by Davies26 and
experimental arrays based upon this principle have been described by Sheleg10
and by Chow and Davies30
Fig. 12.13 shows an experimental result from Reference 30 which relates to the
above type of scanning system when feeding a 16-element circular array of mono-
pole elements located over a flat ground plane at 400 MHz. The excitation involved
amplitude modes of order 0 to 6 with an amplitude taper to achieve — 23 dB side-
lobes. The results relate to a continuous beam scanning at a rate of 100 kHz.
An alternative way of looking at this Butler matrix feed system is to note that
it performs a transformation between the excitation techniques of linear and
circular arrays. Clearly, any technique used to achieve linear progressive phase shifts
across a linear array may be used with a Butler matrix to achieve beam deflection
Circular arrays 317
for circular arrays. The results of Fig. 12.13 were obtained from a single-sideband
type array signal processing system designed for continuous scanning for linear
arrays. This scanning concept is discussed in Chapter 13.
Fig. 12.12 Schematic diagram of arrangement for electronic angular deflection of any
directional pattern formed by a circular antenna array. The pattern is excited
in terms of its constituent phase modes via the Butler matrix and a linear pro-
gression of phase shifts across the input ports of the matrix produces linear
angular shifts of the resultant directional pattern
a = phase shifts to deflect pattern through angle 0
b — values of amplitude and phase shift required to synthesise directional pattern
m = matrix network
Fig. 12.13 Experimental directional pattern of 16-element circular array undergoing con-
tinuous electronic pattern rotation at 100kHz (— 23 dB sidelobe level)
In principle, all designs of multiple-beam networks used for linear arrays may
also be employed to excite phase modes of circular arrays. The phase mode exci-
tation, however, requires constant values of phase shift over the frequency band of
interest. This can only be achieved by the Butler matrix configuration (when
suitably modified for wideband operation) since other forms of network, such as
Maxson matrix and lens configurations, are based upon path length (time delay)
compensation. Wide-bandwidth Butler matrix networks have been constructed for
16-port designs28 and the technique is, in principle, extendable to larger systems31
at the expense of additional complication. A recent design9 for an octave bandwidth
318 Circular arrays
All arrays suffer from the effects of mutual coupling between their constituent
elements and circular arrays are no exception. In general, the effects are more
severe for such arrays since significant coupling can occur between elements located
diametrically opposite each other in addition to the strong coupling to adjacent
element locations. However, in the case of the excitation of a single-phase mode
around the array, the electrical environment of all the elements is identical, due to
symmetry. This means that all the elements will exhibit an identical radiation
impedance. Nevertheless, mutual coupling effects still modify the value of this
radiation impedance, but in the case of a single-phase mode, all the values are
changed by the same amount.32 Such a radiation impedance, under phase-mode
excitation, is often referred to as the phase-sequence impedance of that mode.
An identical argument applies to all the other phase modes so that the effect
of mutual coupling is to modify the various sequence impedances and hence the
matching at the input port of the matrix associated with each phase mode. There-
fore if phase mode excitation is employed with a network (such as a Butler matrix)
having a separate feed port for each phase mode, it is possible to completely com-
pensate for the effects of mutual coupling by compensating for this impedance
mismatch at each input port. Thus a set of fixed impedance compensations at
each input port will be adequate to compensate for the effects of all mutual
coupling, irrespective of the excitation of the array or of the direction of beam
pointing. For wide-bandwidth operation it is necessary to match each mode impe-
dance over the operating band. The use of directional elements serves to reduce
the effects of mutual coupling since it can reduce diametrical coupling across
the array. The above compensation can, however, still be applied.
Correction of mutual coupling effects may be performed theoretically by
calculating the value of the phase-sequence impedance, and hence calculating the
change required in the excitation of the mode. Reference 32 gives detailed calcu-
lations for such mutual coupling. Alternatively, it is possible to obtain such results
experimentally by exciting each phase mode one at a time, measuring the field
strength and phase achieved in the far field and then modifying the excitation
of the appropriate port in order to achieve the desired value of field strength for
Circular arrays 319
each constituent mode. This attractive feature gives powerful advantage to circular
arrays for those applications where mutual coupling can limit the performance of
an antenna. This can apply, in particular, to low sidelobe applications and small
aperture arrays.
In the case of wideband excitation, in principle, it is necessary to produce an
excitation for each mode which varies with frequency in such a way as to maintain
the correct pattern, taking account of mutual coupling. There is little published
work on this topic but the limited evidence on the wideband stability of modes
when using directional elements suggests that this may not be impracticable.
The rather limited number of applications for circular arrays has been mainly due
to the problems and complications of array excitation and beam rotation. In the
case of beam cophasal patterns this has been particularly difficult and, in general,
has been solved by multiple beams or mechanical rotating goniometers. The
development of phase mode theory in the 1960s represented a substantial step
forward, coupled with the use of Butler matrix systems to feed the modes. Antenna
engineers have studied these concepts for several applications including radar, null
steering and DF systems.
multiple input feeds is the HF wire grid Luneberg lens.21 An antenna based on this
principle was constructed for the FAA near Honolulu33 covering the 3-30 MHz
band and consisting of a 600 ft-diameter wire grid structure.
Fig. 12.14 Some measured directional patterns of the 16-element wide-bandwidth circular
array described in Reference 19. Curve (a): 205MHz; curve (b): 480MHz;curve
(c): 1350 MHz
> [first-order
zeroth mode
^ Y phase mode
variable [~7| null-setting
phase-shifter A ^ o . L_jJ attenuator
(c)
Fig. 12.15 Schematic diagram for broadband feed system for producing a single steerable
directional null
linear arrays is still quite limited. Nevertheless, several experimental systems and
subsystems have been constructed. These have included an array employing diode
switching of signals, constructed by Wheeler laboratories, and a further switching
scheme which could also rotate an amplitude taper, developed by the Naval Elec-
tronics Laboratory at San Diego. Such switching systems have tended to be com-
plex, particularly where amplitude tapers have been involved.
Fig. 12.16 Photograph of experimental 4-element null steering array (see Reference 8)
(/ \ X /^n-A /
-10-
/ \ —A./ >4—null width
r58
dB \ X / °
\ f-i
-20-
circular l ^ \ I \ 1
array \ / \ /
-30-
I
0° 90° 180° 270°
angular positions of array elements
Fig. 12.17 Experimental null patterns for 4-element null steering array at 153 MHz. Curve I
shows null pointing in directions corresponding to element locations. Curve II
shows null pointing between direction of element locations
Experimental circular array systems have also been constructed using lens
techniques to feed the array. These, in general, provide beam cophasal type exci-
tation. Lens systems may employ geodesic lens configurations or an R-kR lens. This
latter approach has been adopted by the Naval Electronics Laboratory, San Diego,
employing a parallel plate construction for the lens. Papers covering several such
systems were presented in two conferences on conformal arrays37'38 at NEL San
Diego in 1970 and 1972.
fmder' whose principles were first described by Earp and Godfrey.39 This involved
a circular array of antenna elements whose outputs were sequentially sampled by a
switch. Early versions used a mechanically rotating switch, but electronic switching
was subsequently used.
The output from such a switch, when the array is excited by a single plane
wave, represents a sampled version of a phase-modulated waveform in which the
phase deviation is given by the diameter of the array in wavelengths. By comparing
the phase of the phase modulation with that of the switching control, the angle
of arrival may be determined. Increased bearing accuracy can be obtained from
larger diameter arrays without the need for rotating parts. This technique used
in reverse for transmission is also the basis for the Doppler VOR and Tacan
antennas.22'23
Wullenweber arrays have been extensively used for DF applications in the HF
band and this work is reviewed by Gething.3'40 Several large antennas of this type
have been constructed in various parts of the world. The University of Illinois
system41 covers the 4-16 MHz band employing a single ring of 120 elements on a
circular of diameter 955 ft.
More recently work has been reported on octave-wide microwave DF by the
use of a circular array comprising 16 elements fed by a wide-bandwidth Butler
matrix.9 This provides a DF accuracy of 0.17° rms over the 2-4 GHz band. The
principle involves comparing the phase of signals received by two phase mode
1
to
O
Fig. 12.19 Photograph of NRL cylindrical array employing printed-circuit dipoles and feeds for the linear arrays forming the cylinder
to
Fig. 12.20 Experimental directional patterns for the cylindrical array of Fig. 12.19. The set of patterns correspond to 36Cf coverage in azimuth
when the elevation beam direction is 4<f above horizontal
Circular arrays 327
patterns. If the order of the modes differs by one, then the phase difference gives
an unambiguous output of bearing. Higher accuracy can be achieved by using
more than two modes. This concept is being marketed by Anaren for an ESM
system. A photograph of the antenna is shown in Fig. 12.21.
Fig. 12.21 Photograph of 16-element circular array used for ESM direction finding system
by Anaren
The use of phase modes in this type of DF system emphasises the wide-bandwidth
properties of the phase-mode concept when used with suitable directional elements.
The accuracy of such a system is also affected by the higher order distortion terms.
Fig. 12.10 showed a theoretical example of phase errors for a phase mode pattern
due to the finite interelement spacing. This will translate into a bearing error for
this type of DF system. The solution is to employ many elements and use low-order
modes or to reduce the interelement spacing, or simply to 'calibrate' the mode's
phase versus angle response.
12.7 References
3 GETHING,P. J. D.: 'High-frequency direction finding', Proc. IEE, 113, Jan. 1966, pp.
49-61
4 STARBUCK, J. T.: 'A multiple-beam HF receiving aerial system', Radio Electron Engr.,
37, April 1969, pp. 229-235
5 WYATT, J. E. G.: The design of an omnidirectional UHF wrap-round antenna', /. Royal
Naval Scientific Service, 29, pp. 251-258
6 HANSEL, P. G.: 'Doppler-effect omnirange', Proc. IRE, 41, Dec. 1953, p. 1750
7 GREGORWICH, W. S.: 'An electronically despun array flush-mounted on a cylindrical
spacecraft', IEEE Trans., AP-22, Jan. 1974, pp. 71-74
8 DAVIES, D. E. N., and RIZK, M. S. A. S.: 'A broadband experimental null-steering
antenna system for mobile communications', Radio Electron Engr., 48, Oct. 1978, pp.
511-517
9 REHNMARK, S.: 'Improved angular discrimination for digital ESM systems', Proc. MM80,.
Microwave Exhibitions and Publishers Ltd., 1980
10 SHELEG, B.: 'A matrix-fed circular array for continuous scanning', Proc. IEEE, 56, Nov.
1968,pp. 2016-2027
11 FENBY, R. G., and DAVIES, D. E. N.: 'Circular array providing fast 360° electronic beam
rotation', Proc. IEE, 115, Jan. 1968, pp. 78-86
12 SHELEG, B., and WRIGHT, B. D.: 'A 3-D electronically scanned S-band cylindrical
array', Proc. of conformal array antenna conf., NELC, San Diego, CA. Jan. 1970
13 BOGNER, B. F.: 'Circularly symmetric RF commutator for cylindrical phased arrays',
IEEE Trans., AP-22, Jan. 1974, pp. 78-81
14 SKAHILL, G., and WHITE, W. D.: 'A new technique for feeding a cylindrical array',IEEE
Trans., AP-23, March 1975, p. 253
15 ZIEHM, G.: 'Optimum directional pattern synthesis of circular arrays', Radio Electron
Engr., 28, Nov. 1964, pp. 341-355
16 FENBY, R. G.: 'Limitations on directional patterns of phase compensated circular arrays',
Radio Electron Eng., 30, Oct. 1965, pp. 206-222
17 JAMES, P. W.: 'Polar patterns of phase corrected circular arrays', Proc. IEE, 112, 1965,
p.1839
18 WATANABE, F., GOTO, N., NAGAYAMA, A., and YASHIDA, G.: 'A pattern synthesis
of circular arrays by phase adjustment', IEEE Trans., AP-28, Nov. 1980, p. 857
19 LONGSTAFF, I. D., and DAVIES, D. E. N.: 'A wideband circular array for HF communi-
cations', Radio Electron Engr., 35, June 1968, pp. 321-327
20 HOLLEY, A. E., DUFORT, E. C, and DELL-IMAGINE, R. A.: 'An electronically scanned
beacon antenna', IEEE Trans., AP-22, Jan. 1974, p. 3
21 TANNER, R. L., and ANDREASON, M. G.: 'A wire-grid lens antenna of wide application',
Proc. IRE Trans. PGAP, July 1962, p. 408
22 CHRISTOPHER, E. J.: 'Electronically scanned TACAN antenna', IEEE Trans., AP-22,
Jan. 1974, p. 12
23 SHESTAG, L. N.: 'A cylindrical array for the TACAN system', IEEE Trans., AP-22, Jan.
1974, p. 17
24 TILLMAN, J. D., HICKMAN, C. E., and NEFF, H. P.: The theory of a single ring circular
array', Trans. Amer. Inst. Elec. Engrs., 80, Pt. 1,1961, p. 110
25 LONGSTAFF, I. D., CHOW, P. E. K., and DAVIES, D. E. N.: 'Directional properties of
circular arrays', Proc. IEE, 114, June 1967, pp. 717-718
26 DAVIES, D. E. N.: 'A transformation between the phasing techniques required for linear
and circular aerial arrays', Proc. IEE, 112, Nov. 1965, pp. 2041-2045
27 RAHIM, T.: 'Directional pattern synthesis in circular arrays of directional antennas', Ph.D.
Thesis, Univ. of London, 1980
28 RAHIM, T., GUY, J. R. F., and DAVIES, D. E. N.: 'A wideband UHF circular array',
Proc. 2nd International Conference on Antennas and Propagation, April 1981
29 DAVIES, D. E. N., and GUY, J. R. F.: 'A UHF communication antenna employing
Circular arrays 329
13.1 Introduction
Array signal processing (ASP) represents a class of antenna system where the output
from an array is subjected to various forms of signal processing (e.g. phase and
amplitude control, time delay or frequency translation) in order to produce outputs
which can simultaneously provide angular information relating to several different
directions.
When the outputs of a multiple element array are combined via some passive
phasing network such as a corporate feed or combination of hybrids, the phasing
will usually arrange for the output of all the elements to add coherently for a given
(main beam) direction. It is interesting to consider what happens to the power asso-
ciated with an incident wave arriving from an angle corresponding to a zero of the
directional pattern. If the phasing network is constructed from passive transmission
lines, then such incident power must be re-radiated. If the phasing network contains
hybrids, then this incident power may flow into the 4th port terminations of such
hybrid units. This is an indication that, although the array receives signals from a
wide range of incident angles, the phasing network provides the directional res-
ponse. Clearly, it is possible to utilise this power by designing multiple port phasing
networks which can simultaneously provide several outputs corresponding to differ-
ent beam directions. This aspect has already been discussed in Chapter 6 which also
drew attention to the need for the directional patterns resulting from such multiple
beams to be spatially orthogonal in order that the outputs from these ports may be
electrically uncoupled.
This corresponds to a simple example of an ASP antenna system which could be
used to transmit or receive independent signals to or from different directions. In
this instance, the processing consists of time delays or phase shifts needed to
synthesise multiple beams, but several other techniques including modulation
methods may be employed to control the shape and pointing direction of the
antenna directional characteristic. In some instances the processing schemes may
make use of non-linear techniques (e.g. signal multiplication, square-law detection,
hard limiting etc.).
Array signal processing 331
The principal use of ASP techniques has so far been in radar, sonar, radio astron-
omy and certain navigational aids. In such cases, the properties of the directional
pattern are determined by the signal processing system which is also closely related
to the system application. It is therefore necessary to discuss ASP techniques in
relation to some form of radar or other application area, since the 'antenna system'
can no longer be considered as a separate isolated unit.
A different aspect of ASP which has received considerable attention over the last
few years is adaptive arrays and null steering antennas. Here, instead of defining a
number of wanted directions for receiving signals we can specify some wanted and
some unwanted directions; the latter may contain high-level interfering signals
which must be minimised by means of directional nulls in the patterns. In some
cases the directions of these nulls may be fixed or pre-programmed, but in many
cases the antenna response must be adaptive. This means that the shape of the
directional pattern is determined in response to the angular distribution of received
signals in order to try to meet some desirable criterion.
The simplest form of adaptive and null steering antenna is probably the sidelobe
canceller. This can steer a single directional null within the sidelobe region of a nor-
mal directional pattern. Multiple null adaptive arrays in general define a 'wanted
direction' and then introduce one or more feedback controls to minimise the signals
from all other directions. Null steering systems have applications in radar, sonar and
communication systems. In the latter case, there may be additional complications if
it is not possible to specify the direction of the wanted signals. The field of null
steering and adaptive arrays is covered in Sections 13.6 and 13.7.
sector illuminated
by wide-beam floodlight
transmission
n separate receiving
beams covering sector
The use of multiple receiving beams with a single floodlight transmission in this
configuration has a number of important consequences. First of all, there is a sub-
stantial increase of data renewal rate, since information is now received from all tar-
gets at all ranges and all bearings for each transmitted pulse. The essence of this is
that the ASP array effectively looks in many directions simultaneously. It is this
feature which essentially distinguishes ASP from a conventional electronic beam
scanning system. An amplitude-comparison monopulse antenna may therefore be
regarded as a 2-beam form of ASP.
The effect of the above processing on the maximum range performance of the
radar is rather complex, since in practice it will depend upon the form of pulse to
pulse integration used. This is because the ASP radar receives N times as many
pulses, each having one Nth of the power of the conventional, sequential scanning
radar. Other features of such a radar include the use of fixed antennas so that the
334 Array signal processing
system may be classed as an electronic scanning system, but without the need for
electronic phase control of the high power radar transmissions. The fixed antennas
also give a certain advantage for coherent radars and MTI systems by the complete
elimination of loss and spectrum spreading effects due to scanning modulation (this
is an advantage which does not apply to conventional electronic scanning). An
important feature of such a radar is that the overall sidelobe performance of the
system is determined entirely by the one-way sidelobes of the receiving array. Since
sidelobe levels are inevitably determined by the phase and amplitude accuracy of
the aperture excitation function, such accuracy must also be reflected in the cor-
responding receiver signal processing.
The simplest way of implementing such a multiple beam receiver would be to
employ a reflector or lens incorporating a number of primary feeds. Alternatively,
it is possible to employ a multiple beam matrix network such as the Butler matrix
or the Blass matrix. It would be necessary to provide a separate receiver at each out-
put port as shown in Fig. 13.2, but the directional patterns of such outputs are
necessarily limited by the spatial orthogonality criterion discussed earlier. The
antenna gain of each beam will be equal to the full gain of the array in each direc-
tion (taking into account the cos# variation of effective aperture). In practice, this
gain would be reduced by any loss associated with the multiple beam matrix or
lens/reflector system.
• •
T
Rx Tx
beam-forming network
0
1 ! I
Fig. 13.2 Schematic arrangement associated with the coverage described in Fig. 1
Antenna array feeding multiple beam forming network on reception
If band limiting RF amplifiers are introduced after each antenna element, this
allows the SNR of the system to be defined before any beam forming or processing
is introduced. This offers additional flexibility and opens the way to many new pro-
cessing schemes. The first additional flexibility is that the multiple beam network
may be lossy and may produce non-orthogonal beams without restriction on either
antenna gain, sidelobe level or beam overlap. It may also be more convenient to fre-
quency change the output of all the elements down to some intermediate frequency
where the design of the multiple beam matrix network may involve more conven-
ient electronic components.
Array signal processing 335
It is also worth noting that amplitude tapers applied after amplification will
obviously taper both signals and receiver noise together and hence not worsen SNR
to such a large extent. In practice, this is worth about 1-2 dB in SNR.
Multiple beam networks may also be regarded as performing a sampled spatial
Fourier transform on the output signals from the array, since they provide a spatial
sampled version of the angular distribution of received signals. This obviously
points to the use of on line computers to perform fast Fourier transforms (FFT) to
form multiple beams. Digital FFTs are being used for Doppler spectrum analysis of
radar signals and for both Doppler and bearing analysis for sonar signals. The power
of such processors is increasing, particularly due to the development of dedicated
computer systems, and their performance is now adequate for application to
medium resolution radar systems. FFTs are now a most suitable choice for multiple
beam forming for either CW radars or low resolution pulse radars. Digital FFT
analysers have already been employed for airborne and spaceborne synthetic aper-
ture radars.9
v array
loca I oscillator
bipoI a r
A/ D A/ D video signals
convertor convertor
F.F.T. processing
output
Fig. 13.3 Formation of multiple receiving beams by fast Fourier transform (FFT) processing
of bipolar video signals from receiving array
The use of such digital processing in general involves translating the signals down
to a low IF which in many cases may be a bipolar video (zero frequency IF). The
signals are then sampled and fed via an A/D convertor into a digital FFT processor.
Such an arrangement is indicated in Fig. 13.3. It is interesting to note that the sche-
matic representation of a FFT process is identical to the physical representation of
the Butler matrix network which was developed several years earlier. There are a
number of processing techniques capable of providing either the Fourier transform
or the angular power spectrum of signals received by an antenna array. Many of
336 Array signal processing
these schemes may also be realised in terms of either surface acoustic wave (SAW)
or charge coupled device (CCD) technologies.10 Recent research is also looking at
optical signal processing techniques for beam forming.11
where co0 is the carrier frequency. The output from each filter will have the direc-
tional characteristics of a single fixed beam in space and the set of outputs will cor-
respond to a family of overlapping beams whose positions, beamwidths and sidelobe
levels can be controlled by the choice of filter characteristics.12 A schematic dia-
gram of such a system is shown in Fig. 13.4. In order to ensure that the noise per-
formance is not worsened by the wide bandwidth sampling switch, it is necessary to
define the noise bandwidth of the system by narrow band filters prior to the
sampling. It is possible to employ a stage of amplification at RF or alternatively to
frequency change and incorporate the sampling and the amplification at IF.
amplifier
BW -nlf
array
centre
per filter frequency u>0
Fig. 13.4 Formation of multiple receiving beams by Nyquist rate time sampling of the out-
put from elements of a linear array and subsequent band-pass filtering
The equivalent antenna gain of the system would be the same as for other forms
of multiple beam receiving system of comparable front end noise factor. It is worth
emphasising that the full receiving antenna gain applies, despite the fact that only
one element at a time is being used.
When used as a radar system the sampling rate of the element outputs must cor-
respond at least to the radar bandwidth. Therefore for a pulse duration of r the
bandwidth in the medium is 1/r and after the sampling switch is at least n/r (for an
H-element array). If n orthogonal beams are formed by the filtering process (i.e. a
set of rectangular filter characteristics) then this gives n output channels each of
Array signal processing 337
bandwidth 1/r as for other multiple beam configurations. Filtering can be designed
to provide non orthogonal beams if required and digital filtering may also be used
at this stage.
Although the above concept has been tried only experimentally for a radar appli-
cation13 it is in general use in the navigational aid field but used on transmission.
The Doppler VOR beacons involve switching a transmitting source sequentially
around the elements of a circular array in order to radiate a signal which is effec-
tively frequency coded in space. The more recent developments of Doppler micro-
wave landing systems employ the same techniques on linear arrays.4 Here the signal
received in space from the sampled transmitting array is characterised by a
frequency shift proportional to the angle off boresight. The latter development also
incorporates a range of advanced signal processing schemes to control beam shapes
and minimise errors due to multipath effects.
13.2.3 Equivalence between multiple beams and Nyquist rate beam scanning
A direct alternative to multiple receiving beams is to employ a single receiving beam
and to scan this across the sector of interest, repetitively, at the Nyquist sampling
rate. This is equivalent to using a single receiving beam to sample the fixed beam
positions of a multiple beam receiver. We can again refer to our sector coverage
radar system to describe this concept.
sector illuminated
wide-beam
transmission
single receiving
beam scans across
sector at Nyquist rate
Fig. 13.5 Sector illuminated by broad-beam transmission employing a single receiving beam
scanning across this sector at the Nyquist scanning rate
Fig. 13.5 shows a sector illuminated with a floodlight transmission with a pulse
of duration r. A narrow receiving beam is scanned repetitively across this sector at a
rate corresponding to 1/T. The scanning rate therefore corresponds to the Shannon
sampling rate or Nyquist rate. This means that pulses reflected from each target will
be sampled by the scanning receiving beam. The technique is in operational use in
338 Array signal processing
the sonar field,3 but the radar application has been mainly restricted to experimental
equipments.14'15j 16
There are several alternative system configurations for providing Nyquist rate
scanning of a receiving beam and the most well known is shown in Fig. 13.6. Here,
the output of each array element is subject to single sideband frequency translation
by angular frequencies of:
(where to s , the beam scanning rate, = 2IT/T). Such a scheme effectively adds phase
shift at uniform rates to each antenna element, thus producing a uniform move-
ment of the array directional pattern along the sin# scale. Each diffraction maxi-
mum of this pattern represents a successive beam which scans across the sector.
7 receiving array
I single sideband
| / j ~ l modulators
(n-1)u>s
amplifiers (bandwidth MY)
combined output
(bandwidth n / r )
Fig. 13.6 Modulation scanning system to produce repetitive scanning of receiving beam at
rate of u>s
u?o = 2-nlr
n-element array
RF a m p l i f i e r and filter
( b a n d w i d t h UDS)
single-sideband
modulators
Fig. 13.7 Frequency filtering to apply low sidelobe amplitude tapers to the output of a
receiving array incorporating within pulse modulation scanning
The range resolution is still determined by the transmitted pulse duration r, and
the sidelobe performance is set by the one-way sidelobes of the receiving array. The
scanning receiver has the useful feature that array amplitude tapers may be achieved
340 Array signal processing
by means of filters situated at the output of the combined array. This is illustrated
by Fig. 13.7. Here, the output of a receiving array is subjected to RF amplification
and frequency translation by integral multiples of the beam scanning rate cos. The
signals are then combined and translated to IF. If we consider the IF time waveform
(and its associated spectrum) due to the excitation of the array by a single far field
source, the envelope of the time waveform will be a sin npj'sinp function and repre-
sents the directional pattern of the receiving array. This repeats with a period
2IT/(JOS. This is shown in Fig. 13.7 (output A). The spectrum of this waveform con-
sists of n uniform amplitude spectral lines, one line for each element (or receiver
channel) of the array.
It is evident that if this spectrum is subjected to an IF filter having a character-
istic Tt(co) it acts as a taper function Ti(x) applied to the receiving array. This is
obviously a simple way to add or change a taper, but the laws of circuit theory
present certain constraints on simultaneous control of amplitude and phase, which,
for example, preclude the use of discontinuous taper functions. Such constraints do
not occur for direct control of the phase and amplitude of the signals from the ele-
ments. This situation is depicted at B in Fig. 13.7. These are also examples of post
amplifier tapers.
A further modification of this concept is indicated at C in Fig. 13.7. Here, the
video (post detected) spectrum and wide bandwidth waveform of the scanning
beam is shown. For the simple case of square law detection and a uniform array,
the post detected spectrum is triangular and the corresponding detected beam shape
is (sin np/sinp)2. It is now possible to apply post demodulation filters to suppress
sidelobes. This will now be a low pass response T2(oS) (Fig. 13.7D). The Fourier
transform (taking account of both positive and negative frequencies) will give the
resultant demodulated directional pattern which incorporates the effect of the
amplitude taper.
This type of scanning system is essentially a continuous process at a fixed scan-
ning rate; it is not possible to stop the beam or vary the beam dwell time. It is how-
ever possible to produce the effect of a fixed beam by means of a sample gate at
one of the outputs A-D of Fig. 13.7. If this gate opens repetitively at a rate us for
a period of time r/n corresponding to the instant when the scanning beam is look-
ing in the required direction, its output will correspond to a fixed beam and its
angular location may be controlled by the timing of the gate. This type of output
may be useful to generate a single beam or a pair of tracking beams.
An alternative configuration for within pulse scanning of a receiving beam is that
shown in Fig. 13.8. Here, the output of the elements are mixed with a frequency
swept local oscillator and the difference (IF) is fed into a tapped delay line. The
output of the delay line represents the receiving beam output and the angular direc-
tion of the beam (on a sin Q scale) is linearly related to the frequency of the local
oscillator. The two ends of the delay line correspond to beams deflected either side
of boresight but in practice, it is only necessary to employ one such output. Some
configurations have located the tapped delay line in the feeds from the swept local
oscillator to provide the same effect.
Array signal processing 341
This concept may also be regarded as 'frequency scanning at IF' and in contrast
with the continuous scanning scheme of Figs. 13.6 and 13.7, is also capable of vari-
able scanning rates and stationary beam positions. Since the typical phase deviation
required for wide angle scanning is of the order 2?rrad between adjacent elements,
this can be achieved by means of either a wide sweep Af or a long time delay in
each section of the delay line tx. This may be expressed as:
receiving array
Y Y Y (n sections)
Y Y
bandwidth restriction
Fig. 13.8 Frequency-swept local-oscillator system for electronic scanning of a receiving beam
The directional pattern of a radar system is the two-way spatial product of the
transmitting and receiving directional patterns. In most systems, this merely means
Array signal processing 343
squaring the one-way pattern of the antenna. There are many examples where there
is a substantial advantage in employing different patterns on transmission and
reception. For example, the use of different transmitting and receiving patterns in
radar arrays can lead to improvements in either beamwidth or sidelobe perform-
ance. The systems discussed in the previous section represent examples where the
use of a floodlight transmit pattern results in a considerable simplification of the
electronic scanning hardware but at the expense of sidelobe performance. We shall
now look at cases where the radar employs different patterns on transmit and
receive to achieve various advantages.
°ft2 (13.2)
Here, 6b = the beamwidth of the transmitter and 6S = angular width of the sector.
Fig. 13.9 Floodlight illumination of a sector by means of a narrow beam scanned across the
sector in a time t2
The effective duration of the transmitter pulse sent out in any given direction by
this system is set by the effective transmitter beam dwell time in that direction. The
system therefore provides a pulse compression in angle effect which is illustrated in
344 Array signal processing
Fig. 13.9. The within pulse scanning system shown in Fig. 13.6 may be reversed for
operation as a transmitter to achieve this effect, but the transmitting system will
need high power transmitting amplifiers at each antenna element.
This angular pulse compression transmitter may be used with any of the ASP
receiving systems discussed earlier and there is no need for either the beamwidth or
scan rates to be the same between the transmitter and the receiver. The scanning
transmitter is here being used only to achieve a pulse compression effect and the
overall directional pattern of the radar is still the one-way response of the receiving
array despite the use of a directional transmitter. This is because we are effectively
achieving an omnidirectional coverage of the sector from a scanned transmitter.
The above simplified argument does not take account of the sidelobes of the
transmitted signal which will now give rise to range sidelobes in the resultant radar
performance.
13.3.2 Array thinning on both transmit and receive by the use of ASP
Fig. 13.10 shows an interesting array configuration employing a short filled array
on transmission and a wide spaced long array on reception. By choosing the length
of the short array to equal the inter-element spacing of the long array, the resulting
two-way spatial pattern corresponds to a sin«x/sinx pattern. It will be seen from
this Figure that the diffraction maxima of the receiving pattern are cancelled by
multiplication with the directional zeros in the transmitting pattern. If there are nx
elements in the short filled array and n2 elements in the long array, the resulting
pattern corresponds to a one-way pattern of an array containing n\.n2 elements.
Clearly, this corresponds to a substantial economy in elements for a given angular
resolution, but we also need to examine the limitations of such a scheme.
First of all, it is evident that the combined T & R sidelobe performance of this
system is equivalent of a one-way sidelobe performance (in the absence of ampli-
tude tapers). Furthermore, the sidelobe performance on receive only will contain
diffraction sidelobes at the OdB level corresponding to the diffraction maxima.
This may not be a severe disadvantage for radar performance if the dynamic range
of target echoing area is not large, but it is obviously a severe problem for any inter-
fering signals or jamming.
The directional gain of the receiving aperture will be set by its physical receiving
aperture (taking account of tapers and the wide interelement spacing). This is not
the gain associated with the directivity of the two-way pattern. At first, it would
appear that a further disadvantage with the scheme is that the power in the broad
beamwidth transmitter lying outside the angular coverage of the narrow receiving
beim is wasted. However, we can now recover all this power by generating a set of
multiple narrow beams from the long wide-spacing array, so that these beams cover
all the sector illuminated by the transmitting array. This corresponds to ASP within
the sector of the transmitted beam. Any of the ASP techniques discussed in the pre-
vious sections could be used for this purpose.
It is also interesting to note that the short array may now be subjected to elec-
tronic scanning on transmission alone (requiring only nx phase shifters). The
Array signal processing 345
receiving array would then employ n2 elements and incorporate array signal proces-
sing within the transmitted section. This still gives the required two-way capability
corresponding to n1n2 elements and the ability to steer the coverage sector over a
wide angle (typically ± 45°). This is a very economical combination, but the limita-
tion on the technique is the sidelobe performance against interfering signals. This is
an area where a null steering would be very valuable to reduce such interfering sig-
nals. The topic of null steering is discussed later in this chapter.
TX pattern
RX pattern
V V wv
RX TX
Fig. 13.10 Use of a short filled transmitting array plus a receiving array with wide inter-
element spacing to provide a 2-way directional pattern having the form (S\nnp/
Sinp)
Although the above transmit and receive ASP system has been discussed for one
dimension of electronic scanning, the technique lends itself very easily to a two-
dimensional electronic scanning system (i.e. three-dimensional radar coverage). Fig.
13.11 shows transmitting and receiving linear arrays located at right angles in space
and the two-way product corresponds to a pencil beam at the intersection of the
two fan beams. It is then possible to scan this pencil beam in both azimuth and ele-
vation by separate phasing of the transmit and receive arrays.
This arrangement may be thought of as a transmit and receive version of the Mills
Cross array22 used in radio astronomy. Again, there will be transmitted power lost
due to radiation into directions of the main beam not covered by a corresponding
346 Array signal processing
receiving beam and again, this energy can be recovered if the receiving signal
employs array signal processing with either multiple beams or Nyquist rate scan-
ning.
'TX array
Fig. 13.11 Use of crossed transmitting and receiving beams to provide a 2-way pencil-beam
configuration
, m u l t i p l e receiving
-fan beams
Fig. 13.12 Formation of 2-way pencil beams from product of T& R fan beams without loss
of power
Fig 13.12 shows a section through such a set of beams and indicates how the
multiple receiving beams then generate a set of pencil beams which may be steered
in elevation by the phasing of the transmitting array.
Array signal processing 347
frequency in each direction. This is the basis of the Doppler Microwave landing sys-
tem.
• t * *m*
Fig. 13.13 Radar antenna with linear array feed to generate stacked beams in elevation
= (13-4)
The received spectrum is the product of the transmitted spectrum with the square
of the antenna spectrum at the particular angle of interest. The resultant receiving
beams may therefore be defined by a bank of filters (in the case of AR3D by SAW
filters at IF). This radar system provides an overall 2-way sidelobe performance plus
the ability of measuring range and angle of all targets within the sector from a single
transmitted pulse. In addition, pulse compression is applied to all received signals
and the antenna spectrum helps to provide frequency weighting which reduces the
level of the time sidelobes of the compressed pulses.
dispersive
feed
non-linear all-beam-pulse-
•f0c9
FM transmitter compression equaliser
Fig. 13.14 Basic principles of within pulse frequency scanning radar as employed in Plessey
AR3D radar
decorrelated
transmitters
Fig. 13.15 Directionally decorrelated transmitter beams: separate transmitters per feed horn
beam-forming network
/ \ decorrelated
V /transmitters
Fig. 13.16 Directionally decorrelated transmitter beams: separate transmitters per beam port
Fig. 13.17 Directionally decorrelated transmitter beams: separate transmitters per array
element
Array signal processing 351
transmitted waveform is directly associated with a unique beam direction. With
method (b), the transmitted 'code' in a particular direction is formed from the
uniquely phased sum of the element excitations corresponding to that direction. It
is thus necessary to reconstruct this phased sum for the receiver processor: a general
method of doing this is illustrated in Fig. 13.18. The advantages of method (b) are
that it can avoid the need for a high power multiple beam network.
correlation
detectors
beam 1
13.3.6.1 Separate transmitters per beam port: Basic methods of achieving trans-
mitted beam decorrelation are to employ either:
(a) frequency separation of transmitters or
(b) code separation of transmitters operating at the same carrier frequency.
The first method is relatively easy to implement, but since the available frequency
allocation must be shared between the beam ports, the transmission in each beam
on each pulse must be relatively narrow band. (A set of frequencies for each beam
could be obtained by providing more transmitters than beam ports.)
The second method uses the whole of the frequency allocation for each beam
and is thus capable of much finer range resolution. However, the codes chosen must
be very long if they are to achieve adequate decorrelation. For example, if random
phase codes of m bits duration are employed to simulate noise-type transmissions,
and received signals are recovered by tapped-delay line matched filters, the near-in
time sidelobes in the latter will have an expected amplitude of l/y/m relative to the
peak (the peak values of the time sidelobes may be some 10 dB greater). Thus m
must have a value of 104 to achieve 30 dB time sidelobes. In addition, each receiver
channel will also respond to other transmitted beam codes which illuminate the
target via spatial sidelboes: the level of this unwanted signal will be further attenu-
ated by the cross-correlation process, but will become significant if the number of
transmitted beams exceeds the average spatial sidelobe power ratio.
352 Array signal processing
13.3.6.2 Separate transmitters per array element: In this case, transmitted beam
decorrelation can be obtained by:
(a) frequency separation of element excitations
(b) time separation of element excitations or
(c) code separation of element excitations.
Frequency separation using incoherent transmitters produces noise-like waveforms
in each beam direction. The correlation detectors in the receiver (Fig. 13.18) must
store the amplitude and phase record corresponding to each beam direction. This
implies considerable storage capacity. A more economical method is to employ co-
herent frequency separations for the various transmitter modules, as would be
required to produce a pulse compression in angle-type transmitted scan, but with a
pseudo-random arrangement of frequencies along the array.
radiators
n I I
UUQUQU
phase I random
shifters 1 phase-code
generator
random
pulse firing
modulators I sequence
generator
oscillators
low-power
locking
signal
Fig. 13.19 Phase-coded time-division-multiplex transmissions to produce directionally de-
correlated transmitted signals
systems can incorporate printed circuit antenna arrays. For ASP systems where the
signal frequency channels are at microwave frequencies, most of the microwave
components such as hybrids, circulators and phase shifters can be constructed using
printed circuit technology, which is now in widespread use. Multiple beam matrix
networks such as the Butler matrix or the Maxson matrix can be constructed at
microwave frequencies in stripline and some Butler matrix systems are commer-
cially available in this form. The absence of high power requirements can make
these components comparatively cheap. An alternative approach is the lens fed
array where a lens, such as the Rotman lens can be used to provide a multiple beam
capability.18 These lens systems may be designed in waveguide or stripline con-
figurations.
The continuing improvements in the performance of solid state microwave
amplifiers is encouraging and it is therefore possible to incorporate limited amounts
of microwave amplification in each channel with good phase and gain stability. This
reduces the need for specialised phase control loops for all but the most stringent
sidelobe requirements.
133.8.3 Digital and base band beam forming: The concept of digital beam
forming first grew up in sonar systems though it had been used to a limited extent
in radio astronomy. Both these applications are associated with fairly low RF band-
widths, but modern developments in wide band digital processing have extended
the capability to radar. Reference 20 describes the use of digital processing in sonar
systems and Reference 21 is an example of digital beam forming for radar.
The basic concept of digital beam forming first requires frequency changing the
received signal down to a low IF. In general, this is a zero IF (though modified by
Doppler shifts) and the signal is then fed to analogue-to-digital convertors to pro-
duce two streams of binary baseband signals representing 'in-phase' (I) and 'quadra-
ture' (Q) channels. These sets of digital signals then represent the amplitude and
phase of signals received on each element of the array. The process of beam form-
ing implies weighting these digital signals with the appropriate complex values of
amplitude and phase and summing the resultant to form a beam. The resultant sig-
nal may, of course, be converted back via a digital-to-analogue convertor to provide
a beam output. Note that this process is linear and does not involve demodulation
of the signals which still retain phase information.
There are many possible configurations for performing the digital processing. In
the simplest case, dedicated multipliers are used for each channel to multiply the
element signals by the appropriate beam steering weightings. These instructions can
be driven by software and can produce scanning beams, multiple beams or null
steering, as will be discussed later in the chapter.
It is interesting to note that the flow chart of a discrete fast Fourier transform
programme is exactly the same format as the Butler matrix network which evolved
several years before fast Fourier transform programmes became available. The
digital FFT generates a set of orthogonal beams giving the usual —13 dB sidelobes
from an unweighted aperture. Dedicated FFT machines are becoming available for
various functions and can therefore be applied directly to beam forming at base
band. Since digital processing is of necessity performed after amplification of the
signals, the relevance of orthogonal beams is far less important.
Many different configurations are thus available for scanning beams or generat-
ing multiple beams and this can be done with dedicated special purpose hardware
or in general purpose computers, though the latter are rarely particularly efficient
for this purpose. Sampling rates of about 20 MHz are currently available though this
does tend to limit this form of processing to low or medium range resolution radars.
The number of bits employed in the processing systems and the design of the soft-
ware also places limitations on the dynamic range of the signals and on the resulting
sidelobe levels of the beams so formed. Although the concept has been described
here for reception, it is clearly possible to use digital processing to generate suitable
IF signals which could represent array excitations for any beam direction. They
could thus be amplified for electronic scanning on transmission if required.
Array signal processing 357
Multiplicative signal processing has been used extensively in radio astronomy but
has had only limited application in radar. In radio astronomy it has been used for
three different types of application:
(a) To achieve 2-element interferometer arrays.
(b) To synthesise the pencil beam of a square aperture by multiplying the output of
two crossed linear arrays22 (a special form of interferometer called a Mills Cross
array).
(c) For aperture synthesis by post demodulation processing;2 here the product of
signals from pairs of elements located at different positions on the ground can be
added together to synthesise the output corresponding to a large filled aperture.
In both radar and sonar, multiplicative arrays have been used to achieve small
reductions in beamwidth for various receiving arrays. This technique had also been
adopted in one form of harbour radar.23 A feature of multiplicative arrays with
potential for exploitation in radar systems is the ability to produce normal beam-
width patterns from thinned arrays containing very few antenna elements.
13.4.1 Interferometers
One of the early uses of 2-element interferometer arrays was in radio astronomy
where the need for increased angular accuracy and reduced beamwidth requires
antennas having very large dimensions. This problem was solved by the use of two
antennas having a large spacing between them. If the outputs of two such identical
antennas are combined and fed to a receiver incorporating a detector, the resulting
direction pattern is given by:
•De(9)
|D e (9)|
e(6) 1(0) |
Fig. 13.20 Two-element interferometer array and its directional pattern with modulus detec-
tion
The above concept depends upon the fact that there is only one source within the
broad beamwidth of the element patterns, otherwise the interferometer response
will be a complex signal which is difficult to interpret.
When any directional receiver is scanned across a far field distribution of signal
sources 5(0), the output of the directional receiver is the convolution of the direc-
tional pattern of the antenna with the angular distribution function 5(0).
D(6)*S(0) (13.7)
This situation applies for either single or extended sources and in the latter case can
be used to measure the angular width of the radio source.
Array signal processing 359
Fig. 13.21 Two-element multiplicative interferometer array and its multiplicative (bipolar)
directional pattern
The 2co term may be filtered out leaving the demodulated signal having a directional
pattern of the form:
typical multiplicative
array configuration
wide-aperture
response
Da(6)
short-aperture
response
D b (6)
overall multiplicative
directional pattern
interferometer
function
It can be seen from Fig. 13.23 that the multiplicative pattern produces a reduced
beamwidth but a higher sidelobe level than the additive pattern. However, we have
not taken into account the square law response of the characteristic. Clearly there is
no directional advantage in employing square law detection compared with
modulus detection; indeed, if the first two characteristics were plotted on a true dB
scale (in terms of the variation of the input signal with direction necessary to main-
tain a constant output from the array) they would produce identical responses.
This method of plotting directional responses in dB may therefore be adopted in
order to remove the effect of the detector law (or demodulation law) on the result-
ant directional patterns. Under these conditions the first sidelobe level of the two
additive patterns will be — 13dB, but that of the multiplicative pattern will be
about —7 dB.
When a pattern is subjected to square law detection the number of zeros in the
demodulated response is effectively doubled, due to each zero becoming a double
coincident zero. Multiplicative processing also doubles the number of zeros in the
resultant pattern but they need no longer be coincident zeros and this gives some
increase of flexibility in synthesis of multiplicative patterns. In general, the use of
multiplicative processing only produces reductions of beamwidth at the expense of
some increase of sidelobe level when the effect of the square law response is taken
into account. Claims for improved resolution due to multiplicative processing
should also be viewed in terms of the overall effect of the processing on sidelobes.
Some of the literature has tended to regard multiplicative arrays as fundamen-
tally different from conventional arrays. However, if we make comparisons between
demodulated additive arrays and multiplicative schemes, there is no fundamental
difference. Both systems produce various combinations of self-products and cross
products of signals from the elements.
The problem of pattern synthesis can be handled by applying conventional syn-
thesis techniques to the two sub arrays. The resultant pattern is then the product
Array signal processing 363
of these two patterns and the interferometer directional term. One of the most con-
venient ways of designing any demodulated directional pattern is to analyse it in
terms of spatial spectra.
sinxi 2 detected
x J nresponse
0
spatial frequency spectrum
imimiiimiiiii
square-law
detector
Fig. 13.24 Square-law detected linear array and its directional and spatial frequency respon-
ses
364 Array signal processing
IT ? T 1 MM ?{• M
Fig. 13.25 A 5 x 5 thinned multiplicative array with the constituent directional pattern and
spatial frequency spectrums
as discussed in Section 13.4.3 and shows a theoretical low sidelobe pattern which
simulates the response of such a thinned array using only twenty eight elements to
synthesise a seventy four element array.
The second disadvantage of the above scheme is the complex response of multi-
plicative arrays when illuminated by multiple sources. This will be discussed in
more detail in the next section.
Array signal processing 365
The spatial frequency concept shows how the response of an array is made up
from constituent components each related to the inter-element spacing of pairs of
elements. Since the output from such an element pair spaced say d apart corresponds
to the same output, irrespective of where the pair occurs within the array, we can
argue that each element spacing should not be duplicated as this corresponds to a
waste of elements. In terms of the single source response of an array this is quite
true. An array which does not duplicate its inter-element spacing (or which has only
one spatial component for each spatial frequency line in its spectrum) is known as a
minimum redundancy array. Although many minimum redundancy arrays are also
multiplicative arrays, one simple form of such an array26 is shown in Fig. 13.27 and
corresponds to a square law detected additive array. This array consists of a
7-element linear array in which three elements are missing. The missing elements are
so chosen that the resulting combination contains all the possible inter-element
spacings of the full 7-element array. This means that its spatial frequency spectrum
must be uniform (with the exception of the DC self product term.) If this
4-element array is square law detected, the resultant directional pattern corresponds
to the sin 7p/sin p pattern of the full 7-element array plus the DC term due to the
four self product terms. This pattern is shown in Fig. 13.28 for both linear rectifi-
cation and square law detection.
Fig. 13.26 Synthesis of a 74-element array, 30dB sidelobe response using a 12 x 16 multipli-
cative array
O O •+• • O • O
Fig. 13.27 ARSAC array: a 4-element minimum redundancy array containing all the inter -
element spacings of a 7-element array
1.00r
.- 0.75
2 0.50
0l—
-1.0 -0.5 0 0.5 1.0
sin 6
1.00r
0.75
£ 0.50
I
^0.25
angle and time. The multiple source cross products are reduced, and hence the reso-
lution properties of the receiver are enhanced by any lack of temporal coherence of
the wanted signals. In some situations this can occur naturally due to the ionos-
phere or multiple reflection fading.
A detailed discussion of the noise performance of multiplicative arrays and cor-
relation detection systems is outside the scope of this book. It has received atten-
tion from many authors.30'31 Correlation receivers are used extensively for passive
detection systems in several applications including radar and sonar.
If we consider the simple case of a uniformly phased multiplicative array it is
reasonable to ask whether the multiplicative signal processing affects the detection
process. This corresponds to comparing the multiplicative demodulation process
with other forms of demodulation. A simplified picture can be obtained from the
output SNR of a multiplier in terms of the SNR at its two inputs. Tucker has pro-
vided an expression for this in the form:30
2/? R
R ( 1 3 1 6 )
where R x and R2 are the SNRs of the two input signals. This may be compared with
a combined SNR of (Rt + R2) for optimum combination of a linear additive array.
A limitation of this approach is that the amplitude probability distribution of
the noise outputs are not identical to that of a simple detector.31 The probability
distribution is an exponential and Bessel function dependent upon the correlation
of the two inputs to the multiplier. Nevertheless calculations on probability of
detection performance reveal a loss of about 1-1.5 dB compared with conventional
detection. This small loss is therefore associated with the absence of the self
product terms at the output of the directional reciever. An important consequence
of the above equation is that the output SNR is a maximum when the two input
SNRs are equal. This means that the two sub arrays feeding the multiplier should
have equal gains.
In cases where thinned multiplicative arrays are used, the basic SNR at the two
inputs will be reduced by the thinning factor of the two sub arrays.
d, d2 d3
QJ-Q--Q Q Q O
Fig. 13.29 Two multiplicative arrays having identical directional responses despite the differ-
ent array geometry
For the case of radio astronomy, the far field distribution of sources is fixed so
it is not even necessary to take the output signals from all the elements at the same
period of time. This concept led to the development of synthetic aperture antennas
for radio astronomy by Ryle.2
The linear array synthetic aperture is shown in Fig. 13.30. It is necessary to have
two elements and one multiplier and to move the elements to different locations
along the line and to record the bipolar, low pass filtered output of the multiplier
for all the necessary interelement spacings corresponding to a filled array. These
recorded output signals can then be summed directly to form the signals corre-
sponding to a synthesised beam. If the output of the multipliers were also recorded
for in phase and quadrature phase of the two elements, it would also be possible to
synthesise multiple beams pointing in many different directions.
370 A rra y signal processing
Since the outputs from these multipliers are, in general, broad bandwidth signals,
they may be integrated in order to increase the signal to noise ratio. This provides
a narrow band signal which may be conveniently stored on magnetic or paper tape.
These recorded signals may then be fed into a computer to produce various forms
of synthesised beams at a later date.
positions for
fixed moveable element
element \
Y
I
This is the basis of the post demodulation form of aperture synthesis used in
radio astronomy. The term post demodulation is relevant because it is the demodu-
lated output of the multipliers which are stored and then used to synthesise patterns.
Although the above discussion has been in terms of a linear array, the technique is
quite general and could apply to a planar or even a volumetric array.
If et represents the output of the zth element of an ^-element array, the square
law detected output of the whole array Eo is given by:
2
En = (13.17)
£=1
This expression shows that the detected output can be synthesised by a series of
multiplicative products between the zth and gth elements. This also indicates that
this multiplicative system has an exact equivalence with the additive square law
detected array. This applies even to multiple target performance if the array is syn-
thesised to include all repeated element products.
To perform two dimensional aperture synthesis, the two elements must be loca-
ted at appropriate points in a plane, i.e. (x\9y.i) for element 1 and (x2,y2) for
element 2. The locations within this plane are then chosen such that x{ — x2 and
J i ~ y i cover all required interelement spacings in both dimensions. In many cases,
it is more convenient to employ more than two elements in order to reduce the
total amount of time required to synthesise patterns. Various combinations of
moveable elements and a larger fixed array may also be adopted.
Although the basic concept of synthetic aperture appears to lead to a very low
gain performance, the important parameter is the time taken to collect sufficient
energy to achieve an adequate signal to noise ratio for each direction in space. Here,
a synthetic aperture system scores on the grounds that it can also simultaneously
synthesise multiple beams in different directions, which therefore speeds up the
Array signal processing 371
overall process. The main advantage of the synthetic aperture antenna is its substan-
tial reduction in cost and its ability to synthesise antennas whose dimensions would
be quite impractical by means of a filled aperture technique.
An interesting application of post demodulation synthetic aperture techniques
to radar has been described by Shearman.33 This is an HF radar employing flood-
light transmissions covering a broad sector of interest and is designed to study the
ionosphere at long range.
To achieve adequate angular resolution at these frequencies requires a long array
and consequent expense, but the movement of the targets is quite slow so that
there is ample time to build up a picture. The receiving array consists of a 2-element
multiplicative interferometer having one fixed element and the other moveable (on
rails). The product of these two element outputs is therefore recorded for-various
positions of the moveable element and group of narrow beams are synthesised
covering the sector illuminated by the floodlight transmitter. Again, since the
recorded signals are not dependent upon the phase of the received signals, there is
no need to employ coherent transmissions. An important restriction on the tech-
nique is that if the target field is complex (as it is for ionospheric targets), there will
be a significant distortion of the results due to multiple target cross product effects.
A particularly interesting result of measurements made by Shearman on his radar is
that the effects of such cross products could be substantially reduced by the appli-
cation of frequency agility to the transmissions.34 This indicates that the use of
frequency agility should be able to reduce such cross product effects when mini-
mum redundancy arrays are employed in a radar system.
13.5.1 Principle
The technique of coherently synthesising an antenna aperture is primarily associa-
ted with airborne or satellite-borne mapping radars, as illustrated in Fig. 13.31. The
basic principle is shown in Fig. 13.32. A pulsed radar transmitter feeds a compara-
tively small antenna aperture d, which is transported along a straight line. Received
signals are range-gated and, for each range interval, the amplitude and phase history
is stored. The processor associated with each range interval performs a continuous
weighted summation of signals over a distance D equal to the length of aperture
which it is desired to synthesise. Thus the real aperture d essentially forms the ele-
ment pattern for the synthesised aperture D.
The system differs from the post-demodulation synthetic aperture antennas dis-
cussed previously in three major respects:
(a) The transmitter provides the phase reference, so that absolute phase informa-
tion is stored, rather than relative phase between a pair of elements. The coherent
synthetic aperture antenna is thus only suitable for detecting stationary, non-
scintillating targets.
372 Array signal processing
V = 100 m/s
V = 7500 m/s
, =880 km
Ro =840 km
h = 800 km
(b) Since 2-way phase is involved, angular discrimination is generally twice as good
as that of a post-demodulation synthesised aperture.
(c) Targets of interest are invariably within the near-field range of the maximum
aperture which can be synthesised.
processed output
E(x)
Assuming that the radiation pattern of the real aperture d can be approximated by
a flat-topped beam of angular width \/d, then it is evident from the geometry of
Fig. 13.32 that the maximum length of aperture which can be synthesised as pro-
portional to range, specifically:
(13.18)
374 Array signal processing
For the airborne radar example of Fig. 13.31a, typically d = Im and X = 0.03 m,
and Do becomes 200m at Ro= 10km. The far-field range of such a synthesised
aperture, according to the usual D%fk criterion, is 3000 km. For the satellite
example of Fig. 13.315, typically d = 10m and X = 0.25 m, giving Do = 22km at
880 km range. The far-field range in this case is about 2 x 106km.
When the aperture d has moved a distance x along the track, the additional path
to the target is
8 = [R2 + x2]QS-R » x2/2R (13.19)
and hence the phase variation is
0 = - 4 T T S / X ** -2TTX2/RX (13.20)
where W(y) represents the amplitude and phase weighting applied in the processor.
Essentially, the processor associated with each range interval performs a correlation
between the range-gated received signals f(x) and the stored weighting function
W(y).
The principal parameters under the designer's control are the real aperture length
d, the synthesised aperture length/) and the processor weighting function W(y).
over the synthesised aperture. In this case, it is evident from Fig. 13.33 that it is not
worth while choosing D to be much longer than L, since the signal contributions
outside this interval are largely self-cancelling due to the rapid phase variations.
phase
TTRA
=L
distance along track
(a)
phase
=L
distance along track •
Fig. 13.33 Phase history of target (b)
a Target at short range
b Target at longer range
For the simplest case of constant amplitude and phase weighting, then
(13.25)
and the processor output, using eqns. 13.22 and 13.24, becomes:
which reduces to
376 Array signal processing
E(x) =
2
(fovD-D0l2<x<D0/2-D) (13.27)
where
F(u) = C(u)-jS(u) (13.28)
and C(u) and S(u) and standard Fresnel integrals defined by:
a/— D/2 U-
"U
B i 1
mli D=T/A R
D
/AR 3/AR
A
\
%
B
/ \ D-
pli
/XR
/ V.
E
o
D=2/XR
-/AR
track
(D ft /D)* (
track
(b)
Fig. 13.35 Focal region in a focused synthetic aperture antenna
a Geometry
b Focal region
Array signal processing 379
where 77, £ are the along-track and across-track co-ordinates of Tx from To. Setting
6 = A/8, corresponding to n/2 2-way phase error, gives the maximum values of dis-
placement as
7? = ±(XR/4D) = ±(dl4)(D0/D) (13.35)
and 2 2 2 2
% = ±(XR /D ) = ±(d IX)(DolD) (13.36)
The focal region is roughly elliptical, as indicated in Fig. 13.35Z?. For an optimised
focused synthetic aperture D =D0 and the along-track resolution is simply d/2, i.e.
half the aperture of the real antenna, whilst the across-track resolution is 2d2/X,
which is just twice the Rayleigh range of the real aperture. In practice, the across-
track resolution is enhanced by employing short pulses and range-gating.
The apparent anomaly that the along-track resolution is directly proportional to
d, rather than inversely proportional to d as in a conventional antenna, is resolved
when it is recognised that choosing d small enables a larger value of synthetic aper-
ture d to be realised. The penalty paid for increasing the resolution is that more
processing power is required.
The waveform of the processed output may be obtained by substituting eqns.
13.32 and 13.22 into eqn. 13.24. For the case of uniform amplitude weighting and
an optimised system (i.e. D = Do), the processed output becomes:
This can be expressed more compactly by shifting the origin to the position of peak
response by writing:
X = x-D0/2 (13.40)
380 Array signal processing
Then, noting that DQ = (RX/d), the final expression for the output from the pro-
cessor becomes:
(D<X<D) (13.41)
-Do/2 3Do/2
distance along track-•— x
Do/2)
distance along track —«*•• x
b
Fig. 13.36 Processed output from focused synthetic aperture
a Optimised {D = Do)
b Non-optimised (D < DQ)
Illustrated in Fig. 13.33 from a finite number of samples, the maximum phase shift
between samples should not exceed n. The maximum phase slope occurs at the
edges of the beam from the real aperture, and at these points the along-track sampl-
ing interval s corresponding to a phase shift of n is d/2. In practice, a shorter
sampling interval may be needed, to allow for the fact that the beam from the real
aperture d extends beyond the assumed beamwidth of \/d.
Even when the synthesised aperture D is less than the maximum possible value
of D o , it is still necessary to sample at intervals of d/2 to prevent aliasing effects
(e.g. a short section of the parabolic phase curve at the edges of the illuminated sec-
tor could appear to have a reduced phase slope if it is under-sampled).
It will be apparent that the sampling interval s determines the angular positions
of grating lobes from the synthetic aperture. As shown in Fig. 13.37, grating lobes
occur at intervals of approximately X/2s, and it is clearly desirable that the beam
from the real aperture d should have low side lobes at these angles.
sin 8 = A /2s
h-.-l
sampling
interval
Fig. 13.37 Grating lobes in a coherent synthesised aperture
The minimum number of samples for the processor for each range interval is
M = 2D/d = (2R\/d2)(D/D0) (13.42)
The number of range gates is
N = {Rx~R0)jbR (13.43)
where bR is the range gate width and Ro and Rx are the minimum and maximum
instrumented ranges. The minimum total number of samples or pixels in the proces-
sors of an optimised focused synthetic aperture radar is therefore
382 Array signal processing
P = (K/d^iRl-RDldR (13.44)
For the airborne radar example of Fig. 13.31a, where, say,/?! = 10km,i? 0 = lkm,
d = lm, dR = lm and X = 0.03 m, then P is 3 x 106. For the satellite example of
Fig. 13.316, typically Rx = 880km, Ro = 840km, d= 10m, 8R = 10m and
X = 0.25 m. The number of pixels in the processor is then about 2 x 107.
Although the along-track resolution is independent of wavelength, choosing a
short wavelength helps minimise the amount of processing power required.
Pr = ^ ^ (13.51)
' (47T)3i?4
where Pt is the peak transmitter power, G is the gain of the real antenna aperture d,
and a is the target echoing area. The noise power is
Pn = kTBF (13.52)
where k is Boltzmann's constant, T the receiver temperature, B the bandwidth and
F the noise factor. B is generally approximately 1/r where r is the pulse length.
The vertical beamwidth of the antenna is fixed by the coverage requirements,
and hence the gain G is proportional to d/X. The single-pulse signal-to-noise ratio is
thus independent of the choice of wavelength, but is proportional to d2.
The quality of the processed image may be judged by its resolution, its ability
to reject noise and its ability to distinguish wanted targets from the background sig-
nals contributed by other targets. The resolution, of course, is determined by d and
r.
All samples are integrated coherently; hence:
S/Nimprovement factor = D/s = (RX/sd)(D/D0) (13.53)
less any loss due to amplitude taper introduced to improve the sidelobes (which is
directly analogous to antenna efficiency). For an optimised focused synthetic
384 Array signal processing
aperture, sampled at the minimum PRF, eqn. 13.53 becomes:
S/Nimprovement factor = 2D0/d = 2RX/d2 (13.54)
which is 28 dB and 36 dB at the maximum ranges of the airborne and satellite-borne
examples quoted previously. The effective S/N at the output of the processor
becomes independent of d, but is directly proportional to wavelength and inversely
proportional to R$. Reducing the sampling interval s by increasing the PRF above
the minimum value will improve the processed S/N, but only at the expense of
more processing power.
The contrast between a wanted target and the residual signals from other targets
within the processor depends upon the number of targets being processed simultan-
eously and the sidelobe level achieved in the processor. If S represents the average-
power sidelobe level relative to the peak response, then in the case of an optimised
focused system where the wanted signal duration is d/2 and each resolution cell
produces residual signals over an interval of 2D0-
Contrast factor = d/(4SD0) (13.55)
Thus the contrast factor is essentially half the peak signal-to-mean-sidelobe ratio,
degraded by the S/N improvement factor. The need for low sidelobes from the pro-
cessor is quite apparent. For an under-optimised focused system:
Contrast factor = (dD0)l[2SD(D0+D)] (13.56)
Reducing the sampling interval below d/2 will not affect the contrast factor, since
both wanted and unwanted signals are correlated to the same degree within this
interval.
an additional linear variation of phase with time (i.e. a fixed Dopper offset) within
each cell. This generally results in along-track translation errors, but the target may
be blind if the phase shift between samples is exactly rt. For moving vehicles, there
is further spectrum broadening due to aspect dependency, vibration and multipath
effects.
13.5.10 Applications
The two principal application areas are sketched in Fig. 13.31. It will be evident
that deviations of the flight path from the desired straight line must be assessed
and compensating phase adjustments made to the processors. The compensation is
range-dependent, because the depression angle to the target varies with slant range.
Additionally, the real antenna's beam must be stabilised against vehicle pitch and
yaw, unless its beamwidth is very broad. The residual errors caused by these effects
can be separated into random and periodic errors and treated by the usual methods
for assessing the effects of antenna illumination errors.38 Additional phase errors
will arise if there are any spatial or temporal variations in the refractive index of the
propagating medium.39
Airborne mapping, as in Fig. 13.31a, was initially developed for military surveill-
ance, but is finding increasing scientific applications, particularly in remote sensing
of the terrestial environment.40
A number of satellite-borne synthetic aperture radars are currently operating for
both ground and oceanographic surveillance.41 They usually orbit at a compara-
tively low altitide (c. 800 km) and their processors require additional compensation
for the effects of earth rotation.
Since the basic resolution is independent of wavelength, synthetic aperture
radars can be built at VHF. Such low-frequency systems are valuable in remote
sounding applications in glacial ice42 and dry soil,43 as deep penetration can be
achieved due to the low absorption coefficients at these frequencies.
13.6.1 Introduction
Communications and radar antenna systems are susceptible to degradation in per-
formance because of interference received through their sidelobes. The interference
may be deliberate electronic counter measures (ECM), accidental RF interference
(RFI), clutter returns or natural noise sources.
Typical interference signals received by RF systems are from point sources
radiating a coherent wavefront for at least the signal gathering time of the antenna.
The antenna operating in its 'far field' will receive a plane coherent wavefront. In
order to prevent interference of this type entering the receiver it is necessary to
arrange for a null in the antenna's polar response to coincide with its angle of
incidence.
The existence of a plane coherent wavefront can be identified through the use of
386 Array signal processing
an array of receiving elements. Once identified, the angle of incidence of the signal
can be readily determined. The directional pattern of an iV-element linear array may
possess up to (JV— 1) independent zeros. There are obvious advantages in having
the ability to independently steer the zeros in the array's directional pattern in
order to reduce interference signals normally received by the sidelobes.
In order to independently control the angular location of the zeros it is
necessary to control both the phase and amplitude of signals in the feed system of
each element. The requirement for accurate control of both the phase and ampli-
tude of signals received by each element can in practice be difficult to achieve, and
often limits the use of such systems to experimental and military applications.
Before discussing the properties of zeros further it is worth defining the differ-
ence between minima (commonly called nulls) and zeros. The mathematical analysis
of an ideal array will define angles at which the response is zero. All incoming
signals at those particular angles of incidence and at a single defined frequency will
not produce any output. However, because the formation of zeros relies on many
signals cancelling exactly in the combining network, they are extremely sensitive
to aberrations in the array, and to the bandwidth of the signals. Even if it were
possible to manufacture a perfect antenna, it is likely that mutual coupling effects
would disturb the signal phase and amplitude characteristics. Therefore in practice
zeros are not perfectly formed. They will not have infinite attenuation at one
angle, but a rather less sharp response spread over a range of angles. Mathematical
zeros therefore tend to produce nulls in the actual polar response of an antenna.
=I k=
where *
z = exp (j(2ird/X) sin 6) (d = element spacing)
and the phase reference is taken as the first element in the array. The coefficients of
the polynomial may be complex, thus representing the amplitude and phase weight-
ing of each array element. Thus:
Wk = Akexp(j<Pk) (13.58)
The polynomial may be factorised into (N — 1) roots and each of these represent a
zero of the polynomial
D(d)= WN(z-b1)(z-b2)(z-b3)...(z-bN-1) (13.59)
In general, these roots are complex numbers and may therefore be represented as
points in the complex z plane. Since z is simply a phasor, \z\ = 1; so roots located
Array signal processing 387
on the unit circle \z\ = 1 represent true zeros of the directional pattern (i.e.
This Schelkunoff analysis also shows44 that if d > A/2 the zeros of the pattern
function may repeat over the 180° of real angles and if d < A/2 there may be zeros
of the pattern outside the range of real angles. Although the above indicates the
maximum number of independent zeros to be (N— 1) there is no reason why a pat-
tern may have no true zeros if all the roots are located off the unit circle. Mutual
coupling and tolerances of excitation can give errors in the coefficients of the poly-
nomial for a real array. This displaces the zeros off the unit circle and hence softens
the otherwise sharp nulls in the pattern.
w2 W
3
Fig. 13.38
The relationship between the zeros, the radiation pattern and the element
weights W is sometimes difficult to visualise. Consider the pattern of the uniformly
illuminated array of Fig. 13.39 and assume that there is a jammer in the sidelobe
shown. It would be sensible to move the zero closest to the jammer on to the
jammer bearing. In z-plane terms (eqn. 13.59) this is easy. The root bp correspond-
ing to the closest zero is simply replaced by Bj, where
. 2nd
exp (/-—sin0;- (13.60)
A
and 0j is the jammer bearing. The modified radiation pattern, shown in the Figure,
can be expressed as
Dj(6) = (13.61)
388 Array signal processing
and it is clear that the disturbance of the pattern is mainly in the region of the jam-
mer. However, the effect upon the element weights W is quite complicated and it
will be appreciated that every weight must be modified in order to achieve the new
pattern. The situation becomes more complicated if there is more than one jammer
to be dealt with.
•i
o
<L
-30
-180
Fig. 13.39 Effect of moving one zero on the pattern of an 8-element uniformly illuminated
array
= cos (13.62)
There are three outputs with this directional pattern. These are then combined
together again in adjacent pairs with a phase shifter 0 2 to steer the second zero, and
the process repeated for the third. For an TV-element array, the overall pattern is the
product of (TV' — 1) such terms:
and it will be appreciated that all the zeros are real zeros on the unit circle.
Fig. 13.41 shows a sidelobe region of a directional pattern with zeros at Un and
£/(n+1). If the two zeros are made coincident a rather broader second-order zero
Array signal processing 389
results where the pattern touches the axis. This can be extended to third order and
higher order zeros as shown in Fig. 13.42.
Multiple zeros deepen and broaden the null, and as a consequence reduce the
level of the lobe between nulls. In cases where interfering signals are received by the
array
output
F ig. 13.40 Null Steering Tree
H = Hybrid (4th port terminated)
array from specific directions, multiple zeros are useful in preventing the unwanted
signals entering the receiver. However, it is not always advantageous to overlay nulls
on existing nulls. Superior performance can often be achieved by locating zeros
between existing ones, especially when signals with a wide bandwidth are being
received.
Although the above concept can, in principle, synthesise any directional pattern
by the choice of the location of the zeros (which in turn defines the roots of the
390 Array signal processing
polynomial) the system is not loss free. This is due to the power lost in the resist-
ive loads. The system also has the disadvantage that movement of zeros can also
result in changes in the gain of the resultant pattern in a 'specified main beam'
tion. This may be overcome by extending the concept discussed in reference 46
which uses the weighted combination of a sine and cosine directional response to
steer a zero without changing gain on axis. A modified form of null-steering tree
based upon this concept is shown in Fig. 13.43 for a 3-element array.
D(6)
Fig. 13.41
1st order
D(6)
2nd order
•3rd order
Fig. 13.42
For the arrays shown in Fig. 13.43 the nulls are steered by the variable gain
amplifiers giving a null for row m in direction dm due to the amplifier gain Rm:
output
power patterns
H-i „ J A
spatial
filter
formed by
beam 1
spatial filter
formed by
beam land 2
spatial filter
formed by
beams 1,2 and 3 K/VJ _
F ig. 13.44 Orthogonal beam network
I Bexp(jv b )
phase shifter
B exp j (y*0)
A exp(jy a ) | I
I
3dB hybrid
variable coupler
ganged j 7 2*c (power coupling = sin ex)
phase shifters
3dB hybrid
!
i i C = cos o((A exp v a )
I +sinaf(Bexp \y+<t)
13.38. A scheme which will allow a single zero to be steered in an array of any
number of elements is shown in Fig. 13.48. The single zero is controlled by combin-
ing signals from the two end elements. As in the multiple null steering array, the
beami
(-90°)
beam 2
(45*)
beam 3
020°)
beam 4
r\
11! \; w \..\ .
V
-180
Fig. 13.46 Network constants and radiation patterns for maximum gain beams at —90°, 45°
and 120°
control-amplifier gain is given by eqn. 13.64, the spacing d being the interelement
spacing. The position of the zero is subject to the same restrictions as in the mul-
tiple zero steering array, and its proximity to the broadside position is set by the
maximum allowable value for R. As a zero is steered close to the broadside, the
Array signal processing 395
position of the main beam changes. The main array response and the weighted end
element response of a 10-element array (constructed as shown in Fig. 13.48) is
shown in Fig. 13.49. This gives a zero at — TT/6 radians from the broadside position.
1075*
beam 2
(45°)
beam 3
-90°)
-180* -90°
2nd sin 0
Fig. 13.47 Network constant and radiation patterns for maximum gain beams at 120°, 4$°
and -90°
V V V
It is therefore necessary to vary both A x and 0j in Fig. 13.50 to steer this null. In
order to steer such a zero in the sidelobe region, it is necessary to subtract the
weighted omnisignal from the output of the array. This can also lead to a small
reduction of gain of the main antenna. In order to steer a null in the main lobe
region it is clearly necessary to attenuate the signal from the main antenna (or to
amplify the signal from the omni) prior to combination. This leads to a significant
loss of gain (or SNR of the received signal).
An alternative way of achieving sidelobe null steering in radar systems is to
detect the signals from the two antennas and then to subtract the demodulated sig-
nals.50 This is a form of post demodulation processing and it is necessary to take
care to avoid multiple signal (or multiple target) cross product effects. A simple
example of the above concept is merely to ensure that the detected response of the
omni element is larger than the largest sidelobe of the directional antenna. The sub-
tracted pattern thus has sidelobes, all of negative polarity. This has been termed
'sidelobe blanking'.
With this system a single signal received from the sidelobe region gives a negative
polarity output - which can be ignored (or removed with a diode). The perform-
ance in a multiple target situation, however, is unattractive since the negative signal
can still mask a wanted main beam return (which gives a positive DC polarity out-
put). Such systems are used in radar for the removal of impulsive interference.
Array signal processing 397
13.6.7 Discussion
The previous section has indicated how the zeros of various arrays may be steered
by combinations of amplitude and phase control. Although the discussion has cen-
tred on linear arrays, it may clearly be extended to planar or even volume arrays.
The general result of (N— 1) zeros applies to TV elements in any 3-dimensional
configuration. The particular case of circular arrays is dealt with in Chapter 12.
Although many of the networks discussed (such as the Davies tree) have indi-
vidual control parameters for each zero, the array excitation can, in principle, be
398 Array signal processing
directional
antenna
_omnidirectional
"element
amplitude and
phase control
Fig. 13.50
In many practical arrays the presence of mutual coupling causes departures from
the desired directional response. Although this can be significant for many beam
patterns, it is critical for null patterns where the null depth is dependent upon exact
cancellation of signals.52 An exception to this situation is the field of circular arrays
where the symmetry can permit the compensation of mutual effects.53 Circular
arrays have proved attractive for steering nulls in omnidirectional type responses for
communication applications and this approach is mentioned in Chapter 12.
In addition to the null depth sensitivity to excitation errors, it is also necessary
to consider the effect of changes of frequency or bandwidth. Certain phase control
devices (such as the directional coupler) offer phase shifts which are independent
of frequency. This would result in a pattern in which the null directions changed
with frequency. Replacing the phase control devices with variable time delay
components will achieve directional responses whose nulls remain fixed in direction
with change of frequency.
Generally many null steering applications are limited to systems where it is pos-
sible to derive a feedback signal to verify the locations of the nulls. Such a system
may use an operator to apply the necessary corrections or a processor designed to
automatically steer nulls. Such systems are termed adaptive and are described in the
next section.
Array signal processing 399
13.7.1 Introduction
A major disadvantage of the open-loop null steering arrays described above is that
the angular location and depth of the nulls is extremely sensitive to phase and
amplitude errors in the power splitting networks. In order to avoid this limitation,
an automatic control system is required which steers zeros in desired directions and
which compensates for any errrors due to tolerances in the array construction or
due to mutual coupling effects.
Data on the desired null directions is invariably obtained by processing some
form of measured co-variance matrix which defines the mean values of the cross-
products of the voltages received by the elements of the system. In the case of a
2-element system, the co-variance matrix M is:
*- [ 1 1 1
where the asterisk stands for the complex conjugate and an overbar indicates a time
averaged value, and Vu V2 = voltage received by element number 1, 2, respectively.
For two isotropic elements illuminated by a single narrow-band jammer at angle
d, the co-variance matrix reduces to
(Pt + Qn)
Kj Mn) , _ . , |
L P,exp(--#)
( l 3 6 6 )
where
Pj = mean jammer power at each element
Qn - receiver noise at each element (assumed uncorrelated)
0 = (27r<isin0)/A
In this case it is evident that the angle of arrival of the jamming signal can be
deduced immediately from the ratio of two terms of the co-variance matrix, since
(13.67)
13.7.2.1 Single loop canceller: The simplest form of adaptive null steering
antenna is the 2-element sidelobe canceller. In this type of system, one element is
usually a high gain antenna, which may be an array or a reflector, and the other ele-
ment is virtually omnidirectional. Signals received by the omnidirectional element
are adjusted in amplitude and phase and used to cancel those received by a sidelobe
of the high gain element. (The maximum amplitude available from the omnidirec-
tional channel is restricted to prevent main beam cancellation,) Since the system
has only two basic antennas, only one interference signal can be cancelled
effectively because the combined antenna has only one independently steerable
null, although the high gain element may have many nulls in its radiation pattern.
Consider a simple array such as that shown in Fig. 13.51. The signal processor
consists of a conjugate multiplier, a filter, an amplifier with gain G and a signal
Array signal processing 401
weighting multiplier. The conjugate multiplier and the filter effectively correlates
the output of the omnidirectional element with the output of the array. In practice
conjugate multiplication is achieved by selecting the lower (i.e. difference) sideband
of a mixer.55
u = vo\q= (13.70)
where V\ is the conjugate of V2 and an overbar denotes an average value. Referring
to Fig. 13.51, it follows that
W
z = (13.71)
402 Array signal processing
and combining eqns. 13.69,13.70 and 13.71 gives the differential equation
when G~l < |F 2 | 2 . If it is assumed that a wanted signal from the broadside position
has insignificant average power compared with the interference signal received from
some other angle, then the signal in the omnidirectional element is almost entirely
due to the interference. This is an important assumption because eqns. 13.73 and
13.74 make no distinction between signals received by a sidelobe or the main beam
of the high gain element. If there is sufficient gain in the processor loop, large
wanted signals received by the main beam of the high gain element could be
cancelled.
An interesting property of eqn. 13.73 is that the effective time constant for the
transient response is
r r
Thus the time required for the array to settle or converge is determined by the
interference power in the omnidirectional element, in addition to the loop para-
meters. Therefore the convergence is slow when a small interference signal is
received and fast when a large interference signal is received. If the interference
is very large the system time constant will be small and it is possible that the pro-
cessor will attempt to track the noise and fine structure of the interference signal.
If this happens the output signal will become noisy because of the rapidly varying
signal weighting function applied to signals from the omni-directional element. As
the incoming interference level is increased a point will be reached where the
system becomes unstable and the output oscillates. The rapid variations in output
signal caused by a small time constant occurring in eqn. 13.73 are known as loop
noise.56
In practice, the gain G must be chosen in conjunction with the maximum expec-
ted level of interference |F 2 | 2 to ensure that the loop bandwidth (== 1/2ITTC) is less
than 10% of the input RF bandwidth.
In order to reduce loop noise the time constant of the filter has to be increaseed
or the loop gain decreased, but this would result in even slower convergence when
weak interference signals are received. One solution to this problem is to insert an
amplitude limiting device at point A in the circuit of Fig. 13.51 which would
Array signal processing 403
maintain the value of \V2\2 constant for signals above a set threshold level. If the
device were an automatic gain control rather than a hard limiter the signal ampli-
tude could be controlled without distorting its phase characteristics. If automatic
gain control is used, terms in V2 can be replaced by K exp j<t>2, where K is a constant
and <j>2 is the phase of the signal V2. The convergence would then be determined by
G, r and K2, and would be independent of the interfering signal level.
Consider an interference signal incident at some angle Qt radians to the broadside
of the array shown in Fig. 13.51. For simplicity assume that the interference has
both a plane wavefront and is coherent over the period of time required for the
control loop to converge. In addition, further assume that no wanted signal or noise
is present. In this ideal situation
] (13,5,
V2 = G{d)VJ4)
where G^Of) and G2(6t) are the complex polar responses of the high gain and omni-
directional elements respectively at #,- with respect to an arbitrary reference
element.
VR = voltage induced in the reference element by the incoming interference signal,
2nd
(j) =
A
d = element spacing
X = interference signal wavelength
Inserting eqn. 13.75 into eqn. 13.74 gives
The example discussed is very idealised. In practice there is 'wanted signal' in the
high gain element and some residual in the second. Both elements will also have
some level of random noise. The effect of these additional signals is to: 56
(a) alter the loop convergence
(b) offset the null from the desired position
(c) soften the resulting null
The precise effect of these additional signals can be determined by including them
in the terms for Vx and V2 in eqn. 13.76.
An interesting consequence of the presence of random or receiver noise terms is
that the cancellation of the interference signal is limited by the noise power in the
omni-directional element. The cancellation of the interference cannot exceed the
interference signal-to-noise ratio in the omni-directional element, i.e.
^(MAX) ~ ~ ^ (13.78)
where
Pn = interference signal power received by the high-gain element
Pj0 = interference signal output power of the array
PI2 = interference signal power received by the omni-directional element
N — receiver noise power in the omni-directional element.
When this ratio is small, little cancellation occurs. This problem is not serious
because it is only significant when low level interference is received.
13.7.2.2 Bandwidth: The dispersive effects of the feed system and path length
mismatches limit the canceller's broadband performance. Long feeders between
elements and the canceller require careful matching in both dispersive and fixed
delays to avoid differential phase distortion. The spacing of the antennas can be a
further source of difficulty. When a large high gain element is used it may be
impossible to place the two antennas closer than several wavelengths apart. It is
possible that the subsequent delay introduced in one channel cannot be equalised
to that in the other for all directions. Generally, the frequency sensitivity of the
high gain element's polar response is the dominating factor in determining the
arrays broadband performance.
The solution of this set of simultaneous linear equations can be expressed as the
ratio of two determinants. Thus for the M i loop:
Gi(P1)G2(61)...Go(0l)...GN(61)
Gi(O2)G2(02)...Go(e2)...GN(d2)
Gi(e2)G2(02)...Gk(02)...GN(82)
-80 r
-150
10 15 20 25 30 35 40
angle to moving interferes deg
auxilliary channels
main channel
-*«-jconjugate
adder
A schematic for a multiple loop sidelobe canceller is shown in Fig. 13.53. Limi-
ters can be included in the conjugate branch of each auxiliary channel in an attempt
to provide a fast response.55 However, each limiter is captured by the strongest jam-
ming signal with the result that this signal is cancelled rapidly, but the loop gain to
Array signal processing 407
13 JJ Adaptive arrays
The concept of sidelobe cancelling can be extended to arrays with many elements,
each additional element allowing an extra null to be steered.55'58'59 In a situation in
which an iV-element array is receiving interference from /jammers, then ideally /
nulls would be directed at the jammers, leaving the remaining N — J— 1 nulls to be
adjusted so as to optimise the performance according to some predetermined
criteria. Of course, the number of jammers is rarely known in advance (and may
even exceed the number of elements), and the most widely adopted criterion is
simply that of optimising the ratio of signal power to interference plus receiver
noise power.
Consider a linear array of TV elements such as that shown in Fig. 13.38, which is
receiving a wanted signal from a direction 0. In the absence of any interference, the
signal has to compete only against receiver noise, which can be assumed to be un-
correlated from element to element with equal mean noise powers. It is well known
that in this circumstance, the signal-to-noise ratio is optimised by combining the
element voltages with uniform amplitude weights which have phase shifts which are
simply the conjugates of the signal phases at the elements. In the presence of inter-
ference from one or more directions, the net noise (interference plus receiver) at the
elements exhibits partial correlation and uniform weights are no longer optimum.
408 Array signal processing
convenient to place the fictitious reference element in the 'zero' position located at
—d from the first element. Expressions for sk and Gkq then become:
sk(6) = exp [(J2irkd sinO)l\] (13.85)
and
Gk(0q) = exp [Q'lnkd sindq)/\] (13.86)
When the element voltages are weighted by the factors Wx to WN and added
together, the mean power output due to signal is
I (Wksk) I (WX)\ (13.87)
whilst the mean power output due to interference plus receiver noise is
^±^ £ (13.89)
Pn + APn Pn
- ^ = 5? = H (13.90)
*Pn Pn
where H denotes the stationary value. To be more specific, assume that the weight
Wk is represented by an amplitude and by a phase term:
Wk = Rkexp(jtk) (13.91)
then the derivatives of Wk and of R/£ for amplitude and phase perturbations can be
written as:
whilst
^ = jWk and ^ = -jW*k (13.93)
Consider first a change in amplitude only. Differentiating Ps with respect toRk and
using the above results yields
410 Array signal processing
p=i
Taking the ratio of these last two equations and using the relationship for the
stationary value H yields
P=N P=N
I (WpSp)
l
^ P=l P=l
It will be noted that the two equations above differ only by the signs in their
numerators and denominators and that the right-hand side of each term in a large
bracket is the conjugate of its left-hand side. Hence they can be combined to yield
an equation which has to be valid for all values of k:
(wpSp)
(WpE*kEp) (fc=l,2,...,JV) (13.98)
H
Now the term %(WpSp) is the same in all k equations above and, furthermore, the
value of Ps/Pn will not be affected if all the weights Wx to WN are multiplied by
any arbitrary constant. The above equation can therefore be reduced to
I (WpElEp) (13.99)
p=i
where C is any arbitrary constant. Applying this for all values of k enables the
weights to be related by the set of simultaneous equations:
W1E*1El + W2E\E2 + . . . 4- WNE\EN = Cs\
^ £ I + . . . + ^£,,= . (i3ioo)
The above result is often derived by matrix algebra which enables the set of simul-
taneous equations to be written in compact form as
Array signal processing 411
S* = (13.102)
is a column vector defining the wanted signal direction, i.e. the 'steering vector',
W = (13.103)
= El
E*2E2 . E2EN (13.104)
m
mm NN
M \M\ \M\
An A22 Am
l
M~ = \M\ \M\ \M\ (13.107)
AW Aw
\M\ \M\
where \M\ is the determinant of M and AJk is the cofactor of m-ih (i.e. the reduced
determinant obtained by deleting the /th row and the &th column, with a sign given
by (—l) J+fe ). The formal procedure for finding the inverse is thus to replace each
element mjk by the quotient of its corresponding cofactor and the determinant \M\
and then to interchange rows and columns.
It will be appreciated that, if the determinants are evaluated by multiplying out
the various terms in accordance with Cramer's rule, excessive numbers of multipli-
cations are required and the rounding-off errors can be substantial. It is more usual
to solve the simultaneous equations (eqn. 13.100) directly by elimination tech-
niques which successively subtract multiples of one row from another so as to con-
vert the original equations from the form
Wxmn + W2m12 = Cs\
W2m22 WNm2N = Cs*2 (13.108)
where all the terms below the leading diagonal have been reduced to zero. (The x
and y above represent the modified coefficients for the other terms.) The equations
can then be solved for the weights by back substitution, starting with the last equa-
tion for Wjsf. This method reduces the number of multiplications required, but some
care is required in selecting the order in which the elimination proceeds in order to
minimise rounding-off errors.61 Even so, double precision arithmetic is generally
required in the computations.
For large arrays, iterative methods of solving the simultaneous equations are
sometimes used.62 Although these generally require more computation time than
the elimination methods, they are more economical in the use of computer memory.
Furthermore, they preserve the original matrix (unlike the elimination method
where the original is destroyed as the computation proceeds) as a means of testing
the latest iterative for the weight vector and the rounding-off errors are usually less
severe.
Discussion on the methods used in practical arrays is deferred until after some
further properties of the co-variance matrix have been described.
IS. 7.3.3 Steering vectors incorporating amplitude weighting: For a linear array,
the simplest steering vector for a wanted signal direction 0 is
exp[(--;2m/sin0)/A]
* _ exp[(—j4nd sin0)/X]
(13.110)
exp[(—j2Nnd sint
0 0 Qn
where Qn is the receiver noise power per element. Applying this to equation 13.100,
it can be seen immediately that the optimum weights in the absence of interference
are
C
w = —s* (13.112)
Qn
414 Array signal processing
Thus, as expected, the system reverts to uniform amplitude weights in the quiescent
condition.
In many applications, array designers are willing to sacrifice perhaps 1 or 2 dB in
signal-to-noise ratio in the quiescent condition in order to obtain lower sidelobes.
This can be incorporated readily by including the quiescent design weights b x to b^,
which control the quiescent sidelobe level in the desired manner, into the steering
vector. Thus the new steering vector is taken as:
exp [(-/2m/ sin 0)/X]
[(~-j4nd s
S* = (13.113)
(Qn+Pl+P2)
(Qn +
(Qn P X + P 2)
(13.114)
=
where Pu P2 mean interference power per element due to jammer number 1, 2
and 0x = (2nd sin #i)/A and 0 2 = {^d sin 02)A-
The theoretical adapted patterns are obtained by evaluating the inverse of ilf and
multiplying by the appropriate steering vector. The first diagram in Fig. 13.54 cor-
responds to uniform weighting, i.e.
(13.115)
whilst the second diagram shows the effect of incorporating binomial amplitude
weighting into the steering vector:
Array signal processing 415
10 dB
-30
-180
^ = 10 dB
416 Array signal processing
-20
-30
-180° -90* 90° 180°
2tTd sine
(c)
Fig. 13.54 Adapted radiation patterns of a four element array with two jammers
a Uniform amplitude steering vector S*T = [ 1 , 1 , 1 , 1 ]
b Binomially-weighted steering vector S*T = [ 1 , 2, 2,1 ]
c Omnidirectional steering vector S*T = [ 1 , 0, 0, 0]
quiescent pattern
adapted pattern
Adapted weights
(a) Uniform (b) Binomial (c) Omnidirectional
0.4056+/0.0210 0.3837 +/0.0852 0.7535+/0
W2 0.5627+/0.1359 0.5778+/0.1080 0.0433 -/0.0193
0.5627-/0.1359 0.5778-/0.1080 0.4886 -/0.2777
0.4056-/0.0210 0.3837 -/0.0852 -0.2349 +/0.2426
o*
(13.116)
The nulls in these example patterns are rather broad, since there are only four ele-
ments in the array. With a greater number of elements, much sharper nulls are
obtained.
This constraint ensures that the RMS weight value is unity and maintains that part
of the output due to receiver noise alone at a constant value. Since large jammers
are usually cancelled to well below receiver noise, this constraint is fairly effective
in maintaining constant false alarm rate properties for radar receivers.
The above constraints are essentially single point constraints and do not affect
the operation of the maximum signal-to-noise-plus-interference criterion, apart
from the change in the scale factor C. However, they may become inadequate if the
direction of wanted signals is now known precisely. In radar applications, for exam-
ple, the measured co-variance matrix is often allowed to include wanted signal
418 Array signal processing
terms, because they are difficult to exclude and it is argued that their average
energy is small compared to the interference and receiver noise energy. In some
modern high-duty-cycle pulse Doppler or pulse-compression radars, the signal
energy may approach the level of receiver noise energy and in this case the adaptive
circuits will begin to cancel the signal.
This situation is illustrated in Fig. 13.55, which shows the main beam of the
quiescent pattern, which is applicable when the signal energy is much smaller than
the receiver noise energy, and the adapted response (labelled 'zero-order con-
straint') as a function of signal direction for a signal-to-noise ratio of OdB. The
effective beam width is reduced to about one tenth of the nominal.
/ f l \
S 10 I \ Vi-^^^-quiescent beam ( l o w S/N)
-first-order constraint
20
zero-order constraint
30
Expressions for the mean signal power and for the mean interference plus noise
power corresponding to eqns. 13.87 and 13.88 can then be set up by making the
substitutions
and } (13.124)
can be regarded as a constant multiplier. Eqn. 13.126 indicates that each adapted
weight Wk is equal to the quiescent weight minus a uniform amplitude weight (\B\)
phased to direct a beam at the jammer. Eqn. 13.127 indicates that the amplitude of
the retrodirective beam is proportional to the jammer power Px and to the ampli-
tude of the final adapted beam at <plt The absolute phase of the retrodirective beam
is such as to achieve substantial cancellation of the jammer.
420 Array signal processing
The concept above can be extended to cover / jammers, even if / exceeds the
number of array elements N. Thus the adapted weight vector can be written as
(13.128)
Qn i=l
where
(13.129)
jammer 1 jammer 2
lOdB
-ieo #
Fig. 13.56 Analysis of adapted pattern as quiescent pattern and two non-orthogonal retro-
directive beams
Quiescent pattern
Retrodirective beam 1
Retrodirective beam 2
Adapted pattern
Whilst the above analysis gives a simple physical picture of the operation of an
adaptive array, it does not readily reveal the amplitudes of the various 'uniformly
illuminated' retrodirective beams (until the actual values of the adapted weights
Array signal processing 421
have been computed). The amplitude of each retrodirective beam depends in a com-
plicated way upon all the jammer power levels and directions, and the various
beams are generally not orthogonal to each other.
Further insight into the operation of the array can be obtained by analysing the
retrodirective beams as a weighted set of particular orthonormal beams known as
the eigenbeams, as discussed below.
JW> i :7 03-132)
(/*/)
The first condition above simply normalises the gains so that the output power due
to receiver noise alone is Qn for all the eigenbeams. The second condition (i%j) is
the mathematical criterion for orthogonality, but it is worth noting that it produces
strictly orthogonal beams only if the radiation pattern of each element of the array
has a special form. If
422 Array signal processing
E(d) = complex voltage gain pattern of each array element (13.133)
then the overall radiation pattern is
J Gi(pyGj(d)* dd = 0 (13.135)
-7T/2
Substituting eqn. 13.134 and changing the variable of integration from 0 to 0 yields
J0 (13.136)
fVcosJ (\sm6\<\l2d)
\E(B)\ - (13.138)
I 0 (|sin(9i>A/2d)
The last equation represents Wasylkiwskyj's Ideal pattern,64 which has a eos0
power pattern, but is zero beyond |sin 6 \ = \/2d. (This corresponds to the scanning
angle at which grating lobes just begin to appear.) Much of the theoretical treatment
of adaptive arrays implicitly assumes this ideal element pattern, but practical arrays
rarely achieve it due to mutual coupling effects. However, as noted earlier, mutual
coupling effects are automatically compensated for by the maximum signal-to-
noise-plus-interference criterion and the idealised conditions assumed enable the
basic mechanisms of adaptation to be understood.
Returning to the definition of eigenvalues, if the output noise plus interference
power Pn is regarded as a function in the 2Af-dimensional space defined by the
weights, then each eigenvalue \t corresponds to a maximum value of Pn.
Eqn. 13.88 gives the expression for Pni but in order to take account of the
normalisation given by eqn. 13.132, the quantity to be maximised is
A rray signal processing 423
fe-iV p=M
I (xkEk) I
k=l
(X;E*P)
P — (13.139)
kN \
k
I (xkxt)\j
Maximising the ratio (interference-plus-noise)/(noise) in this way enables each
weight to be considered as a completely independent variable and the interrelation-
ship inherent in eqn. 13.132 to be accounted for later.
The maximisation of Pr can proceed along the same lines as before, differentiat-
ing with respect to amplitude Rk and phase \pk of each weight Xk. (cf. eqn. 13.91
et seq.). The details are omitted here, but the final result is another set of simul-
taneous equtions:
(13.140)
1
where Xs is a maximum of Pr (i.e. an eigenvalue).
Alternatively, these equations can be expressed in the deceptively simple matrix
form:
\SX = M-X (13.141)
Written out in full, the equations are
(E\E2)
(E2EO (E2E2 ~ (E*2EN) = 0 (13.143)
(E*NE2)
usually large when there are closely spaced eigenvalues. For this reason, eigenbeam
analysis is not used in simple practical adaptive arrays, but is a useful tool for theo-
retical analysis and for more complicated arrays where a detailed analysis of the
jamming environment is required. Several iterative numerical procedures exist
which determine eigenvalues and eigenvectors simultaneously.62
Since the eigenvalues correspond to power ratios they are always real and posi-
tive. Mathematically, this property is associated with the fact that the co-variance
matrix is Hermitian, i.e. terms in complementary positions on each side of the lead-
ing diagonal are complex conjugates.
Since Xs appears only in the diagonal terms of the determinant (eqn. 13.143), it
follows that when the determinant is expanded as a polynomial:
coefficient of Xf = ( - 1 ) * (13.145)
and fe=N
coefficient of xf" 1 = (~ l ) ^ " 1 £ E%Ek (13.146)
(The coefficients of lower powers of Xs are more complicated). The negative of the
ratio of these two coefficients gives the sum of the roots for any polynomial and
hence S=N U=N
I Xs = I E*kEk (13.147)
Thus the sum of the eigenvalues is the total interference plus noise power incident
on the array elements.
Having determined the eigenvalues, each corresponding eigenvector X^l) can be
obtained by substituting the value of Xf back into the simultaneous eqns. 13.142
and solving these for the various weight components X\1^ to X$. Since these N
equations are essentially an overdetermined set in the iV— 1 variables X2/Xi to
XJV/XJ, any N — 1 of the equations can be solved for these weight ratios and the
actual weight values found from the normalising condition
fe = iV
I \Xk\2 = 1 (13.148)
Ut = I 4 % (13.149)
fe = i
The complete set of eigenbeam voltages can thus be represented by a vector
U = LE (13.150)
where
Array signal processing 425
\f
• /
E
w2
1
I
add
T r. ,-T.
2 (W k E k ) = W E = JE W
(a)
Fig. 13.57 Weighted linear array (a) and its eigenbeam equivalent (b)
426 Array signal processing
L = (13.151)
WN (13.152)
Now for / = / the eigenvectors have been normalised by eqn. 13.148 and hence
UfUi = W=h (13.156)
Hence the sum of the eigenbeam powers is equal to the sum of the eigenvalues,
which has already been shown to be the total interference plus noise power incident
on the array. It therefore follows that
UfUj = 0 (13.157)
and that the eigenvectors form an orthonormal set, satisfying eqn. 13.132. (If the
eigenvectors were not orthogonal, it would be possible to form some combination
of eigenbeam voltages which increased the total power.)
Mathematically, L is the transpose of the 'modal matrix'. Since L is an orthogo-
nal matrix, its inverse is equal to its conjugate transpose:
L1 - Z* T (13.158)
and it transforms the co-variance matrix M of the element voltages E into a diagonal
Array signal processing 427
0 0
For adaptive arrays, the advantage of an eigenbeam analysis is that the co-variance
matrix i$, reduced to the above simple diagonal form and that the various eigen-
beam voltages are decorrelated. This topic is addressed in more detail later, after a
few examples of eigenvalues and eigenbeams have been discussed.
=0
(13.161)
where 4>l = (2n sin 0^/X.
The determinant can be reduced by the usual rules for operating upon determin-
ants. First, multiply the first column by e]k<t>l and subtract from the (k + l)th
column (k = 1,2,3). This gives
0 0 0
- o o = 0 (13.163)
0 (Qn-\) 0
0 0
428 Array signal processing
The determinant can then be expanded to give the characteristic equation as:
(Qn + 4Pl-\s)(Qn-K)3 = 0 (13.164)
which indicates that there is one unique eigenvalue
*i = Qn + Wi (13.165)
and three equal roots or non-unique eigenvalues
X2 = X3 = X4 = Qn (13.166)
The above reasoning can clearly be extended to an iV-element array with a single
jammer to show that the unique eigenvalue is
Xi = Qn + NPt (13.167)
The unique eigenvalue corresponds to the maximum interference power which can
be abstracted (NPX) plus receiver noise power corresponding to one element. The
non-unique eigenvalues each correspond to receiver noise for one element, and
hence the sum of the eigenvalues represents the total interference plus noise power
of the system.
Formally, each eigenvector XS1* can be obtained by substituting the appropriate
eigenvalue \ into the simultaneous equations and then solving these equations for
the various weight components X*p to X$.
However, in the present example it is easier to proceed by writing eqn. 13.142 in
the form
XXPX
X2PX
(13.170)
which represents uniform amplitude weights, phased to direct a beam at the jam-
mer. The radiation pattern of the unique eigenbeam at any angle 6 is thus
Array signal processing 429
xcxp/[5(0-00/2] (13.171)
which is, of course, the usual (siniVx/sinjc) shape of a uniformly illuminated beam.
The phase angle of this beam is a result of having chosen the hypothetical phase
reference element in the 'zero element' position of the array. The peak amplitude
of the beam is 2, corresponding to a power gain of 4, equal to the number of
elements.
For the three non-unique eigenvalues Xs = Qn and all the equations 13.168
reduce to
= 0 (13.172)
which indicates that the corresponding eigenbeams must all have a zero at 0 = 0X.
Apart from the orthonormalising constraint (eqn. 13.132), eqn. 13.172 is the only
other constraint on these beams and it is clear that their eigenvectors are indeter-
minate. The orthonormal set can be completed by 'inventing' (sinNx/sinx) beams
of the same shape as the unique eigenbeam, spaced in 0 by n/2 as in a 4-element
Butler matrix. Thus the non-unique eigenvectors could be chosen as:
0 5 e"1 0" -
05
(1 = 2,3,4) (13.173)
05
However, it must be emphasised that this choice is arbitrary. For example, it can be
readily checked that the three weight vectors:
1 _,„ Q
0 -O.Se"' 301
0 0.5 e"
(13.174)
satisfy eqn. 13.168, are mutually orthogonal and are orthogonal to the unique
eigenvector (eqn. 13.170). Thus they, and many other possibilities, could be con-
strued as completing the set of orthonormal eigenbeams.
430 Array signal processing
In summary, there is only one unique eigenvector for a single jammer illuminat-
ing an TV-element linear array, given by
1 _,*
(13.175)
The other eigenvectors are indeterminate, but all non-unique eigenbeams have a
zero in the jammer direction. Thus, in theory, this direction can be determined
exactly. In practice, it will be disturbed by noise in the measured co-variance matrix
due to the finite, rather than infinite, measurement duration.
= (K~Qn)X\
+ .. . + X = (ks-Qn)X2
N
(13.176)
where Pt and fa refer to the zth jammer and the summations extend from i = 1 to / .
It is convenient to introduce the function
p=N
= e
-4(0) 2- (13.177)
which is simply the pattern of a uniform amplitude constant phase excitation, with
phase referred to the first element. Note that A(— <j>) =A*(<f>). Now define
p=N
(13.178)
P=I
Array signal processing 431
so that the various Bt represent the complex amplitude of the as yet undetermined
eigenbeams at the various jammer directions 0,-. If each kth row (k = 1 to N) of the
equations is multiplied by e^k~1^1 and all the equations are then added together,
the result can be written as
i^ ( 0 2 - 0 0 = 0
(13.182)
Thus for theoretical problems where the jammer powers and directions are speci-
fied, the above reduced form of characteristic equation can be used to find the
unique eigenvalues. Note that these eigenvalues depend upon the jammer powers
and the differences in their bearings, rather than upon the absolute bearings. (The
N—J non-unique eigenvalues are given by Xs = Qn, as before.) Having determined
the unique eigenvalues, the corresponding eigenvectors can be found by substituting
back into the original equations 13.176.
Applying eqn. 13.182 to the case of two jammers, the determinant can be
expanded, giving the quadratic equation
(NPi + Qn- \WP2 + Qn - \) = PlP2\A(4>2 - 0OI2 (13.183)
2
The quantity l^4(0 2 ~0i)l is simply the power pattern at angle 4>2 — <i>l for a uni-
formly illuminated array and is reasonably small for jammer separations exceeding
one basic beamwidth of the array. In that case, the two unique eigenvalues are
432 Array signal processing
and (13.184)
The relations are exact when the jammer separation is a multiple of one beamwidth.
CQ
90° 180°
relative jammer spacing , I ^ - ^ I
Fig. 13.58 Eigenvalues for two jammers (Px/Qn = 20dB: P2/Qn = WdB)
4 element array
8 element array
Fig. 13.58 shows how the eigenvalues vary as a function of jammer separation. For
small separations, there is one large eigenvalue, approximately Qn + N{PX + P2), and
the second eigenvalue is just greater than receiver noise Qn. At large separations, the
lower eigenvalue is more affected by the exact jammer separation than is the higher
eigenvalue.
Array signal processing 433
beami
beam 2
beam 3
jammer 1
30r
-180
2TTCJ
sin 9
Fig. 13.59 (b) Power observed with uniformly illuminated scanning beam {beam port 1)
Jammer 1 alone
Jammer 2 alone
Observed resultant
434 Array signal processing
Some physical appreciation of eigenbeams and eigenvalues can be obtained with
the help of the orthonormal beam-forming network of Fig. 13.44. Consider a
4-element array feeding such a network and assume that there are two jammers
present, as in Fig. 13.59. Suppose that a power meter, with adequate smoothing to
ensure it measures true mean values, is connected to beam port 1. Assume that the
first row of power couplers in the network has been set to the values 1/2,1/3 and
1/4, corresponding to uniform amplitude weighting, and that the first row of phase
shifters are adjusted in sympathy, with values \p, 2$ and 31^, so as to slowly scan
the beam throughout the full range of angles. The measured mean power output
will be as shown in the solid curve of Fig. 13.59.
At this stage, it will be apparent that there is a least one jamming source present,
but the response corresponding to the lower-powered jammer may be mistaken for
a sidelobe. Suppose that the phase shifters are returned to the peak power position
and that the various power couplers and phase shifters in the first row are then
gently perturbed in turn so as to attempt to increase the power meter reading.
Eventually, after several iterations, a true maximum will be found. This is the
largest eigenvalue and the weights corresponding to the coupler and phase shifter
settings in the first row form the eigenvector. The radiation pattern of beam port 1
is the corresponding eigenbeam.
The power meter can now be transferred to beam port 2 and a second scanning
beam set up with the couplers and phase shifters in the second row. For each scan
angle, the couplers and phase shifters should be set so as to produce maximum gain.
(The maximum realisable gain will generally be less than N owing to the spatial
filtering effect of the first eigenbeam.) These settings will involve somewhat messy
computations, since they depend upon the values set into the first row. Having
completed the scan, the settings can be returned to the greatest measured peak and
the effect of minor perturbations in the settings investigated. Since only two jam-
mers have been assumed, no further adjustment of the settings will be necessary.
(Had a third jammer been present, further adjustment would have been required.)
The output of the second beam is the second eigenvalue and the element weights
corresponding to the combined settings of the first and second row form the second
eigenvector.
If the power meter is now transferred to the third beam port and the procedure
followed to set up a third scanning beam, it will be found that the third beam out-
put power is constant (at receiver noise level) and is not affected by the settings of
the third row of controls. The first two eigenbeams have abstracted all the jamming
power incident on the array and the remaining two non-unique eigenbeams are in-
determinate. Any setting of the third row of controls produces two orthogonal
beams at ports 3 and 4, which have zeros in the jammer directions and which com-
plete the set of eigenbeams.
In theory, the network described above could be used to determine all the jam-
mer bearings, since the non-unique eigenbeams have zeros at all jammer bearings.
The third row of controls, which forms the only degree of freedom left in the
example, could then be set to produce a maximum realisable gain beam in a given
A nay signal processing 435
signal direction. (In a more general case N — J—l degrees of freedom would be
available to optimise signal reception.) In practice, such an open loop method of
control would be unsatisfactory because of difficulties in measuring mean powers
with sufficient accuracy and because of the interdependence of the various control
elements. However, the eigenbeam network, even when set up only approximately,
is a very useful pre-processor to put in front of a simple adaptive processor using
the maximum signal-to-noise-plus-interference algorithm, since it improves the over-
all convergence time.
_ I (13.185)
where \t is the fth eigenvalue. For a two-jammer situation, the positive contribution
to this cross-product from one jammer is balanced by an equal negative component
from the other jammer (except, of course, for the self products / = k).
Mathematically, the eigenbeam network shown in Fig. 13.576 is equivalent to an
orthonormal matrix L which is the N by N matrix having the eigenvectors for its
rows.
Thus
L = (13.186)
436 Array signal processing
The relationship between a particular eigenbeam voltage Ut and the element volt-
ages Ei.. .EN is
Ut = J * 1 % + Xi2i)E2 + . . . + X ^ N (13.187)
and hence the vector representing all the eigenbeam voltages is
U = LE (13.188)
Similarly, the steering vector 5* at the elements can be transformed into an equi-
valent steering vector r\ * representing the weights which should be applied to the
eigenbeam channels to obtain the desired quiescent pattern. Thus
TI* = L*S* (13.189)
The same transformation gives the weights co which must be applied to the eigen-
beam voltages U before summation in order that the output of the array shall be
the same as the output of the original array with the element voltages E weighted
by W. Since the output voltage is
ETW = UTv = (LEfu = ETLTo> (13.190)
then T
W = £ co (13.191)
and
co = L*W (13.192)
It will be recalled that the derivation of the maximum signal-to-noise-plus-interfer-
ence condition for the original array:
W = CM~l'S* (13.193)
permitted the elements of the array to have differing patterns. If the eigenbeam
voltages Ux... UN are regarded as the element voltages of a new array, then it will
be evident that the maximum signal-to-noise-plus-interference condition in 'eigen-
beam space' will be of similar form:
co = CM"1-*!* (13.194)
where
U\U2 N 0
¥
U 2Ul U*2UN 0 x2 (13.195)
U*NU2 UNUN 0 0
is the co-variance matrix of the eigenbeam voltages and which, as shown earlier, is
simply the diagonal matrix of the eigenvalues. Since each component of eqn.
13.195 is now decorrelated, the optimum weights in eigenbeam space can be writ-
ten down immediately as
(13.196)
Array signal processing 437
which is the 'matched filter' condition referred to above. The optimum weights W
required for the original array can then be obtained from eqn. 13.191.
To apply the above result to the previously considered case of a 4-element array
illuminated by a single jammer, for which there is one unique eigenvalue:
Qn + 4Pi (13.197)
and three non-unique eigenvalues equal to receiver noise Qni it is first necessary to
choose a set of eigenvectors for the non-unique values. This unique eigenvector is
given by eqn. 13.170, and eqn. 13.173 will be used for the non-unique eigen-
vectors. (It can be shown that the exact choice does not affect the final result and
eqn. 13.174 could also have been used for these eigenvectors.)
The orthonormal transformation matrix representing the eigenbeam former for
this example is then
L = 0.5 (13.198)
which can be recognised as being equivalent to a Butler matrix steered so that one
beam points directly at the jammer.
For simplicity, assume a uniformly weighted steering vector at broadside, i.e.
S* = (13.199)
The transformed steering vector in eigenbeam space is, from eqn. 13.189:
"sin (200
sin + n) oJ(5n/4)
sin(01/2 + ir/4)
n* = o.5 sin (20! + 2TT)
(13.200)
438 Array signal processing
which is simply the set of weights to be applied to the (sin Aft/sin x) shaped eigen-
beams in order to produce maximum gain at broadside. The phase terms in eqn.
13.200 are the result of having chosen the phase centre of the array in the zeroeth
element position and of having selected arbitrary absolute phases for each eigen-
vector. Fig. 13.60 illustrates the weighted eigenbeams for the quiescent condition.
From eqn. 13.196, the optimum weights in eigenbeam space, neglecting common
arbitrary phase and amplitude constants, are:
1 sin (200
\ sin(0i/2)
sin (20i + 7T) eJ (Sir/4)
Qn sin(0!/2 •f it 14)
w = (13.201)
sin (20i •
2}
Qn
Qn
sin(0i/2-
sin (20!
sin (fc/2-
VL )
Since there is only one jammer, the eigenbeam structure is not affected by the jam-
mer power level in this case. Only the <JO\ component of the eigenbeam weight
vector varies if the jammer power changes, since the other eigenbeams have a null in
the jammer direction. Inspection of the equation above indicates that the jammer is
not cancelled completely, but its amplitude is attenuated by the factor
Qn
Qn + &1
The signal-voltage contribution from eigenbeam channel 1 is attenuated by the
same factor. The adapted beam's radiation pattern has a small residual response in
the jammer direction which balances the net signal received from the jammer
against the receiver noise and against the loss of gain in the broadside signal
direction (see Fig. 13.60).
The weights required at the element level to maximise the signal-to-interference-
plus-noise ratio can now be found from the transformation:
W = LTco (13.202)
The result is:
Pi ) (sin(20Q|
W = (13.203)
] (sin(01/2)j
from which it is evident that, even with a single jammer, all element weights have to
Array signal processing 439
jammer
J - = 10 dB
-non-unique Eigenbeams
Eigenbeam
(A, =41)
A* = 1
be adjusted. In the above form, the weight vector is simply the quiescent uniform
weight minus a weighted eigenbeam pointing at the jammer. The (siniVx/sinx) fac-
tor is the amplitude of the quiescent beam in the jammer direction.
Fig. 13.61 shows the result of an eigenbeam analysis for the slightly more com-
plicated case when two jammers are present. In this example, the eigenbeams are no
longer of (siniVx/sinx) shape. The adapted pattern is the same as that shown in Fig.
13.56: the two (sinTVx/sinx) beams in that diagram have merely been resolved into
their eigenbeam components.
440 Array signal processing
quiescent beam
adapted pattern
90°\
JL
F ig. 13.60 Eigenbeam analysis o f adap tive 4-etemen t array with one jammer
a Quiescent pattern
b Eigenbeams corresponding to single jammer
c Eigenbeam components of quiescent pattern
d Adapted pattern and its components
jammer 1 jammer 2
lOr
=
\
/ Eigenbeam 1 . Eigenbeam 2 \
f
\ .*' A 1 = 403.219 A2=38.781
m
V
•10-
•D -1
C
o
en
\l
\
-20
1/ \
-30
180° 90° 90° 180°
(a)
quiescent beam
-A0-
\
• - I weighted . ,
U" I ''-. Eigenbeam 1 II
-50
180°
i ii '"< . ii
90° 0° 90° 180°
(b)
Fig. 13.61 Eigenbeam analysis of adaptive 4-element array with two jammers
a Unique Eigenbeams corresponding to the two jammers
b Adapted pattern and its components
442 Array signal processing
beam former
and the outputs of all the correlators can be represented by the column vector
= V*(WTV) (13.208)
Array signal processing 443
where V is the input element voltage vector (noise plus interference) and WT is the
transpose (i.e. corresponding row vector) of the column vector of weights W.
The correlator output vector s can now be written in terms of the co-variance
matrix M as:
€ = MW (13.209)
From eqns. 13.205 and 13.206, ek can be expressed in terms of and its time
derivative Wk by
e _ c._ (13.210)
and hence the complete set of differential equations governing the system can be
written in vector form as
= GS* (13.211)
where / is the identity matrix. Thus the steady-state solution (i.e. when all the com-
ponents of the derivative W have vanished) should approach the desired solution
W = M~X*S* so long as the gain G is sufficiently high.
The manner in which the steady-state solution is reached is not immediately
revealed by eqn. 13.211 because of the interrelationships between the signals in the
various feedback loops. The interdependence can be removed by transforming the
differential equations from the 'weight space' defined by the weight vector W into
'eigenbeam space'. The transformed differential equations can then be solved
readily and the inverse transformation applied, if necessary, to reveal the behaviour
of the actual weights. Paralleling eqns. 13.188 et. seq., let
« = L*W (13.212)
be the transformed weight vector and
(13.213)
be the transformed steering vector, where L is the orthonormal matrix which
defines all the eigenvectors of the co-variance matrix M. Then the transformed
differential equations become
T W + ( G M + / ) oa = (13.214)
where
0 0
X2 0
i = L*MLT = (13.215)
0 X3
0 0 0
has transformed the co-variance matrix into the diagonal form where each element
is an eigenvalue. Each component of the differential equation is now independent:
444 Array signal processing
exp ( m i 7 )
(—;—)•
where « 0 = L*W0 and Wo is the initial set of array weights.
The time constant with which the kth component of eqn. 13.217 converges is
r/(GXfe + 1). The value of G has to be chosen such that the loop does not become
unstable for the highest eigenvalue X max . (The greatest loop bandwidth should be
restricted to less than, say, 10% of the signal bandwidth.) The lowest unique eigen-
value X min then determines the rate at which the system finally converges and the
speed at which it can follow dynamic changes in the interference field. Large differ-
ences in jammer power and closely spaced jammers can cause the eigenvalues to be
widely separated and this prolongs the settling time of the system. The larger eigen-
values are cancelled rapidly, but the lower eigenvalues can take perhaps 106 times as
long. Examples of convergence times for single and multiple jammers are shown in
Fig. 13.63.
Convergence of an adaptive array with a single jammer is illustrated by the pat-
terns of Fig. 13.64. The 15-element array is receiving interference from a narrow-
band jammer in the first sidelobe, approximately 11° from broadside. At (a),
shortly after switch-on, the pattern is virtually the quiescent pattern. A little while
later (b) the sidelobe begins to be suppressed and finally a deep null is achieved at
(c).
Figs. 13.65 and 13.66 show the steady state response of a 17-element array to
spatially distributed interference and also show the effect of introducing amplitude
tapering into the array. In both cases the processor has used three zeros to provide
some 50 dB rejection over the region of jamming. In Fig. 13.65 amplitude tapering
has been included in the steering function, whilst in Fig. 13.66 the taper has been
applied at the elements, before the adaptive processor. The sidelobe structure of the
two patterns differs because the steering function in the first case attenuates both
interference and noise from the edge elements, whilst in the second case the taper is
applied to the interference only and the front end noise is unchanged. In the latter
case, less interference power is available to the processor and the eigenvalues of the
covariance matrix are reduced, increasing the convergence time slightly.
30
DO
15
time —-
where k is the iteration number and g is a parameter which controls the stability
and rate of convergence. As in the analogue system, g has to be made small enough
to ensure that the algorithm does not go unstable for the highest expected eigen-
value X max . The disadvantage of this simple step-by-step technique is that it is
inevitably slow. The size of the perturbations has to be restricted to prevent excess
446 Array signal processing
-60 -30 0 30 60 90
ang. from broadside, deg
(b)
'perturbation noise' and the number of samples averaged for each step has to be suf-
ficiently large to ensure reasonable accuracy in the gradient measurement. Improved
response times can be obtained by modulating all the weights at different frequen-
cies simultaneously and recovering the gradient vector by employing phase sensitive
detectors. Such systems have found application in fixed point-to-point communica-
tion antennas where the interference is essentially stationary.65
Much faster response times are obtained with the widely-used least-mean-square
(LMS) algorithm. This uses digitally implemented versions of correlation detectors
Array signal processing 447
90 -60 -30 0 30
angle from broadside.deg
ic)
Fig. 13.64 15-element adaptive array receiving interference 11° from the broadside (no
amplitude taper)
a Pi/PN = 60dB; G/T = 250; T = 0.1 ms
b PI/PN = 60 dB; G/T = 250; T = 0.75 ms
c Pj/PN = 60dB; G/T = 250; T = 10ms
performing the same function as those shown in Fig. 13.62 to obtain a direct assess-
ment of the gradient vector without needing to perturb the weights. The algorithm
(eqn. 13.218) can be updated at each data sample,66 although some systems average
the gradient measurement over a number of data samples.67 When updating at each
sample is employed, the gain g has to be smaller than 1/Xmax to ensure stability and
low loop noise'. Typically, the effective time constant for the highest eigenvalue
can be as low as 2.57V data samples, where N is the number of array elements. With
a single jammer, the system will then have fully settled in about ION data samples.
However, with multiple jammers the response to lower eigenvalues is more protrac-
ted, the time constant being inversely proportional to the eigenvalue. The perform-
ance with the LMS algorithm is thus similar to that of the analogue array system.
The requirement for algorithms which converge more rapidly for the lower
eigenvalues has stimulated research into efficient methods of performing more
direct matrix inversion.68 In the sample matrix inversion technique (SMI), an esti-
mate is made of the co-variance matrix by averaging successive data samples:
Mk = \ (13.219)
k
and the matrix is inverted by a fast algorithm to compute the weights required.
k = IN samples are usually considered adequate.60'69
In a stationary interference environment, this algorithm can be allowed to run
448 Array signal processing
• spatially distributed
: interference
'-30
-40
-50
Fig. 13.65 17-element adaptive array with raised cosine amplitude taper applied to the steer-
ing function
G/T = 500
Pl/PN= 70 dB
: spatially distributed
interference
Fig. 13.66 17-element adaptive array with raised cosine amplitude taper applied to the prim-
ary feed
G/T = 500
Pl/PN= 70 dS
Array signal processing 449
(13.222)
180(M/)
For example, a 10-wavelength aperture array has an 'adaptive null bandwidth' of
0.05% (e.g. 1 MHz a t / = 2000 MHz).
Eigenbeam
r~" former
I'
| I j | AGC amplifiers
i 'standard' adaptive
processor
time
To represent broader band jamming, each phase term eHk~l)<i>1 has to be replaced
by a correlation coefficient. Defining the normalised correlation function of a volt-
age E{t) as the complex function
T
lim M E{t)E%t~r)dt
-T
R(r) = (13.223)
1 T
lim — f E(t)E*(t) dt
then the contribution of each jammer to each element of the co-variance matrix
becomes
mik = EfEk = P1R[(i-k)r] (13.224)
where
dsin91
T = (13.225)
A rray signal processing 451
and c is the velocity of EM waves, r is simply the time delay between adjacent ele-
ments for the particular angle Bx. For a single jammer, the covariance matrix is:
R(T) = (13.227)
R(T) = exp
-rV exp (JCO0T) (13.229)
Px + Qn
M = (13.230)
Pl+Qn
where
-rV (13.231)
p = exp
P
lP n-K PlP
PlP
PlP
(13.233)
Fig. 13.68 illustrates how the eigenvalues vary as a function of correlation coef-
ficient p for a 4-element array with a Gaussian spectrum jammer. For high correla-
tion coefficients (narrow-band jamming) there is one unique eigenvalue and three
non-unique eigenvalues at receiver noise level. As the bandwidth is increased, first
one and then the other low eigenvalues increase. For very broad-band jamming
(p = 0), the eigenvalues become equal.
« or 3"
0.5
correlation coefficient,p
high-gain beam in the wanted signal direction. Multiple reference inputs to the
canceller are derived from each element. These inputs are fed to a bank of adaptive
transversal filters comprising tapped delay lines with individual control of the
weights on each tap. The combined filter outputs are subtracted from the main
signal. Feedback control of the adaptive transversal filters is obtained from correla-
tion detectors.
main beam-former
(delays)
adaptive transversal
filters
control signals
output
Fig. 13.69 Adaptive array with time delays
13.8 References
1 FRIIS, H. T., and FELDMAN, C. B. 'Multiple unit steerable antenna', Proc, IRE, 25,
1937,pp.841
2 RYLE, M.: 'A new radio interferometer and its application to the observation of weak
radio sta.is\Proc. Roy. Soc. A., 211, 1952, p. 351
3 VOGLIS, G. M.: 'Design features of advanced scanning sonars based on modulation scan-
ning - Part 1', Ultrasonics July 1971
4 BARTON, P., and KELLY, K. M.: 'Doppler MLS Signal Processing', IEE Conference on
the future of aircraft all weather operation Nov. 1976
5 RADFORD, M. F.: 'The future role of electronic scanning in ATC radar' UK Symp. Elec-
tronics for civil aviation, September 1969
6 RADFORD, M. F., and STEVENS, A. D.: 'A new three-dimensional surveillance radar'.
Military Microwaves 78 Conference proceedings p. 193, Microwave Exhibitions & Pub-
lishers
7 MILNE, K.: The combination of pulse compression with frequency scanning for 3-dimen-
sional radars',/. IERE 28, Aug. 1964
8 MOTKIN, D. L.: 'Practical aspects of a frequency scanning 3D radar'. Conf. Proc. Military
Microwaves, October 1978, p. 203
9 VAN DE LINDT, W. J.: 'Digital techniques for generating synthetic aperture radar
images',/£M/. Res. & Dev. 21,1977, pp. 415-32
10 JACK, M. A., MANES, G. F., GRANT, P. M., ATZENI, C, MASOTTI, L., and COLLINS,
J. H.: 'Real time network analysis based on SAW chirp transform processors', IEEE
Ultrasonics Symposium Proc. 76 CHI 120-5SU, pp. 376-381
11 Proc of conference on 'Optics in radar systems,' Society of Photo-Optical instrumentation
Engineers (Vol. 128) 27-29 Sept. 1977, Huntsville, Alabama
12 SHANKS, H. E.: 'A new technique for electronic scanning',IRE Trans., AP, March 1961,
pp.162-166
13 KUMMER, W. H., VILLENEUVE, A. T., and TERRIO, F. G.: 'Scanning without phase
shifters', Electronics, 29 March 1963, pp. 27-32
14 DAVIES, D. E. N.: 'A fast electronically scanned radar receiving system'/. Brit. IRE, 21,
April 1961, pp. 305-351
15 EDGAR, A. K., and JONES, I. L.: 'Flood-lighting with Nyquist rate scanning', AGARD
1970
16 JOHNSON, M. A.: 'Phased-array beam steering in multiplex sampling', Proc. IEEE, 56,
Nov. 1968
17 WILLIAMS, P. L.: 'Multiple channel receivers for IF beam forming', IEE Conf. Pub. 155.
Radar 77, pp. 427
Array signal processing 455
18 MILNE, K.: Trends in array signal processing*. IEE Conf. Publ. 144. (The impact of new
technologies in signal processing). Sept. 1976., pp. 121-134
19 MILNE, K.: 'Principles and concepts of multistatic surveillance radars'. IEE Conf. Publ.
155 (Radar 77)., Oct. 1977, pp. 45-52
20 CURTIS, T. E.: 'Digital beam forming for sonar systems' IEE Proc, 127, Pt. F, Aug. 1980,
pp.257-265
21 BARTON, P.: 'Digital beam forming for radar', ibid. pp. 266-277
22 MILLS, B. Y., and LITTLE, A. G.: 'A high resolution aerial system of a new type\ Aust.
J. Phys., 6, 1953
23 BLOMMENDAAL, R.: 'A note on multiplicative receiving systems in radar',/. Brit, IRE,
28,1964,pp.317-325
24 SHAW, E., and DA VIES, D. E. N.: Theoretical and experimental studies of the resolution
performance of multiplicative and additive arrays', /. Brit. IRE, 27, Dec. 1965, pp.
323-35
25 DAVIES, D. E. N., and WARD, C. R.: 'Low sidelobe patterns from thinned arrays using
multiplicative processing', Proc. IEE, 237 Pt. F, Feb. 1980, pp. 9-15
26 MOFFET, A. T.: 'Minimum redundancy arrays', IEEE Trans., AP-16, pp. 172-75
27 KSIENSKI, A. A.: 'Multiplicative processing antennas for radar applications,' Radio and
Electron. Eng., 29, 1965, pp. 53-67
28 KSIENSKI, A. A.: 'Signal processing antennas', Microwave J., 4, 1961, No 10. pp. 77-86
(Pt. I) and, No 11. pp. 87-94 (Pt. II)
29 PEDINOFF, M. E., and KSIENSKI, A. A.: 'Multiple target response of data processing
antennas',//^ Trans., AP-10, 1962, pp. 112-126
30 TUCKER, D. G.: The signal/noise performance of electro-acoustic strip arrays', Acustica,
8, 1958,pp. 53-62
31 COOPER, B.C.: The probability density function for the output of a correlator with
bandpass input waveforms',IEEE Trans., IT-11, 1965, pp. 190-194
32 SHEARMAN, E. D. R.: 'Non-colinear and cylindrical multiplicative arrays' /. Brit. IRE,
26, Dec. 1963, pp. 481
33 SHEARMAN, E. D. R., and CLARKE, J.: 'Aperture synthesis in ionospheric radar',
Nature, 219, 1968, pp. 143-144
34 SHEARMAN, E. D. R., BICKENSTAFF, P., and FOTIADES, L.: 'Synthetic aperture sky-
wave radar: techniques and first results', IEE Conf. Publ. (Radar present and future),
London, Oct. 1973, pp. 414
35 HARGER, R. O.: 'Synthetic aperture radar systems', Academic Press, 1970
36 KIRK, Jr. J.C.: 'A discussion of digital processing in synthetic aperture radar', IEEE
Trans., AES-ll,May 1975,p. 326.
37 SKOLNIK, M. I.: 'Radar handbook', MacGraw Hill, New York, 1970
38 BROWN, W. M., and RIORDAN, J. R.: 'Resolution limits with propagation phase errors',
IEEE Trans., AES-6, Sept. 1970, p. 657.
39 PORCELLO, L. J.: IEEE Trans., AES-6, Sept. 1970, p. 636
40 'Remote Sensing of the Terrestrial Environment', Proc. 28th Symposium of Colston
Research Society, Bristol, 5-9 April, 1976, Butterworths, London, 1977
41 CUTTERING, E., et al.: 'Mission design for Seasat-A, an oceanographic satellite' AIAA
77-31, Jan. 1977
42 GUDMANDSEN, P. E.: 'Layer echoes in polar ice sheets', /. Glaciology, 15, 1975,p. 95.
43 PORCELLO, L. J., et.al: 'Apollo lunar sounder radar system', Proc. IEEE, 62, 1974,
p. 769-783.
44 SCHELKUNOFF, S. A.: 'A mathematical theory of linear arrays', Bell System Tech. J.,
22,1943,pp. 80-107
45 DAVIES, D. E. N.: 'Independent angular steering of each zero of the directional pattern
of a linear array,'IEEE Trans., AP-15, Mar. 1967 pp. 296-298
46 HICKS, D. S.: 'Null steering in linear arrays by using amplitude control of signals in the
456 Array signal processing
feeder network/ Electron. Lett., 13, Mar. 1977, pp. 198-199
47 NOLEN, J. C: 'Synthesis of multiple beam networks for arbitrary illuminations.' Bendix
Corporation, Radio Division, Baltimore, Maryland, 21st April 1975
48 WHITE, W. D.: 'Cascade preprocessors for adaptive antennas'. IEEE Trans., AP-24, Sept.
1976,pp.670-684
49 WHITE, W. D.: Adaptive cascade networked for deep nulling*, ibid, AP-26, May 1978, pp.
396-402
50 FIELDING, J. G., et al.\ 'Adaptive interference cancellation in radar systems'. IEE Conf.
Publ. 155 (Radar 77), 1977, pp. 212-217
51 HICKS, D., and RAYMOND, G.: 'Adaptive arrays and sidelobe cancellers for communica-
tions and radar applications'. Conf. Proc. Military Microwaves, MM78, Oct. 1978, pp.
366-378
52 MELLORS, M., DAVIES, D. E. N., and WITHERS, M. J.: 'Zero steering in the directional
pattern of a linear array in the presence of mutual coupling', Proc IEE^ 117, Jan. 1970
53 DAVIES, D. E. N., and RIZK, M. S. A. S.: 'A broadband experimental null steering an-
tenna system for mobile communications', Radio & Electron. Eng., 48, Oct. 1978, pp.
509-517
54 APPLEBAUM, S. P.: 'Adaptive arrays with main beam constraints', IEEE Trans., AP-24,
Sept. 1976, pp. 650-662
55 GABRIEL, W. F.: 'Adaptive arrays - an introduction,' Proc. IEEE, 64, Feb. 1976, pp.
239-271
56 BRENNAN, L. E., PUGH, E. L., and REED, I. S.: 'Control loop noise in adaptive array
antennas,'IEEE Trans., AES-7, Mar. 1971, pp. 254-263
57 GOBERT, J.: 'Adaptive beam weighting', IEE Trans., AP-24, Sept. 1976, pp. 744-748
58 APPLEBAUM, S. P.: 'Adaptive arrays', ibid., AP-24, Sept. 1976, pp. 585-598
59 WIDROW, B.: 'Adaptive filters I: Fundamental', Stanford University Elec. Labs., Syst.
Theory Lab. Centre for Syst. Res. Report SU-SEL-66-126, Technical Report 6764-6,
Dec. 1966
60 HUDSON, J. E.: 'Adaptive array principles', Peter Peregrinus, 1981
61 RALSTON, A., and RABINOWITZ, P.: 'A first course in numerical analysis', McGraw Hill,
1978, 2nd edn.
62 WILKINSON, J. H.: 'The use of iterative methods of finding the latent roots and vectors
of matrices', MTAC, 9, 1955, pp. 184-191
63 WARD, C. R., and HARGREAVE, P. J.: 'The application of sub-optimal control methods
to adaptive antennas for airborne communication systems'. IEE Conf. Pub 195
(Antennas and Propagation) April 1981, pp. 174-178
64 WASYLKIWSKYJ, W., and KAHN, W. K.: 'Scattering properties and mutual coupling of
antennas with prescribed radiation patterns', IEEE Trans., AP-18, Nov. 1970, pp.
741-752
65 WINDRAM, M. D., et al: 'Adaptive antennas for UHF broadcast reception', IEE Proc,
127, Pt. F, Aug. 1980
66 WIDROW, B., and McCOOL, J. M.: 'A comparison of adaptive algorithms based on the
methods of steepest descent and random search', IEEE Trans., AP-24, Sept. 1976, pp.
615-637
67 JENKINS, R.W.: 'Adaptive antenna studies for HF communications'. IEE Conference
Pub. 195. (Antennas and Propagation) Apr. 1981, pp. 190-194
68 HUDSON, J.E.: 'An adaptive antenna with computer assisted convergence'. IEE Conf.
Publ. 169 (Antennas and Propagation) Nov. 1978, pp. 46-50
69 REED, I. S., MALLETT, J. D., and BRENNAN, L. E.: 'Rapid convergence rate in adaptive
mays,9 IEEE Trans., AES-10, Nov. 1974, pp. 853-863
Chapter 14
Radomes
R. H. J. Cary
14.1 Introduction
14.1.1 Definitions
Radomes are defined as electromagnetic windows, consisting of covers or housings
that serve to protect an antenna from damage and environmental conditions, or
modified metallic surfaces such as to permit antennas to radiate. They are required
to have the necessary structural strength, and be so designed as not to exceed some
specified maximum deterioration in the electromagnetic performance of the
antenna under operational conditions.
14.1.3 Constructions
Radomes can be manufactured in a variety of structural forms:
(a) air pressure supported dielectric structure
(b) self supporting dielectric structure
(c) widely spaced metallic or non-metallic frames supporting dielectric windows
(d) (seldon used at present) metal surfaces with dielectric-covered slotted or other
shaped windows.
solid wall
A sandwich
low dielectric constant foam core
A sandwich
low dielectric constant honeycomb core
liiiiiiiiij ihighsandwich
dielectric constant core
C double sandwich
Commonly used constructions, shown in Fig. 14.1, are of the following forms:
(a) A dielectric single-layer wall, either electrically very thin or near multiples of a
half-wavelength thick in the dielectric, such that reflections cancel.
(b) An A-type sandwich consisting of two skins, and electrically very thin or near
multiples of a half-wavelength thick, which are spaced by a lower dielectric con-
stant core of such thickness as to effect substantial cancellation of the skin reflec-
tions.
Radomes 459
(c) A B-type sandwich with outer skins of appropriate dielectric constant and thick-
ness, chosen to match a higher dielectric constant core. This is analogous to optical
blooming, where the outer skins act as quarter-wave transformers.
(d) A C-type sandwich consisting of two A-type sandwiches back to back, forming
five layers, enabling the residual reflections from the individual A sandwiches to be
further cancelled.
(e) More than five layers, with very thin skins and suitable cores, which can give
great strength and wide-band transmission properties for small angles of incidence.
The choice of radome construction will depend on the requirements. A single
'thin' skin may be adequate at low frequencies, but multiples of half-wave thickness
may be necessary to provide sufficient strength at the higher frequencies, where the
sandwich types may be preferred as they give high strength/weight ratio and wide
bandwidths. Metal wires or discs may be included in the construction, e.g. to im-
prove the bandwidth or to match a thinner structure, thereby saving weight.
Ground radomes may be air-pressure supported using inflatable materials, or of
rigid construction, either using solid dielectric, foam dielectric, or sandwich mater-
ials, or using spaced frames with thin dielectric material covers. Airborne and ship
radomes are usually of rigid construction with sandwich or solid dielectric mater-
ials. Underwater radomes are usually of solid materials (Figs. 14.2 and 14.3).
Fig. 14.2 Nose, tail and fin radomes on Hawker Siddley Nimrod aircraft {Courtesy of British
Aerospace)
Dielectric fairings are used to cover suppressed antennas such as slot or notch
antennas energising wings or aircraft tail structures. Radomes are not often em-
ployed with simple unipole or dipole antennas, but rather are used with narrow-
beam microwave antennas, such as scanners on aircraft or large ground-based
antenna systems.
460 Radomes
Transmission loss: resulting from the attenuation in the radome wall, reflections
from surfaces, both internal and external diffraction and refraction effects, and
polarisation shift.
Fig. 14.3 Domed cylindrical radomes on HMS Bristol {Courtesy of HM Royal Navy)
The result can be a loss in range performance, reduced bandwidth and tracking
accuracy, an increase in noise temperature and a greater susceptibility to interfer-
ence. This degradation has to be set against possible advantages. With airborne sys-
tems, a radome is often essential to maintain the aerodynamic shape of the vehicle.
With ground and shipborne antennas it has to be established in each case whether or
not a radome is overall an advantage.
The following advantages may be claimed:
Operational advantages
(a) The antenna is not affected by weather, resulting in all-weather operation. How-
ever, the radome itself may be affected by weather.
Radomes 461
(b) The antenna, being enclosed, requires less-frequent and simpler maintenance.
(c) The overall system may be more reliable and accurate.
(d) The working area for personnel is more congenial.
Structural advantages
(a) The antenna structural loads are reduced, due to the removal of environmental
forces. This permits the antenna structure to be simpler and lighter, requiring less
drive power and improved life for the bearings, motors, etc.
(b) The antenna is less prone to corrosion.
(c) The antenna will have more uniform rotation, easier stopping and starting, due
to removal of the environmental forces.
Economic advantages
(a) A cheaper antenna can be produced, requiring less maintenance, but the
additional costs of the radome must be considered.
(b) Small ground antennas, or very large slow-moving or fixed systems, may on
balance favour operating without radomes. But for others, particularly large anten-
nas with high rotational speeds and those operating at the shorter wavelengths, the
balance may be in favour of a radome, provided its electrical degradation, structural
maintenance problems and reliability and cost are acceptable.
A+B-C-D
Voltage reflection coefficient^ =
A+B+C+D
Radomes 463
COS0
The matrix describing a single layer =
j—sin0 cos0
sin 8
equivalent
transmission
lines
longitudinal
2 n cos 8 2n cos 9 propagation
constant
effective
zocos e cos 8 impedance
for Ell
effective
ZQ sec 9 Zosecoc sec 9 impedance
for E l
Fig. 14.4 Transmission through a lossless dielectric
14.3.3 Transmission of an A-sandwich construction
Sandwich constructions comprise a multiplicity of layers of materials of different
dielectric constants. The three-layer A-sandwich with its low dielectric constant core,
has found wide application.1'5 A typical construction is with resin glass-fibre skins
of dielectric constant near 4 and loss tangent 0.015, and core material, such as
honeycomb or foam, with a dielectric constant near 1.15 and loss tangent 0.002.
The transmission characteristics of such a sandwich with skin thicknesses of 0.03
thickness/wavelength ratio, and with variable core thickness is shown in Fig. 14.10,
for perpendicular and for parallel polarisation. The skin thickness chosen is equiva-
lent to lmm. at I band (10 GHz) and forms a lighter structure than the alternative
half-wave thickness of a solid resin glass-fibre construction of dielectric constant 4,
which from Figs. 14.4 and 14.5 would be near 9mm {02S\^x). In consequence,
the weight of the sandwich is about a third of that of the solid single layer. The
Radomes 465
curves indicate reasonable transmission at low angles of incidence with both polar-
isations, but at the higher angles of incidence the transmission is more sharply
tuned, particularly for perpendicular polarisation. Optimum transmission occurs at
widely different core thicknesses for the polarisations. The phase delay through an
A-sandwich is shown in Fig. 14.11. At high angles of incidence there can be large
466 Radornes
100
90
80
c
70
d
Io 60
Q.
cn
d
50
40
30 1 1
005 0-10 0-15 0-20 0-25 0-30
thickness to free-space wavelength ratio
Fig. 14.5B Parallel-polarisation percentage power transmission through dielectric sheet;
Dielectric constant = 4; loss tangent — 0.015.
100 &
100
90
80
70
5
1
60
50
30
20 JL _L J_ _L _L
005 0 10 0 15 020
thickness to free-space wavelength ratio
Fig. 14.7 Parallel-polarisation percentage power transmission through dielectric sheet: dielec-
tric constant = 9; loss tangent = 0.0006.
A-sandwiches have zero reflections, and also at the angle where the reflection for
each A-sandwich is of opposite phase. If the individual A-sandwiches give maximum
transmission at one angle of incidence and the combined sandwich at another, then
wide-band coverage over a reasonable range of incidence angles can be obtained. In
consequence, the C-sandwich has found application on high-incidence streamlined
airborne radomes such as that shown in Fig. 14.14. Typical transmission and phase
curves are shown in Figs. 14.15, 14.16 and 14.17 where it may be seen that for a
core of 0.2 thickness/wavelength, high transmission out to high angles of incidence
Radomes 469
is obtained for both polarisations. However, the phase difference of some 40° is
likely to produce significant aberrations.
Sandwich constructions consisting of more than five layers have not been widely
used. Calculations shown that by using thin skin layers and low dielectric constant
cores, useful performance can be obtained, particularly for low angles of incidence,
over very wide bandwidths.
0 01 02
thickness to free-space wavelength ratio
200
80"
0 0 1 0 2 0 3
thickness to f r e e - s p a c e wavelength ratio
Fig. 14.8 Phase delay versus thickness of dielectric sheet at angles of incidence: Dielectric
constant = 4; loss tangent —0.015
The input data for any radome design problem is likely to be: the position, size
scan angles and electrical characteristics of the antenna, together with the environ-
mental conditions and loads which the radome will encounter. The desired external
shape of the radome may also be specified, based on aerodynamical considerations,
with the possibility of some minor modification. The initial geometry determines
470 Radomes
perpendicular polarisation
200
80 c
en
<h
parallel polarisation
100
a
.c
Q.
_L _L
005 010 0 15 0 2
thickness to free - space w a v e l e n g t h ratio
Fig. 14.9 Phase delay versus thickness of dielectric sheet at angles of incidence: Dielectric
constant — 9; loss tangent = 0.0006
Radomes 471
0-1 02 0-3 CU 06
core thickness to f r e e - s p a c e wavelength r a t i o
Fig. 14.10 Transmission through A-sandwich with skins of dielectric constant = 4; loss tan-
gent — 0.015 and thickness/wavelength ratio = 0.03; core dielectric constant =
1.15; loss tangent = 0.002.
At various angles of incidence
polarisation perpendicular
polarisation parallel
472 Radomes
200 i
perpendicular polarisation
100-
parallel polarisation
S*
50-
0°
i—
0-1 0-2 0-3 0-4 0-5 0-6
core thickness to f r e e - s p a c e wavelength ratio
Fig. 14.11 Phase delay versus core thickness of A-sandwich with skins of dielectric constant
— 4, loss tangent — 0.015 and thickness wavelength ratio — 0.03; core dielectric
constant = 1.15; loss tangent - 0.002. At various angles of incidence
Radomes 473
the approximate range of angles of incidence. A study of the best choice of mater-
ials, constructional techniques and dimensions that meet the structural and environ-
mental requirements having been made, the resultant electrical performance, trans-
mission loss aberration etc. can be assessed from the transmission equations. By
studying the anticipated performance over a frequency band, dimensions can be
optimised for the particular system bandwidth. It is unlikely that the first optimisa-
tion will be the best solution, and it is usually necessary to vary the radome wall
thickness longitudinally and transversely to refine the electrical design. The best
solution for transmission does not necessarily give the least aberration, and, accord-
ing to the required specification, a compromise in design may be necessary.
20 / \
o
Q_
/ \
c
o
60
Ku
30 GHz
Ka
L35
10 15 20 25
Fig. 14.14 Rotating high-incidence C-sandwich AWACS radome (Courtesy of The Boeing
Aerospace Company)
Radomes 475
1001-
90
80
u 70
<u
o
Q.
.60
d
50
J_ JL _L _L
01 02 03 0-4 0-5 06
core thickness to free-space wavelength ratio
Fig. 14.16 Parallel-polarisation percentage power transmission of C-sandwich, with skin di-
electric constant = 4; loss tangent = 0.015, thickness/wavelength ratio = 0.03
outer, 0.06 centre, and core dielectric constant = /. 15, loss tangent = 0.002; at
various angles of incidence
The main materials for air supported radomes are rubberised fabrics, or more
generally Hypalon coated Dacron with a dielectric constant near 2.9 and loss tan-
gent of about 0.008.
Ground, ship and airborne requirements for rigid radomes are largely met by
resin glass-fibre laminates, or sandwiches with core materials of resin fibres honey-
combs or foams. Small radomes may use other dielectric materials such as PTFE,
polyethersulphone etc., with or without fibres. Space frame constructions have
Radomes 477
0-1 02 03 OA 0-5
core thickness to free-space wavelength ratio
150 80c
parallel polarisation
O)
(V
'.100
(V
in
d
CL
50
J_ JL _L
01 02 03 04 05 06
core thickness to free-space wavelength ratio
Fig. 14.17 Phase delay versus core thickness of C sandwich with skin dielectric constant = 4,
loss tangent = 0.015, and thickness to wavelength ratio = 0.03 outer, 0.06 centre,
and core dielectric constant — 1.15, loss tangent, 0.002; at various angles of
incidence
478 Radomes
used resin glass-fibre or metal for the frames, supporting resin glass-fibre laminates
or sandwiches, or skins of organic materials. Missile radomes may call for alumina
or glass ceramics.
materials electrical
environment construction calculation
input data radome shape thickness of of radome
antenna data radome performance
speed 8> flight
profile if applicable stress
loads
ok
not suitable
change change
material material
thickness thickness
The fibre yarns can be either knitted, woven or filament wound, with E glass
being the most common, though D glass, quartz and some organic fibres have occas-
sional application where special properties are required. The fibre material is usually
combined with resin by one of the manufacturing methods listed in Table 14.2.
The resultant dielectric constant change with resin loading of a polyester E glass
laminate is shown in Fig. 14.19, and is governed closely by the following empirical
equation:
VR VF
Loge = — — log eF
Y rp Y HP
Radomes 479
where e = resultant dielectric constant
eR = resin dielectric constant
eF = reinforcement dielectric constant
VR = volume of resin
VF = volume of reinforcement
= Vp+ VR+ V, where V is the volume occupied by voids.
The most commonly used resins are polyester and epoxy, with typical dielectric
constants of about 2.7 and loss tangents of 0.013 at 10 GHz, changing with tem-
perature as shown in Fig. 14.20. When combined with glass or quartz, the overall
480 Radomes
dielectric constants are about 4 and 3, with loss tangents of 0.014 and 0.004, res-
pectively. The parameters of a polyester glass composite vary with frequency as
illustrated in Fig. 14.21, and with temperature as in Fig. 14.22<z. A modified poly-
ester8 of lower loss tangent has the property of forming a glass-fibre laminate whose
6i
5-
C
O
o
u 3-
u
<V
2 -
1 -
0 20 40 60 80 100
Fig. 14.19 Dielectric constant of E glass laminate versus percentage polyester resin
29- -0-04
° 28-
002 2
-001
,£- —
-©-©•
40- •002
•0-015,0
o o
X5^°
3-0- -0-01
•o
25- 0005
0 10 20 30 40
frequency, GHz
Fig. 14.21 Dielectric constant and loss tangent versus frequency (37% polyester SR 17449
resin glass cloth laminate)
482 Radomes
0025
002
-0015
c
O
•0005
600-
500-
100-
0.002, at I band (10 GHz), but increases at high incidence angles. Loss tangents for
Nomex resin cells are usually lower.
The mechanical properties of reinforced plastic honeycombs are shown in Table
14.7 for typical call size, densities and resins. Thy polymide has the useful property
of retaining strength to much higher temperatures than most core materials. (Fig.
14.25).
Radomes 485
Table 14.6 Electrical properties of Hexcel nylon reinforced phenolic honeycombs
3/16 in (0.48 cm) cells
Density, Dielectric constant
lb/ft 3 : g/cm2 Perpendicular polarisation Parallel polarisation
E par. L E perp. L E par. L E perp. L
4.5 0.072 1.11 1.07 1.10-1.13 1.07-1.12
6.0 0.096 1.14 1.09 1.13-1.16 1.08-1.14
9.0 0.144 1.20 1.13 1.19-1.24 1.12-1.19
flexural
500-
£00-
compressive
1 300
200-
100
Fig. 14.24 Strength versus temperature of polyimide laminate (E glass cloth: half hour tem-
perature soak)
14.5.3.2 Foams: A foam core material which has been commonly used is
expanded nitrile ebonite. This has a limited temperature range, softening at 70°C,
and lacks homogeneity. It has been largely superseded by polyurethane, which is
available in various densities and can be accurately controlled to give a close toler-
ance on its dielectric constant. Foams must be of closed-cell construction to avoid
moisture pick up and be homogeneous throughout. This can be achieved when low
486 Radomes
30min.
lOOhr.
30min. HRP
phenolic
N
HRP lOOhr. \ 30min.HRH10
* nomex
I 1*0
20-
0 100 200
temperature, deg C
molecular weight polyois are used with isocyanate, resulting in a high degree of
cross linkage to give a rigid urethane foam. Water is added during the process to
provide the foaming and to control the ultimate density. Dielectric constant, loss
and mechanical strength all increase with density. In general the material strength
falls off rapidly above 100°C.
Radomes 487
9
A development of a high temperature polyurethane foam, PIO, prepared from
TDI and alkyd triallycyaniviate copolymeric material, has a much higher strength,
it retains its rigidity to 200° C, where in general its strength is still as good as normal
polyurethane foam of the same density at 20°C. The material is closed cell and
softens for forming near 220°C, enabling double curvature shapes to be made.
1-20- -0002
1-10- - 0 001
105 -00005
1 01
0 005 01 0 15
density, g/cm
Fig. 14.26 Dielectric constant and loss tangent of polyurethane foam versus density
While polyurethane foams are obtainable in almost the whole range of densities,
those typically used for radomes are from 0.05 g/cm3 (3 lb/ft3), to 0.21 g/cm3
(121b/ft3). The dielectric constant and loss tangent variation with density is shown
in Fig. 26.
A comparison of the plastic yield deflection test for normal and P10 polyure-
thane foams of 0.14 g/cm3, (8 lb/ft3), using ASTM Spec. D648-56, is:
Temperature Deflection P10 foam
Normal foam.
20°C 0 0
100°C 0.125 cm (0.05 in) 0
120°C collapse 0
160°C — 0.05 cm (0.02 in)
220°C — collapse
A comparison of strength versus density for normal and P10 foam is shown in
Fig. 14.27. Fig. 14.28 compares their compressive strengths with temperature varia-
tion.
488 Radomes
Polyimide foams are being developed to extend the temperature range to 300°C.
301
P10 compressive
20-
10-
normal tensile
2 U 6 8 1b/ft 3
i i
0035 007 0-105 0-U g/cm 3
Typically the dielectric constant of 2.78 and density of 1.2 g/cm3 for an epoxy
resin are reduced to 1.9 and 0.6 g/cm3, respectively, by the addition of micro-
balloons. This low density material, with a flexural strength of 30MN/m2, has appli-
cations in wideband and multiband radome design.
30 -i
25-
P10
20-
15-
normal
10-
5-
100 200
temperature^
14.5.4.1 Alumina (A12O$) The conventional resin glass fabrication has limitations
in operating temperature and in withstanding high-speed rain impact. Alumina is
one of the few current materials which overcomes these failings, and in conse-
quence is widely used for vehicles such as missiles of Mach 3 speed and above.
Electrically, its dielectric constant (about 9.6) is well suited for narrow-band
transmission, particularly where minimum aberration is required. Its loss tangent is
extremely low. Whilst its dielectric constant is nearly constant with frequency, it
changes rapidly with temperature, as shown in Fig. 14.29, limiting its performance
to a narrow temperature range. It can be corrected by adding titanates as shown in
Fig. 14.29, where the resultant dielectric constant is higher but is nearly constant.
The specific gravity of dense alumina, about 3.28 g/cm3, generally results in a
heavy radome, through its high dielectric constant helps to reduce the thickness
Table 14.8 Properties of inorganic radome materials
Property Alumina Pyro- Silica Cordierite Silicon nitride Boron nitride Beryl
i
99% ceram Rayceram Oxide
Slip with Dense Reac- Dense Pyro- 99%
cast Woven tion litic
fused Quartz Bonded
Density g/cm3 3.9 2.6 2.2 1.8 2.45 3.2 2.4 2.0 1.25 2.95
1
lb/ft3 244 162 137 112 153 200 150 125 78 185
X/2 Weight Factor
(g/cm3)/Ve 1.26 1.1 1.2 0.96 1.1 1.15 1.02 0.97 0.72 1.25
20°C10GHz
Dielectric constant
10GHz25°C/43°F 9.6 5.65 3.42 3.05 4.85 7.9 5.6 4.5 3.1 6.6
500°C/832°F 10.3 5.8 3.55 3.04 5.05 8.2 5.7 4.6 3.2 7.2
1000°C/1632°F 11.4 6.1 3.8 3.02 5.8 4.78 3.3 8.0
%e change v.
100°C/180°F 1.2 0.5 0.2 0.1 0.8 0.75 0.4 0.6 0.6 1.5
Loss tangent
10GHz25°C/43°F 0.0001 0.0002 0.0004 0.0009 0.002 0.004 0.001 0.0003 0.0003 0.0003
500°C/832°F 0.0005 0.001 0.001 0.001 0.008 0.0045 0.0025 0.0006 '0.0006 0.0005
1000°C/1632°F 0.0014 0.0008 0.0014
Flexural strength
25°CN/m 2 Xl0 6 270 235 44 125 400 100 100 260
500°C 250 200 54 120 400 60 200
1000°C 220 75 66 100
43°Flb/in 2 Xl0 3 40 34 6.3 18 57 25 14 14 35
832°F 37 29 7.8 17 57 9 29
1632°F 35 11 9.5 14
Young's modulus
25°CN/m 2 X 109 380 120 48 18 128 300 70 13 320
500° C 350 120 48 125 300 50 300
1000°C 285 100 120 210
Table 14.8 Continued
Property Alumina Pyro- Silica Cordierite Silicon nitride Boron nitride Beryl
99% ceram Rayceram Oxide
Slip with Dense Reac- Dense Pyro- 99%
cast Woven tion litic
fused Quartz Bonded
Young's modulus
43°Flb/in2X 106 54 17 7 2.5 18 43 15 10 1.75 46
832°F 50 17 7 — 17.6 43 7 42
1632°F 41 14 17 30
Poisson ratio 0.07-
0-800°C 0.28 0.245 0.15 0.18 0.26 0.23 0.34
Coef. thermal
conductivity
BTU.ft.h°F 20 2.2 0.46 2.4 1.4 12 6 14 16 120
cal.cm.s°C 0.09 0.009 0.0019 0.01 0.006 0.05 0.06 0.07 0.5
Coeff. thermal
expansion
10-6in/in/°F 4.5 2.5 0.30 1.3 1.8 1.4 1.8 2.1 4.6
10-6cm/cm/°C 8.1 4.0 0.54 2.3 3.2 3.2 3.8 8.3
Specific heat
cal.gr.°C 0.27 0.2 0.18 0.25 0.18 0.2 0.2 0.3 0.29 0.26
Thermal shock Fair Good Very Very Good Very Very Very Very Very
good good good good good good good
Water absorption 0% 0% 5% 26% 0% - 23% 0% Closed 0% 5
cell
Rain erosion Excel- Very Poor Poor to Very Very Good Very Good Very
!
lent good fair good good good good
492 Radomes
titanate alumina
11-
99H-7o
10-
500 1000
temperature,deg C
Fig. 14.29 Dielectric constant of alumina and titanate alumina versus temperature
14.5.4.2 Glass ceramics Glass ceramics are the other main candidate inorganic
materials for radomes requiring high-temperature capability. The glass ceramic
mainly used is Pyroceram 9606, u a product of the Corning Glass Co., though other
variants are available with various emphases on particular properties.
Radomes 493
1500-
compressive
1000-
300-
flexural
en
S 200
tensile
100-
500 1000
temperature ,deg C
9 5-
90-
8 5-
3 6 3-7 38 39
d
density, g/cm
Fig. 1431 Alumina dielectric constant versus density at room temperature and 9.5 GHz
494 Radomes
.a 5 5-
10 20 30
frequency, GHz
55-
500 1000
temperature.,deg C
Fig. 14.32 Dielectric constant versus frequency at 20° C of Pyroceram 9606 and versus tem-
perature at 9.5 GHz
Mechanically the material is hard, with good rain erosion resistance, and thermal
shock properties. Its specific gravity of 2.62 g/cm3 makes it lighter than alumina,
but for the half-wave thicknesses radome design the net weight is only slightly less
than that of alumina. Some of its mechanical strength properties, Fig. 14.33, are
lower than those of alumina, but on balance, due to the increased physical thick-
ness of the half-wave sheet usually chosen for the design, its properties are
adequate.
A glass ceramic material based on Zn, Al, Si and Ba oxides, called Mexim, has
been fabricated into radomes by GEC (UK) with a dielectric constant varying
according to formulation, but typically near 6 with a loss tangent of 0.001 at
500°C. Mechanically and thermally, the material has properties somewhat similar to
Pyroceram.
1500 -
compressive
1000 -
300
CM
E flexural
2
200
_c
"en
c
<b
100
0 500 1000
temperature, deg °c
u, 3 75 -
325
2 21
density , g/cm 3
Fig. 14.34 Dielectric constant versus density of silica at 9.5 GHz and 20° C
dense
slip cast
3 5-
0 500 1000
temperatureJdeg°C
very dense
t 100H
less dense
0 500 1000
temprature,deg C
max density
25 2-75 30 325
density, g /cms
Fig. 14.37 Dielectric constant versus density of silicon nitride at 9.5 GHz and 20° C
498 Radomes
(2.5g/cm 3 ) it is near 5.5 (Fig. 14.37). It changes considerably less than that of
alumina with temperature (see Fig. 14.38) and the material exhibits low loss. The
mechanical strength changes little with temperature (Fig. 14.39). Rain erosion resis-
tance is good, and improves with increasing density. Thermally, the shock resistance
is excellent, which is the main advantage. The densest material requires heavy
pressures and costly tools for its manufacture, while the reaction bonded silicon
nitride is porous, presenting sealing problems.
9 -
_L _L
500 1000
temperature , deg °c
Fig. 14.38 Dielectric constant versus temperature of silicon nitride at 9.5 GHz
500 -
dense
300
5 200
reaction bonded
100
J_
0 500 1000
temperature, deg °c
a = sphere radius
s = sphere spacing
= 14-16a°
3S 2 L
= disc radius
disc spacing
layer spacing
1-1
volume fraction of
elements in medium
/ length \
\ diameter /
n /f 2 length \
Structural design can be a very complicated procedure and may require many com-
promises to be made before all aerodynamic, electrical, operational and environ-
mental requirements are met. The basic approach, after selecting a preliminary
design, is to determine the loads and evaluate the structural integrity and, if unsatis-
factory, modify the design until an acceptable solution is found. It is a fundamental
requirement that the structure be safe; hence the flow diagram of Fig. 14.18
attempts to find such a safe structure which is closest to meeting the electrical
requirements. Simplified, the structural design may consist of first evaluating the
loads, taking into account the appropriate wind, aerodynamic and blast pressures,
inertia, shock, temperature range stresses, ice, snow etc.; secondly examining the
materials and configurations that might meet the mechanical and electrical require-
ments; and thirdly calculating the dimensions of the materials, and possibly modify-
ing the radome shape to best advantage to satisfy the structural, aerodynamic and
electric requirements.
The non-rigid radome, relying for support on internal air pressure, requires simi-
lar consideration. These radomes are usually ground based and the important loads
which enable the internal air pressure to be determined are those due to wind, ice
and material weight.
other point. The air flow around the sides of the sphere is constricted, with the
streamlines closer together, causing a reduction in pressure. The result is that the
pressure pattern over the body may take the form showin in Fig. 14.41, with both
positive and negative pressures. If the maximum pressure, e.g. at the stagnation
point, is one half the air density times the velocity squared, then the required thick-
ness can be solved from buckling theory.
With a space-frame radome,15 the assembly of beams and panels effectively acts
as a shell, whose effective modulus and thickness is given by:
EA
Ee
_ and te
ft^~
" 467 jnT ~~ v T
\Jn N A
which by substitution into the buckling equation gives:
/nAI
P = 0.225£-
R3
when E — elastic modulus of framework beam
/ = maximum moment of inertia of beam
A = beam cross-sectional area
R = radius
n = number of panels in spherical radome
Hence if the pressure is known the minimum beam cross-section is determined. In
addition to evaluating the general stability of the radome framework, it is necessary
that the individual beams do not buckle under the design loads, and that the
strengths of both the membranes and the beam to membrane attachments are ade-
quate. The whole is best achieved by computer analysis.
to be obtained. The shear force diagram is found from S = f£ Ldx, e.g. at the lm
station back from the nose, the shear load, shown in Fig. 14.43&, is near 8 kN. The
bending moment diagram is similarly determined from BM = f$Sdx, and at l m
station, the bending moment, as shown in Fig. 14.43&, is near 2kNm. These dia-
grams, together with other forces, enable the total stresses to be calculated and the
failure modes determined.
non-uniform pressure
on cross section
The choice of shape, material and construction of the radome available to the
structural designer may be limited by microwave or aerodynamic requirements. For
a nose radome, a procedure may be to consider the radome as a shell, comprising a
series of rings which each have to be strong enough to carry their individual loads.
Radomes 505
r- 2m. up surface
2m along surface
J from nose apex
stagnation pressure
115kN
local aerodynamic
pressure loading, kN
20- m
shear force, kN
10-
T I
0 1 2
distance from nose apex, m
Fig. 14.43A Aerodynamic pressure distribution on nose radome (maximum upload condition)
Fig. 14.45B Pressure distribution and loads on nose radome (maximum upload condition)
506 Radomes
The inertial loads, when integrated, are likely to be greatest at the base of the struc-
ture, and lead to an analysis of the base attachment, for shear and bending
strengths.
To save time, load points, which are considered likely to cause failure are inspec-
ted first. Primary shell failure modes are tensile fracture and compression insta-
bility. Tensile fractures may be simple, a creep rupture, or a fatigue crack. Simple
rupture occurs when applied loads cause excessive tensile stress. Creep rupture is
time dependent and occurs when a material continues to stretch under load until
failure occurs; this tends to happen with resin-bonded materials. Fatigue crack
occurs when loads are applied a number of times, and is more evident with brittle
ceramic materials. Instability failure occurs when the structure collapses under direct
compressive stress, as can happen with thin-walled structures. Local instability can
occur with resin glass-fibre laminates, due to delamination, bond or cure failure,
and with inorganic compounds, such as alumina, due to local gradient stresses.
With advanced radomes the analysis of shell structures becomes very complex
when using realistic loading patterns and constructional details. To cope with this
the radome is often analysed with aid of computer programs, generally using finite
elements17 (Fig. 14.44). Structural elements commonly used are rods, beams, rings,
membranes, conical and cylindrical shells and flat plates. Mixtures of these simple
elements may be used to approximate to structural details. A coarse network of
elements may be used in a preliminary analysis to minimise computation; this will
often suggest changes to the model to improve the stress distribution. The input
data required is:
(a) geometric data giving co-ordinates of connecting points of the structure ele-
ments
(b) boundary conditions of location and type of constraints acting on the network
and the type of load interactions which may occur between elements
(c) stiffness data for the type of element, its cross-sectional dimensions, variations
of properties through the cross-section, and appropriate integrated stiffness para-
meters of cross-section, (moment of inertia of beam, area of rod, etc.)
(d) loading data such as pressure, inertia, temperature variations over and through
the cross-section, concentrated loads, etc.
In spite of the confidence engendered by comprehensive computer-aided predic-
tions of actual stresses, the design is usually verified by practical tests on the
radome under simulated loading and environmental conditions, to confirm the
radome's safety factor. Some commonly performed structure tests are:
(a) instability tests under static and impulsive loads
(b) dynamic tests under harmonic and random vibration and loads.
If the failure mechanism is not easily quantifiable, for example the effects of
rain or hail impact and erosion, high-temperature charring of the surface, thermal
shock resistance, etc., then functional tests should be performed on specimens to
establish likely failure effects.
Radomes 507
14.7 Airborne radomes
Airborne radomes may vary widely in size and shape, may call for near normal
and/or high incidence-angle transmission, and may be used on low- or high-speed
vehicles in a variety of flight and environmental conditions. Transmission data of
the form referred to in Section 14.3., together with structural prediction tech-
niques, are used to find solutions to the requirements. Radome requirements may
include not only adequate transmission, but wide bandwidths, minimum aberration,
low crosspolarisation and low reflection.
Fig. 14.44 Aircraft nose radome finite element model (Reference 11}
Fig. 14.45 Bandwidth improvement of resin laminates with low-loss temperature-stable poly-
ester
(a) with E glass
(b) quartz reinforcement
The sandwich type of radome relies on maximum transmission when each skin is
spaced such that reflections cancel. The bandwidth of the A sandwich can be deter-
mined from a presentation such as that shown in Fig. 14.46#, where constant
power transmission curves for perpendicular polarisation for particular angles of
incidence are shown for varying core and skin thicknesses. Parallel polarisation is
not shown as it usually gives higher transmission up to the Brewster angle.
Operation over angles of incidence 0° to 60°, with greater than 80% transmission, is
obtained along the line 0 to A, (Fig. 14.46a), with a core up to 0.4 wavelengths
thick and corresponding skins up to 0.03 wavelengths thick. If the band was to
cover all wavelengths down to 3 cm (0-10 GHz), the sandwich would have a core of
1.2 cm and skins of 0.9 mm thickness.
Radomes 509
\^50° 70°
S0.3 • \ 70°^^*^
vei
\ ^v. ^^^-
o 70° \ ^^-^
**—^^-22°
— ^ •^^__o°
80.2-
c \ / —
" " ^ "^ —
D
__ 50°
i o.i •
o
rati
/ 3Q2- -
/ \ /
0
0.01 0.02 0.03
skin thickness
wavelength
0.4 I
0.03
skin thickness
wavelength
Fig. 14.46 Angles of incidence curves of 100 and 80% transmission of perpendicular polaris-
ation of (a) lossless A and (b) C sandwiches (a: skin, 4; core, 1.15)
100%
80%
510 Radomes
If a reduced bandwidth of 4-10 GHz was required, but with angles of incidence
up to 70°, the bandwidth would be determined from C to B, (Fig. 14.46a), with a
thickness to minimum wavelength ratio of 0.3 for the core and 0.025 for the skin.
If very high angles of incidence were required, it would be necessary to include a
study of parallel polarisation, which, beyond the Brewster angle, starts to limit the
performance of wideband A sandwiches.
The C sandwich has the advantage that it can have perfect transmission at more
than one angle of incidence, e.g. point P in Fig. \4A6b, where the 100% transmis-
sion curves for 0° and 60° angles of incidence intersect. The curves indicate that
wide-band transmission properties can be obtained, even at high angles of incidence.
e
~
I = angle of incidence
e = dielectric constant
d = distance back in dielectric from interface
X = air wavelength
whilst for parallel polarisation
__ eW/
e — sin2/
The susceptance of a grid of parallel wires embedded in a dielectric sheet where
the electric vector is parallel to the wires, and the polarisation is perpendicular to
the plane of incidence, is:
B —
+ 0.6 I - 1 (e + 2sin2/)
\ A/
S 10
"09
08
0 7
06
05
0U
03
0 2
0 1
01 0 2 0 3 04 05 06 07
wire pitch , wavciength
Fig. 14.47 Susceptance versus pitch of wire grid in dielectric material of dielectric constant
of 1.15
(a) 0.01 X diameter wires
(b) 0 . 0 0 3 \ diameter wires
(c) 0.001 X diameter wires
0° incidence
180°
1180°
Radomes 513
0.1
02 0.3 0.1* 0.5 0.6 0.7
wire pitch,wavelengths
Fig. 14.48 Susceptance versus pitch of wire grid in dielectric material of dielectric constant
of 4
(a) 0.01 X diameter wires
(b) O.OO3X diameter wires
(c) O.OO1X diameter wires
0° incidence
180°
|| 80°
514 Radomes
(i)
c
o (ii)
So
20 30
frequency, GHz
(iii)
-1
•u
10 20 30 50
frequency, GHz
b
Fig. 14.49 Matching and loss of quartz fibre resin laminate (e = 3.15, tan 8 = 0.004, 1.7mm
thick) with wires
(a)
(i) 0° incidence dielectric only
(ii) compensated dielectric with wires
(iii) wires only, parallel to E vector
(b) angles of incidence
(i) 0°
(ii) 45° 1
(iii) 45° ||
Radomes 515
The A and B sandwich can, by choice of the core dielectric constant, give cover-
age at passbands which are not multiples of each other. The A sandwich can give
initial passbands in the ratio of more than two to one and the B less than two to
one. Fig. 14.51 shows several multi-band designs with skins of dielectric constant 4.
A core of e = 2 gives two passbands in the ratio 2.5/1, e = 3 gives 2.2/1 and
e = 6 gives 1.76/1. Increasing e to 10 provides three bands in the ratios of 2.5/1.6/1
(G, I and J bands.)
skin £ = 3.2
coil in core e = 1.1
skin e = 3.2
circumferential
wire matching in core
100
I 80
in
O
60
2f 3f
frequency
1 i
s0.8 •
0.6 -
2f 3f
frequency
Fig. 14.50 Loss and eHipticity versus 4:1 frequency band and over ± 30° antenna scan
0-3 i
5S.2.2
0.2-
OG:OI:OJ
\ 1:10:25
Fig. 14.51 Contours 100% transmission for lossless sandwiches, perpendicular polarisation,
O°, 50°, 70°, angles of incidence.
Angles of incidence: • # 0°; 50°; 70°
Dielectric constant: skin = 4; core = 2, 3, 6, and 10
Ideally, the phase delays for both parallel and perpendicular polarisations, irre-
spective of angle of incidence and wall thickness, should be identical. A study of
the phase-delay curves for dielectric panels, as exemplified in Figs. 14.8 and 14.17,
shows different and irregular phase delays, and hence a radome presenting a variety
of angles of incidences and/or different polarisations and/or operating over a band-
width will inevitably have aberrations.
Radomes 517
73°l 63°1
- I* m radians
aberration
h3
1-5 -i
1-0
AQ centre
wavelength
0-5
fineness ratio —
Fig. 14.53 Aberration slope and antenna size versus radome shape
L/D — fineness ratio
d/K = antenna aperture in wavelengths
R = sum of ± aberration slopes in degrees per degree
C
d
- 3
C
o
JO
15 10 10 15
azimuth scan,deg
80 80 ~
V) 60
<u
O)
CD
a
a a.
JZ
Q.
20 20
40 60 80 40 60 80
angle of incidence, deg angle of incidence, deg
Fig. 14.55A Insertion phase difference of A sandwich with and without wires. Skins: e = 4,
tan 6 = 0.015. Core: e = 1.15, tan 6 = 0.005
a No wires
skins: 0.03X
core: 0.3\
b Wires: 0.008A. diameter
0.5\ spacing
outer skin: 0.03X
inner skin: 0.05\
core: 0.3\
10 10 azimuth scar^deg
90-
80-
70-
60-
r 50-
en
o
x:
QL
30-
20-
10-
-I 1 1 1 1 r- -i 1 1 r~
30 60 90
angle of incidence,deg
Fig. 14.56 Matched phase delays of A sandwich with anisotropic core.
perpendicular
parallel polarisation
thic ness
Skins: e = 4, tan 6 = 0.015, * = o.O3
wavelength
plane of
incidence
cross-polar plane
A N 1 + k2 •
where k = TJT\\ as before.
14.7.4.2 Cross polarisation of a flat single layer and an A sandwich Fig. 14.58a
shows the cross polarisation for a flat single-layer radome of dielectric constant 4,
designed to be a half-wave thick at 0° incidence, for various angles of incidence and
polarisation angles, with linear polarisation incident. The cross polarisation
increases with angle of incidence and is greatest for a polarisation angle of 45°. Also
shown for comparison are the cross polarisation characteristics of an A sandwich
panel with skin and core dielectric constants of 4 and 1.2, and thicknesses of
0.08 cm and 0.9 cm, respectively, at I band. The A sandwich cross polarisation is a
few decibels greater than that of the solid half-wave radome. The behaviour of these
two panels with circular polarisation is shown in Fig. 14.58&. Also shown is the
improved performance obtained by modifying the thickness of the solid panel for
optimum performance at 60° incident angle.
CD
T3
10-
C
o
2
5 20
S
30-
Fig. 14.58 Crosspolarisation of half-wave solid and A sandwich panels at angles of incidence
a Linear polarisation
A sandwich
A./2 solid at 0° incidence
e= 4
b A sandwich
(i) \ / 2 solid at 0° incidence
(ii) \ / 2 solid at 60° incidence
Radomes 527
Section 14.7.4.2., the crosspolarisation at 60° tilt increases to 20 dB for linear and
to 18 dB for circular polarisation.
(c) Conical: Main and cross polarised patterns for a conical half-wave radome with
the antenna on axis are shown in Fig. 14.61a, for linear and circular polarisations.
With the antenna off axis (Fig. 14.616), there is some lOdB increase in cross polar-
isation level. An A sandwich design as in Section 14.7.4.2 gives a general increase of
cross polarisation of 15 dB compared to the single layer half-wave design, as for
instance in the off axis antenna case illustrated in Fig. 14.61c.
50dB
v 3dB contours of
2dB main beam
(d) Ogive and paraboloidal: The diagrams are similar to the conical radome but tend
to give less cross polarisation, due to the averaging effect of different incident
angles.
The half-wave solid radome generally gives less cross polarisation than the A
sandwich, though off-frequency the solid can be worse than the A sandwich.
It is unfortunate that the radome cross polarisation effects are additive to the
'go' and 'return' path through the radome, thus the cross polarisation two-way is
6dB worse than one-way. For rain discrimination of 15 dB overall with circular
polarisation, the radome must introduce less than — 21 dB one-way with a perfect
antenna, and even less if the antenna is not perfect.
solid - radome e = 4
\ 12A diameter
main beam
Since in any practical measurement of ellipticity the radome and the antenna are
usually measured together, the actual performance of the radome can only be deter-
mined if the antenna alone is measured as well. In both cases the angle of the ellipse
as well as major to minor axis ratios must be measured so that the true contribution
from the radome alone can be established.
main beam
centre and outer circles -3 and -20dB levels
/ dB levels of
crosspolarisation
Fig. 14.61 Crosspolarisation values of a half-wave solid and an A sandwich conical radome
with antenna on and off axis, for circular and linear polarisation
a A solid conical radome with antenna on axis
b Half-wave solid radome with antenna off axis
c A sandwich conical radome with antenna off axis
(i) Linear polarisation, 0° scan
(ii) Circular polarisation, 0° scan
530 R adorn es
Reflections from the radome also degrade the antenna pattern and the tracking
accuracy. Typically, a reflected power level less than 1% or 2% is specified.
Amplitude and phase of reflections versus angle of incidence for both parallel
and perpendicular polarisation on various panels are shown in Figs. 14.62, 14.63,
and 14.64. All these designs have been chosen for minimum reflection near 70°
angle of incidence, as might be required for a high-speed nose radome.
22-
20- / -400
18- 1
;
16-
-300
V. 14-
-o
o
Z. 12- T
power r
-200
6-
-100
4-
2-
n
0 20 40 60 80
angle of incidence,deg
The power levels reflected by the half-wave construction are less than 1% near
70° angles of incidence for both polarisations. Both increase at other angles, par-
ticularly for perpendicular polarisation. The A sandwich gives a minimum power
reflected near 70° angle of incidence for perpendicular polarisation, and near 63°
for parallel polarisation (Brewster angle). Its reflections increase considerably more
than those of the half-wave construction at other angles of incidence. The C
Radomes 531
sandwich reflection level, optimised near 70° angle of incidence, gives very low
levels of reflection over a wide range of angles of incidence, though for parallel
polarisation it increases at very high angles of incidence. The order of preference for
minimum reflection is generally the C sandwich, the half-wave solid, and lastly the
A sandwich, but if wide-band operation is required, the order could be altered.
400
300
-200
-100
20 L0 60 80
angle of incidence deg
The reflections are less than 1% for both polarisations from 0° to 80° angle
of incidence.
If the dielectric constant is low, the reflections are smaller at low angles of inci-
dence, but can increase more rapidly at high angles of incidence for parallel polar-
isation, since the Brewster angle is lower. The choice of dielectric constant for mini-
mum reflection is thus dependent on the required range of angles of incidence.
20
1 - 400
16 11 1
- 300
I
i II
£12
1
power
1 - 200
1
8 1 ;
- j- ' --phase ^/
•^ > ^.
11 - 100
i
^ " \
/ y-
11 power
0
~ — • — .
/
T^— I
0 20 40 60 80
angles of incidence , deg
Internal reflections in which rays reflected by the radome remain collimated will
produce subsidiary beams. This can be minimised by shaping the radome to avoid
linear surfaces (Fig. 14.66). Reflection from discontinuities can result in other
secondary radiation, as for instance at the apex of a nose radome.
Radomes 533
14,7,6 Effect of radome hardware (rain erosion cap, pitot systems and lightning
protection)
The foregoing discussion has been concerned with the dielectric transparency
effects on the antenna performance. Rain-erosion protective metallic caps, pitot-
static systems and lightning protection, which may be called for on airborne
radomes, will all contribute additional loss, reflection, aberration, pattern distor-
tion, side-lobe degradation and cross polarisation.
24-
20- 400
16
phase 300
V
12 \\
200
J-
-100
power
60 80
angles of incidence, deg
(a) The rain erosion metallic cap A metal cap is usually placed on the nose of a
resin glass-fibre radome installed on a high-speed aircraft to protect the tip from
rain erosion. The effect of this aperture blockage can be represented by a new
source, similar in amplitude to the main antenna illumination, but in antiphase,
radiating with a pattern dependent on its dimensions. As the cap may be 3 cm in
534 Radomes
50
10 20 30 50
angle from rodome axis.deg
Fig. 14.66 Comparison of polar diagrams of antenna looking along axis of conical and ogive
radome
Radomes of half-wave solid dielectric e = 4, tan 6 = 0.015
£j_ to reflecting surfaces
(b) The pitot-static system When a pitot-static probe is mounted on a radome tip,
the rain erosion cap is not necessary, but since the mounting requires a larger di-
electric or metal support to be built into the radome nose, the aperture blockage is
typically at least 8 cm in diameter. Also due to the probe length, when the antenna
rays are not on the pitot axis, a shadow is produced which increases the blockage.
Radomes 535
The system requires three pipes, usually metallic, for pitot and static pressures and
probe heating, to run the length of the radome, and this causes further blockage.
Typical patterns for an antenna in an ogival radome are shown in Fig. 14.67. Loss
in directivity and increased sidelobes, particularly close to the main beam, are evi-
dent, together with a high reflection lobe from the metallic pipes. Aberration and
cross polarisation effects are greatly increased by this installation.
pitot static
antenna 15 X dia.
50
Fig. 14.67 Effect of pitot installation on polar diagram of antenna looking along axis of
ogive radome (half-wave solid: e = 4, tan 5 = 0.015)
Ei to reflecting surfaces
ogive radome only
with pitot-static system
Moisture absorption by the radome will cause increased loss, since the high di-
electric constant of water will introduce high reflection losses, and the high loss
tangent will increase the aberration. It therefore is important that materials should
be impervious to moisture, and not deteriorate with time such that they become
absorbent.
Moisture accumulation and rain on the radome surfaces can produce similar
losses, which become increasingly serious at the shorter wavelengths, as the thick-
ness of water becomes more comparable to the wavelength. If moisture and rain
538 Radomes
form a film on the surface, Gibble30 has suggested the following formula to deter-
mine the thickness of the film: t = (3uRr/2w)1/3, where u and w are viscosity and
density of water. R is the radome radius, and r is the rainfall rate. Calculations of
the resultant attenuation at microwavelengths indicate somewhat higher values than
Fig. 14.68 Space-frame radome, Onsala, Sweden. Courtesy of Electronic Space Systems Cor-
poration
14.8.1.3 Snow and ice and lightning protection of ground radomes The collec-
tion of snow depends on the shape of the radome. On spherical shapes, snow tends
to collect mainly on the top and lee side. It is often blown off, but, if not, it can be
removed by internal heating or pressure changes. Dry snow is not serious in contri-
buting to loss. For instance at 10 GHz, the dielectric constant is 1.25 and the loss
540 R adorn es
tangent about 0.004. This results in an attenuation of 0.25 dB, and about 5°K
increase in noise temperature, for a layer of 30 cm of dry snow. Wet snow can have
a considerable quantity of water present and tends to stick to most surfaces; and
consequently can result in significant attenuation, unless removed by heaters.
.c 80-
"E
E 60- 16m
2 20-
002 004
effective water thickness, mm
10 20 30
frequency, GHz
10 20 30 40
rain rate,, mm / h r
Ice films formed by freezing rain or moisture will give losses dependent on the
depth. Provided that no water is present, dry ice has a dielectric constant of 3.3 and
loss tangent of near 0.001, at I band. A build up of 1 cm depth, at 8 GHz, is likely
to give a loss of about 1.5 dB, and a noise temperature increase of 17°K. When ice
and wet snow removal is accomplished by internal heating, the surface should per-
mit it to slide off in the least time.
Radomes 541
Often the antenna is on an exposed site subject to lightning, and in this respect
the metallic space frame radome, properly earthed, behaves as a metallic protective
screen to the antenna. A purely dielectric radome can be punctured by lightning
unless some protection is incorporated.
2 5-
20-
1-5-
C
o
10-
0-5-
10 20
rain rate, mm/h
Fig. 14.70 Rain path attenuation versus rain rate and frequency
14.8.3.1 Rigid foam shell radomes Radomes, usually in the form of truncated
spheres, have been constructed of polyurethane foam. Foam panels can be jointed
with epoxy or polyurethane resin, such as to form a homogeneous low dielectric
Radomes 543
constant structure. The resulting radome has good transmission properties with very
wide bandwidths, particularly for the shorter wavelengths. The foam is required to
be moisture proof, and have a surface that permits rain to have low adhesion.
Examples of this type of radome are not common, and appear to be limited in
size.34
14.8.3.2 Rigid single skin radomes Panels, typically of resin glass-fibre laminate
construction with suitable flange attachments so that they can be jointed together,
have been used to construct radomes in numerous shapes, notably truncated
spheres. A limitation with this construction is that the larger the radome, the thicker
the skin has to be for structural reasons, and this limits the bandwidth and conse-
quently the operational versatility. For low reflection levels, the skin thickness
should preferably not exceed a twentieth of a wavelength in the dielectric material.
Reflections can be minimised by making the radome a half-wave thick or multiples
thereof, but only with bandwidth limitations.
14.8.3.3 Rigid sandwich radomes Small sandwich radomes have been made
without joints, but the larger types have usually employed shaped panels with
strengthened edge flanges to permit jointing. Large part-spherical radomes up to
45 m diameter have been constructed in this manner, with high transmission
efficiency, small boresight errors, and good performance up to 6 GHz.35 At higher
frequencies, the electrical performance falls off due to the joints in the panels and
to the increased thickness required for the skin for structural reasons. At the
shorter wavelengths, the quarter-wave thickness of the core of the A sandwich
becomes so thin, e.g. 3 mm at 25 GHz, that it presents structural problems which
generally preclude its usage.
14.8.3.4 Rigid space frame radomes The rigid space frame radome relies on the
frame for structural integrity. The windows are usually non-structural and of
relatively thin dielectric materials. In consequence, the electrical degradation of the
enclosed antenna is mainly determined by the aperture blocking of the frame and
the resulting induced currents, if metallic, and by the transmission and reflection
losses of the dielectric window materials.36
Such radomes are usually of cylindrical, spherical, or part spherical shape, and if
small, the frame structure may be sufficient if it consists of vertical dielectric or
metallic members, equally spaced round the circumference, and jointed together at
the top and base either at a point or to a ring. The effect on the electromagnetic
performance may be determined by model experiments, or by calculation of the
effect of individual frame members on the electromagnetic field and combining this
with the effect of the dielectric membrane windows. The aperture blocking of the
frame members will result in interference effects, introducing losses, sidelobe per-
turbations, and boresight errors, particularly if regularly spaced. With large diameter
radomes, the number of vertical members increases and for structural reasons hori-
zontal tie members have to be spaced round the radome to form a grid type space
544 Radomes
frame. Regular spacing of such horizontal and vertical elements tends to give rise to
a coherence of the scattered fields and maximises the deleterious effects on the
antenna pattern.
A considerable improvement in performance can be achieved by designing the
framework of random geometry such that reflections are incoherent ensuring much
smaller sidelobe perturbations, boresight errors, and an insensitivity to polarisation,
either linear or circular. Often the frame members are of metal rather than dielec-
tric, as they can usually be thinner, giving less aperture blockage. Typical radomes
with frameworks supporting dielectric windows are those of a quasi-random tri-
angulated construction, as shown in Fig. 14.71. These can be assembled on site to
the shape required, usually truncated spheres, taken apart for transport if required,
or lifted on or off the antenna as an entity as shown in Fig. 14.72. Large space
frame radomes have been constructed,15'37 mostly of a truncated spherical shape,
but occassionally toroidal, of up to 48 m diameter.38'39
Fig. 14.72 ESSCO 68 ft-diameter radome being lifted over communication antenna, Hawaii
Courtesy of Electronic Space Systems Corporation
Sangliolo41 emphasises the importance of the transfer of the loads through the
joint, where there is a potential of additional deflection above that which occurs
from the stretching of the skin and the frame, and has shown that increasing the
546 Radomes
joint stiffness and decreasing the membrane modulus will decrease the skin stress. A
nearly threefold reduction in the stress levels of the skin has been achieved with a
new membrane material, (Esscolam VI),15 which has a fourfold decrease in elastic
modulus. The material is non-absorbent and more flexible than conventional resin
glass-fibre laminate types of material and results in improved crack resistance. The
membrane's surface finish is of Tedlar, giving 90% rejection of solar radiation, thereby
helping to keep the enclosed antenna at an even temperature. The outer surface is
hydrophobic such that rain concentrates in beads and runs off quickly in rivulets.
The physical dimensions and dielectric properties of the struts and panels and
their distribution will have a direct effect on the electromagnetic performance of
the radome. A balance has to be struck between the size of the dielectric
membranes and the framework. The panel size should not be so large as to cause
the framework to produce a significant unbalance in the aperture blockage, for, as
the antenna scans this can produce excessive boresight shift, sidelobe deterioration,
and variable transmission coefficients. Neither should the panel size be so small as
to require such a large number of frames as to cause inordinate blockage. The
design of the frames, whatever their size and material, should consider their thick-
ness, shape and jointing so as to achieve the best electromagnetic performance.
A dielectric spaced frame has the advantage of having no fundamental frequency
cut off and gives the additional low-frequency coverage.
However, it has been found that reinforced plastic frames have more variable
characteristics and less suitable stress factors than metal frames and the latter are
invariably preferred for operation at frequencies above their electromagnetic cut
off. This situation could change with the possibility of more suitable dielectric
materials becoming available.
Over a wide range of frequencies, transmission loss will vary due to fundamen-
tally different mechanisms. As the windows of the metal spaced frame structure
become less than half-wave wide they will have a low frequency cut-off. Above this
low frequency cut-off, the transmission performance increases to a maximum, with
the metal framework tending to contribute a constant loss equivalent to that of the
physical blockage. At higher frequencies, the membrane losses become significant,
except at frequencies where the surface reflections cancel. Within the radome pass-
band the combination of the losses of the space frame and the windows are such as
to give, in typical designs, a maximum loss of ldB between 0.3 GHz and 30 GHz.
The loss is attributed to the physical blockage scattering some 5% of the energy,
and to the membrane giving 98% transmission at 0.3 GHz and 85% at 30 GHz. The
membrane loss is due to its thickness, typically 0.75 mm, its dielectric constant 2.8,
and its loss tangent 0.01, and the angle of incidence being near normal. At lower
frequencies, as the space frame approaches cut off (typically near 0.15 GHz) the
loss rises quickly. At higher frequencies, above 30 GHz, the loss fluctuates, as the
membrane becomes a half-wave or an odd quarter-wave thick, giving losses of 0.5
and 2 dB at the first pass and stop bands respectively with increased losses at even
higher frequencies. For minimum losses it is advantageous for the membrane mater-
ial to have a low dielectric constant and loss tangent, and be of minimum thickness.
Radomes 547
The space frame radome tends to give performance independent of polarisation,
due to the angle of incidence being near normal on the near-spherical surface, and
to the triangular frame members having a fairly even distribution of horizontal and
vertical blocking components.
Boresight errors will result from unsymmetrical disturbances of the phase front
of the antenna. They can be minimised by installing the antenna symmetrically
within the radome and for the radome to appear as symmetrical as possible to the
antenna over all scan angles. With the space frame radome, the large number of
metallic members ensures that the energy scattered from the frame structure,
typically 5%, is randomly polarised and distributed over wide angles. The boresight
shift errors are minimal, and with a 20m-diameter radome are found to be typically
less than 0.05 mr over the microwave pass band (3 to 30 GHz).
Sidelobe patterns of an antenna can be perturbed by the scattered energy from
the frame elements and windows of the radome. Minimisation of the perturbations
is achieved by the quasi-random structure and low reflections from the windows,
giving typically 5 to 10% non-coherent energy scattered over wide angles in both
planes. On a high-gain antenna, this scattered energy is likely to be at least at the
— 30dB level. A — 26 dB sidelobe of the antenna would be perturbed to a maxi-
mum of perhaps — 25 dB.
Noise temperature contributed by the radome is directly related to transmission
loss, which over most of the microwave band is not ohmic, but almost entirely
scattered energy. Most of the energy is scattered skywards, either directly or by the
antenna reflector, and only a small fraction reaches the 'hot' earth. Typically a 5°K
increase at 10 GHz is to be expected, decreasing at lower frequencies and increasing
at higher frequencies.
A summary of the main features of typical truncated-sphere metal space frame
radomes of diameter up to 50 m is:1S
Mechanical:
Antenna diam. Radome max. diam. Base diam. Height Weight
5m 7m 6m 6m 800 kg
10m 16m 14m 12m 5400 kg
20 m 33 m 30 m 12m 36000 kg
30 m 48 m 40 m 36 m 90000 kg
40 m 60 m 52 m 45 m 110000 kg
Frame elements: typical length 1.7 m of 6061-T62 aluminium
Membrane material: Tedlar bonded Esscolam VI, 0.95 mm typical thickness
Electrical: (with linear or circular polarisation)
Membrane material: dielectric constant 2.8, loss tangent 0.01
Transmission loss: 0.4 to 30 GHz, less than 1 dB
Sidelobe change: an increase of 1 dB on — 26 dB level
Noise temperature: an increase of 5°K at 10 GHz
548 Radomes
Bore sight shift: Radome diam. lGhz 3 GHz 10 GHz 30 GHz
(maximum) 5m 0.25 0.1 0.05 mr
10m 0.15 0.05 0.03 mr
20 m 0.17 0.05 0.03 0.02 mr
40 m 0.15 0.03 0.02 0.015 mr
-2 1
0.10 i r20 -?
frequency GHz
Fig. 14.73 Metal space frame performance
Radome 68 ft diameter; membrane 0.045 In; and space-frame geometry Essco
7000 series
14.10 Acknowledgements
The author wishes to acknowledge the kind assistance of those who gave permission
to publish photographs and data, and particularly to Dave Conti and Geoff Meades
of the Reinforced and Microwave Plastics Group, Stevenage, of British Aerospace
for their contributions to airborne radomes, and to Albert Cohen, President of
Essco, Concord, Mass., and his colleagues and users of their ground radomes, and to
A. M. Munro of RSRE, Ministry of Defence Procurement Executive, UK, for his
comments.
14.11 References
1 CADY, W., KARELITZ, M., and TURNER, L.: 'Radar scanners and radomes', MIT Radia-
tion Laboratory Series Vol. 26, McGraw-Hill, New York, 1948
2 HANSEN, R. C : 'Microwave scanning antennas', Academic Press, New York, 1964
3 WALTON, J. D.: 'Radome engineering handbook', Marcel Dekker, New York, 1976
4 SKOLNIK, M. L: 'Radar handbook', McGraw-Hill, New York, 1970
5 JASIK, H.: 'Antenna engineering handbook', McGraw-Hill, New York, 1961
6 'State of Radome Technology'. Proc. Symposia on Electromagnetic Windows, Georgia
Inst. of Technology, Atlanta, 1955-78
7 International Conferences on Electromagnetic Windows, Paris, 1967-75
8 MUNRO, A. M., VICKENDEN, B. V. A. et al\ 'Recent advances in radome design'. Con-
ference of Military Microwaves, London Oct. 1978
Radomes 551
9 BRYDON, G. M., and CARY, R. H.: 'Some radome materials with superior dielectric and
temperature characteristics'. First Int. Conf. of Electromagnetic Windows, Paris, 1967
10 EMERSON, X. and Cummings, X.: 'Data Sheet Eecospheres', Canton, Mass., USA
11 WALTON, J. D.: Ref. 3 Chap. 5 'Inorganic Radomes'
12 MOOREFIELD, S. A., and ROGERS, T. R.: 'The development of a lightweight advanced
aircraft radome'. Proc. Tenth Symp. on Electromagnetic Windows, Georgia Inst. of
Tech., 1970
13 CARY, R. H. J.: 'Results of dielectric and metallic loading of radomes'. Proc. Eleventh
Symp. on Electromagnetic Windows, Georgia Tech. 1971
14 WALTON, J. D.: Ref. 3, Chap. 8. FYALL, A. A.: 'Rain erosion - a special radome prob-
lem'
15 'Technical proposal for the Essco metal space frame radome'. Thiird Edition. Essco, Con-
cord, Mass., USA
16 Essco: Ref. 15 p. 27.
17 BERTRAM, H., BINI, P. and COOK, T.: 'Development of MRCA radome'. Third Int.
Conf. on Electromagnetic Windows, Paris, 1975
18 CARY, R. H.: 'Some novel techniques for avoiding obscurations', Radar-77. IEE Confer-
ence Pub.155
19 CARY,R. H.:Ref. 13
20 CARY, R. H.: 'Some design problems of an aircraft multi-band radome', Electromagnetic
Windo Symp. Georgia Tech., 1966
21 PARIS, D. T.: 'Computer aided radome analysis'. Proc. Tenth Symp. on Electromagnetic
Windows, Georgia Tech., 1970
22 DELFOUR, A., DUPUTZ, J. F., and SARREMEJEAN, A.: 'New method of calculation of
aberration'. Third Int. Conf. Electromagnetic Windows, Paris, 1975
23 SELLER, D. C : 'Aberration of dielectric patches and rings'. First Int. COnf. on Electro-
magnetic Windows, 1967
24 MUNRO, A.M., WICKENDEN, B. V. A., and WARD, F. S.: 'Artificial and anisotropic
materials'. Third Int. Conf. on Electromagnetic Windows, Paris, 1975
25 DOWSETT, P.: 'Cross polarisation in radomes: A program for its computation', IEEE
Trans. AES9, p. 42
26 CARY, R. H.: 'Cross polarisation produced by radomes', Aerospace Antennas, IEE Conf.
Pub. 77,1971
27 'Engineers handbook', Micro wave J. Feb. 1968, p. 133
28 CARY, R. H., and CONTI, D. A.: 'The protection of aircraft radomes against lightning
strike'. Proc. 13th Symp. Electromagnetic Windows, Georgia Tech., June 1976
29 CARY, R. H., and CONTI, D. A.: 'Radome obscurations and their equivalent sources',
Proc. 12th Symp. Electromagnetic Windows, Georgia Tech., June 1974
30 GIBBLE, D.: 'Effect of rain on transmission performance of a satellite communication
system'. IEEE Int. Convention Record, Part VI, March 1964
31 HENSEL, S. L., and SMOLSKI, A. P.: 'Performance of Essco metal space frame radomes
during precipitation'. Essco, Concord, Mass., Report MTP-77-0011
32 COHEN, A., and SMOLSKI, A. P.: 'Effect of rain on satellite communication earth
terminals and rigid radomes', Microwave J. Sept. 1966
33 BIRD, W. W.: 'Large air supported radomes for satellite communication ground stations'.
Proc. OSU-RTD Symp. Electromagnetic Windows, June 1964
34 LAVRENCH, W.: 'Polyurethane foam radome tests at C Band'. Nat. Res. Council Canada,
ERB-5 33 July 1960
35 CURTIS, R. B.: 'Survey of ground radomes'. Rome Air Dev. Center RADC-TDR-64-127
36 KAY, A. L.: 'Electrical design of metal space frame radomes', IEEE Trans., AP13, March
1965
37 'Technical data on five types of ground radomes made by Goodyear Aircraft Coy'. GER
10064 Nov. 1960
552 Radomes
38 COHEN, A., DAVIS, P., NILO, S. C, and ORABONA, J. F;: 4A 150 ft metal space
radome', Proc. OSU-WALDC Radome Symp. WADC Tech. Rpt. 57-314
39 WEISS, H. G.: 'The Haystack experimental facility', MIT Lincoln Lab. Tech.Rpt. 365,
Sept. 1965
40 D'AMATO, R.: 'Metal space frame radome design', Int. Symp. Structures Technology for
Large Radio and Radar Telescope Systems, MIT, Oct. 1967.
41 SANGIOLO, J. B., and ROHWER, A. B.: 'Structural design improvements of Essco
radomes and antennas'. IEEE MERS-77, Washington, DC, Nov. 1977
42 KAUFMANN, P., and D'AMATO, R.: 'A Brazilian radio telescope for millimetre wave-
lengths', Sky and Telescope, 45, March 1973
43 MENZEL, D. H;: 'A new radio telescope for Sweden', Sky and Telescope, 52, Oct. 1976
44 ARNY, T., and VALERIANI, G.: 'The Five College radio astronomy observatory', Sky and
Telescope, 53, June 1977
45 ESSCO: Ref. 15, p. 7
Chapter 15
15.1 Introduction
The frequency band of interest in this Chapter, 10 kHz to 3 MHz, extends over a
very large range of wavelengths, from 30 km to 100 m. This means that realisable
structures are small with respect to the radio wavelength at VLF frequencies,
whereas at MF the electrical height of antennas employed can be one-half wave-
length or more. Typical tower heights employed at VLF are 300 m, which at
15 kHz are only 5.4 electrical degrees in height. Equivalent tower heights at 1500
kHz would be only 3 m high. Towers of heights up to 300 m are sometimes
employed at MF; however, since such a tower would be 360° high at 1000 kHz, it
would be sectionalised with each half being fed in phase so that its directivity
would be towards the horizon. Propagation in this frequency range is dominated by
vertical polarisation (ground and skywaves), cf. Belrose11, and so the object in
medium- and long-wave transmission is to radiate vertical polarisation, and to
provide as great a field strength as possible in a horizontal plane.
Since antenna dimensions and frequencies scale in inverse proportion, the
antennas discussed are not unique to the lower frequency bands. In fact some
of the results presented in this Chapter were measured at scale model frequencies
on an antenna pattern range. The same antenna fundamentals apply at any fre-
quency, but with a few important differences. At the higher frequencies, a quarter-
wavelength or a half-wavelength becomes a practical dimension, and at HF, VHF
and UHF monopole antennas can be elevated, remote from the ground beneath.
The radiation pattern for vertical elevation angles below the horizon must therefore
be considered, and the 3 dB gain attributed to ground-based monopole antennas,
because they radiate over a hemisphere rather than a sphere does not apply. Also,
ground loss resistance does not scale directly with frequency. The emphasis in this
Chapter is on vertical antennas, or antennas that radiate dominantly vertical polar-
ization, for use at the lower frequencies, and therefore discussion about ground
losses, radiation efficiency, etc. refer to this application.
554 VLF, LF and MF antennas
15.2 Analysis
. — e)
Fig. 15.1 Current distribution on typical monopole antennas and their images
The sketches shown in Fig. 15.1 (except d) illustrate the antenna as though it
were grounded, and such a radiator is usually described as being a 'grounded-
antenna' to differentiate it from the ungrounded antenna shown in Fig. 15.Id.
VLF, LF and MF antennas 555
Actually, the antenna is not directly grounded, but it is resonated and matched,
and fed against ground through the output impedance of the transmitter. In some
circumstances, the base of the antenna is indeed grounded and the antenna is not
base-fed, but it is fed at some distance above the ground plane. This arrangement
not only eliminates the requirement for a base insulator, but it can alter the current
distribution on the radiator, and hence radiation from it.
mV/m
TYPE OF CURRENT PATTERN AT I km dBd
ANTENNA DISTRIBUTION SHAPE Eo GAIN
ELECTRICALLY
SHORT ANTENNA
"ftySf /I 1 - ^ MO 4 . . .
39
ANTENNA 1 I — ^ « 50'
ATOM*
M A I C UU/VV/CT
ANTENNA CENTRE V \/J 456 6.27
X/Z ABOVE GROUNO
TWO END ON
HALF WAVE 1 ««7
53?
? p.a
7 68
IN PHASE ANTENNAS » ' '
{FRANKLIN ANTENNA)
Fig. 15.2 Current distribution, vertical pattern shape, field strength and gain, on the assump-
tion of sinusoidal distribution of currents and a perfectly conducting ground, for
monopole antennas of various heights. The gain is referred to a dipole antenna
in free space
At VLF and LF, practical antennas are electrically short, and, as we shall discuss
below, to achieve reasonable radiation efficiencies and input impedances, they are
usually top-loaded to produce as nearly as possible uniform current on the radiator.
The radiation patterns for such antennas closely approximate that for a hypotheti-
cal vertical current element; for a lossless electrically short antenna, the field
strength at 1 km for 1 kW of transmitter power is 300 mV/m.
As the frequency is increased, for a vertical antenna having a particular physical
length, the radiation resistance referred to the base of the antenna increases, the
antenna reactance decreases and the radiation efficiency increases. At the lowest
frequency for which the input reactance is zero (this is the fundamental frequency
of the antenna system) the antenna is one-quarter wavelength long or 90° in electri-
cal height. The field strength at the horizon, see Fig. 15.2, is slightly greater over
that for the vertical current element.
556 VLF, LFand MF antennas
These facts are well known, since the publication of two historic papers by
Ballantine6'7 in 1924 led to the present development of vertical antennas for MF
broadcasting. In these papers, it was shown that for vertical antennas higher than
one-quarter wavelength, the radiation resistance referred to the base continued to
rise and went through a very high value when the height of the radiator approxi-
mated to one-half a wavelength. This then pointed to a method of increasing the
radiation efficiency by using antennas having a radiation resistance very large with
respect to the ground loss resistance. Since the antenna pattern is more squashed
towards the horizon, its gain is increased (see Fig. 15.2). The second paper dis-
closed that there was an optimum height for a grounded vertical antenna, a height
of 0.64 wavelength or 231°. The gain of a 0.64A radiator is 6.03 dBd (dB gain
relative to a dipole antenna in free space). This is the maximum gain for a linear
grounded antenna. Increasing the height beyond this optimum value results in a
decrease in the field strength measured at the horizon in front of the antenna, and
an increase in the high-angle lobe. When the antenna is X in height, the low-angle
lobe has decreased to zero, and there is only the high-angle lobe.
The field strengths and gain figures quoted apply to an idealised infinitely thin
wire antenna, along which current propagates with the velocity of light. For practi-
cal antennas, the velocity of propagation is reduced, as is the maximum gain (cf.
Surutka and Popovic90).
Diligent research and experiments have been conducted for other possible cur-
rent distributions (cf. Brown20) that might equal or surpass the optimum design
discussed by Ballantine, but only two types, the half-wave radiator elevated A/4
above the ground plane and the Uniphase antenna, surpass the straight vertical
antenna whose height is about 5X/8. For the elevated half-wave radiator, both the
mutual impedance betweeen the radiator and its image, and the phases between
the direct and ground reflected waves are optimum, so that they combine and
provide a maximum gain (6.27 dBd). The gain of two end-on half-wave in-phase
antennas (a simple Franklin type antenna) is 7.68 dBd, and this radiator has only
a small high-angle lobe. Antennas having greater vertical extent become imprac-
tical for the frequency range of interest. At MF, multi-feed is sometimes employed,
to control more accurately the current distribution on the radiator and hence its
vertical pattern.
determining the effect of the finitely conducting ground on the terminal impedance
of the antenna, and this is especially difficult from a theoretical point of view when
the antenna is fed against a ground screen, the 'connection' to the ground being
through this radial-wire ground system.
The finitely conducting ground also influences the vertical polar diagram, but
this effect is somewhat easier to deduce theoretically. It is the ground many wave-
lengths in front of the antenna, where only the radiation fields are significant (the
electrostatic and induction fields are negligible), that affect the vertical pattern.
Various approaches to the antenna problem are: (a) the Poynting vector method;
(Z?) the induced EMF method; (c) the equivalent transmission line method; and
(d) the numerical moment method of analysis.
In the induced-EMF method, the electric field produced parallel to the antenna
by a known or an assumed current distribution is calculated and used to compute
the distributed EMF along the length of the antenna. By taking the product of this
EMF and the current at every point, and integrating over the length of the antenna,
the complex flow of power out of the antenna is obtained. The complex impedance
at every point along the antenna is also obtained directly. Since the computations
deal with the near fields of the antenna, the method can also be used to calculate
the mutual impedances between antennas (Jordan and Balmain46), or between
re-radiating structures that affect the pattern of the antenna. The method is very
suitable for computer analysis. It is, however, only useful for 'thin' structures.
rs
JL
A simpler, yet practical approach which gives answers with sufficient accuracy
for engineering design, is based on the analogy of an open-circuited transmission
line, see Fig. 15.3. On the assumption of a sinusoidal current distribution, we can
calculate the antenna reactance, and the radiation resistance. If greater accuracy
is required, a closer approximation to the current distribution can be obtained
by assuming a dissipative transmission line, whose losses just account for the
radiation losses. A further refinement is to include some estimated value of dead-
loss resistance, i.e. earth losses etc., in the evaluation of the equivalent dissipative
transmission line. This method is used extensively in this Chapter.
VLF, LF and MF antennas 559
The most rigorous method is based on a numerical moment method of analysis.
The basic approach to the solution of the antenna integral equation it to rewrite
it as a set of linear algebraic equations, in which the unknown quantities are a
suitable expansion of the current distribution. After dividing the conducting
antenna wire or tower into subsections, the unknown current on each is related to
the incident field by an impedance matrix. Inversion of this matrix gives an admit-
tance matrix, that allows the current distribution to be found for any source field.
Although not the first to use matrix solutions, there is little doubt that Harrington's38
unified treatment of matrix methods for field problems marked a turning point in
the direction of antenna analysis (Ramsdale71). Since then, numerical techniques
have been used for a wide variety of antenna problems. Numerical methods used to
calculate some of the results given in this chapter employed NEC (numerical
electromagnetic code), developed by Burke.24'25 The calculations were, in general,
carried out for dipole antennas and re-radiating structures twice their actual size.
Employing the methods of images, the results obtained therefore correspond to
monopole antennas over a perfectly conducting ground plane. The inclusion of a
finitely conducting ground beneath the antenna and its radial ground system has
also been numerically modelled by Burke et al, 26 but their solution is only rigor-
ously correct if the antenna system does not make physical connection with the
ground plane. This will be considered later.
there would be great uncertainty in calculating the radiated power using this
expression at a point on the antenna where the current is small or near zero.
In the case of electrically short antennas, the radiation resistance is
Ga = y - (90°) degrees
However, for Gv < 20°, Rr (Ga = 90°) is within 3% of Rr(max) and at LF and
below, when Ga is slightly greater than 90°, the input impedance is low and
inductive; hence difficult to tune.
11 o—
^ ^
/
§90- —
\
/ \
70-
/
/
50-
/
30-
/
Fig. 15.4 Radiation resistance of a monopole antenna referred to the point on the antenna
where the current is a maximum, as a function of the equivalent electrical height
of the antenna
When the height of a straight vertical radiator is greater than about 0.2X (72°),
a more exact expression, based on the induced EMF method, should be used to
calculate the radiation resistance. In this case Rr (loop), the radiation resistance
referred to the loop current, may be obtained with good accuracy from the approxi-
mation (Jordan and Balmain46)
Zo = 6oAn£-ij (15.9)
where h = height of radiator
a = radius of radiator
In the derivation of this formula, it was assumed that the antenna had a cylindri-
cal cross-section. However, it can be used for structures having other cross-sections
by application of an equivalent radius. Since the replacement of a complex tower
shape by a cylindrical antenna conductor is not an exact one, it is expected that no
precise relation exists between the conductor radius a and a tower of artibrary
cross-section. Hallen,36 Lo60 and others have calculated the equivalent radius of a
polygon by the in variance of the capacitance per unit length, and, for a triangular
and a square tower of side width b, they calculate that a is equal to 0.42b and
0.59&, respectively.
Jaggard43 describes the application of isoperimetric inequalities to calculate
equivalent radii, a method which can be employed for structures of any shape.
Here we bound and approximate the equivalent radius aeq by the following
inequality:
i
where Area is the area of the tower, and Per is the perimeter. Taking the arithmetic
mean of the above bounds yields an approximation to the equivalent radius, which
for the simple polygonal structures that have been analysed is as accurate as the
more exact solution.
Applying this transmission-line analogy, the reactance of an open-circuited
transmission line of characteristic impedance Zo and electrical height Ga is
Xh = -jZS cot Ga ohms (15.11)
With capacitive top-loading of a vertical radiator of electrical height Gv and charac-
teristic impedance Z%9 with a flat-top reactance X at a frequency/, the reactance at
the feedpoint would be (see Laport56)
VLF, LF and MF antennas 563
_ XcotGv+iZSmiGv
Xb Zo (15J2)
- ZZcosGv+jXsinGv
Both X and Gv are functions of frequency, so that computations can be made for
all values of / . At the fundamental frequency f0 of the antenna, Xb = 0, under
which condition the numerator of this equation must be zero. The terminating
reactance for resonance X = Xo is:
Xo (at/o) = -j'ZS tan Gv (15.13)
An alternative and perhaps simpler method of calculating the base reactance is
to equate the top-load reactance X to an equivalent lengthening of the antenna
(0) of characteristic impedance ZJ, given by
X = ZS cot 6 (15.14)
Since X has been calculated, we can solve eqn. 15.14 for 0, and the base reactance
is then
Xb = -j2%cot(Gv + 0) (15.15)
These equations strictly apply for electrically short antennas (h < 0.2X).
For straight vertical antennas more than 0.2X high, the more exact expressions for
low-loss but not loss-less transmission lines must be employed. In the transmission-
line representation, the radiated power is replaced by an equal amount of power
dissipated as ohmic loss along the line. This power loss is assumed to be due to a
series resistance r ohms per unit length, and a shunt conductance g mhos per unit
length. Furthermore, it is assumed that the I2r loss per unit length at a current
maximum is equal to the V^g loss at a voltage maximum. This approach to the
antenna problem was first given by Siegel and Lab us,83 and with slight modification,
the essentials of the method are detailed in Jordan and Balmain.46 Their application
of the method, however, gives incorrect results due in part to the fact that they
employ a formulation for the characteristic impedance of the antenna which is not in
accord with experiment, but in the main to the fact that they incorrectly estimate
the electrical height G of the monopole from the physical dimensions of the tower
height H and effective tower diameter D expressed in electrical degrees. Modified
expressions for the impedance of antennas calculated on the transmission line
method, for h > 0.2A follow:
2 <1516>
coshrT^l-cos Ga
Zn
- fo
cosh2 l - 1 - ^ — ^ j - c o s 2 GQ
564 VLF, LF and MF antennas
Design examples showing the agreement between calculated and measured values
of antenna impedance are given in Sections 15.3 and 15.4.
measurements, labelled data points 2, are given in Fig. 15.5. The other selected
data are those of Morrison and Smith (data points 1), and measurements made by
Feldman (data points 3) for second resonance (given by Schelkunoff,81).
Fig. 15.5 Empirically determined relation between the electrical height of a monopole,
graphed as percentage increase over its physical height, H, as a function of its
electrical diameter, D, and height, G.
The curves for G = 90°, 180° and 270° give the percentage increase of the
electrical height G over the physical height H as a function of the diameter D
measured in electrical degrees. The data for 90° and 270° show little scatter,
whereas, even performing the measurements with great care, the data for 180° have
considerably more scatter.
Fig. 15.6 Mutual coupling between monopole antennas: (a) Antenna configuration; (b)
Equivalent circuit
Z 21 = - ^ (15.18)
I
v2 — Az 2 i + / 2 Z 2 2
If we let r = lx\l2 (following Jordan and Balmain46), where in general r is a com-
plex number, then:
j - = rZ2l+Z22 (15.21)
It is clear that the operating input impedances, Vx\lx and F 2 // 2 , are dependent on
the current ratio r, as well as on the mutual impedance which is a function of
physical parameters of the antenna system.
The mutual impedance can be measured by means of an RF impedance bridge,
by measuring (a) the base impedance of antenna 1 with antenna 2 open-circuited;
and (b) the base impedance of antenna 1 with antenna 2 short-circuited.
The first measurement gives the mesh impedance of antenna 1, since, if/ 2 =
0, in eqn. 15.19, the driving-point impedance is
VLF,LF and MFantennas 567
J~ = Zn (15.22)
r ~ zn -y~ (15.24)
I
I ^22
SPACING
300° 340° 380° 420° 460°
-240°
-260°
-280°
-300°
-320°
-340°
-360°
-260
40 80 120 160 200 240 280 320 360
SPACING BETWEEN ELEMENTS "DEGREES
Fig. 15.7 Mutual impedance in magnitude (\Zm\) and phase (&m) for antennas of equal
electrical height G, as a function of spacing between them (degrees)
VLF, LF and MF antennas 569
for H between 90° to 180° the difference between H and G increases as G increases.
The experimental values in Fig. 15.9 were tabulated by Wright (private communi-
cation) from data measured at many sites, data from broadcast station proof of
performance records on file with the Canadian Department of Communications.
600
2 5 10 20 50
SPACING BETWEEN ELEMENTS - DEGREES
Fig. 15.8 Mutual impedance between closely spaced monopoles of equal height, mutual
reactance Xm and res/stance (graphed as a ratio Rm/Rr)
A final check on the calculation of the base operating impedances of the towers,
in a multi-tower array, is obtained during the adjustment of the antenna system and
measurement of its pattern, by use of an in-line operating bridge. This permits
impedance measurement with the towers fed in the desired manner to provide the
required pattern.
\\J\J— J
/
tzzj
/
(1=150^—• H /
•/
t/> /
rp ID
-y?f
I r e CV oronirK I^A 1
o j|-)cr»ocTirAi
3
2 y
1.0-
50 60 70 80 90 100 110 120
PHYSICAL HEIGHT (H) DEGREES
Fig. 15.9 Measured values of mutual impedance \Zm\, as a function of the physical height, H,
of radiators of equal height at a fixed spacing (150°). The graph shows the error
that results if the theory neglects the thickness of the antenna
Since the losses which occur when the antenna current returns through the
ground system is one of the most important factors that limits radiation efficiency
at VLF and LF, many theoretical and experimental studies have been made to
minimise these losses in practical antenna design. Wait has been one of the most
prolific authors of theoretical papers on the subject; however, Abbott1 and
VLF, LF and MF antennas 571
63 100
Monteath had studied the subject before Wait and Pope carried out their
analysis. Monteath used the electromagnetic compensation theorem, a method
which has since been used by Wait, Anderson, King and others (cf. Teng and
King,93) to study surface fields and radiation patterns of vertical electric dipoles
over radial ground systems laid on finitely conducting ground. A detailed summary
of Wait's work is given by Watt.102 The NEC computer code referenced earlier (see
Burke et a/.,25'26) has incorporated modelling of antennas near the ground. It is a
rigorously correct mathematical model for structures that are entirely above ground,
but very close to ground. For wires that connect to the ground, the NEC code
assumes that the derivative of current along the wire is zero at the connection
point, to obtain a boundary condition necessary to proceed. This assumption is
rigorously correct only for perfect ground, but it is probably reasonable for mono-
poles that are well grounded. The problem is currently under investigation.
y
5
*
S /
X
1 ~—
A
/
1 JIM CREEK
2 BALBOA \ VL F
J ANTENINA<
/ 3
4
RUGBY
250" UMBRELLA )\ \
C AW A
Y
5 500' TRIANGULAR
6 300' T 1AL IF/\X
7 140' SPIDERWEB
10 100 1000
FREQUENCY kHz
Fig. 15.10 Typical values o f ground loss resistance Rg for electrically short antennas, measured
as a function of frequency for various LF and VLF antenna installations
The various curves in Fig. 15.10 show measured values for Rg as a function of fre-
quency for electrically short antennas. These data are considered to be typical of
good installations. The ground loss resistance can be less if the ground screen is
very extensive, or is an ocean; it can easily be much higher for poor installations.
It is seen that Rg varies approximately linearly with frequency.
0.01
GROUND SYSTEM RESISTANCE, R g , OHMS
Fig. 15.11 Ground loss resistance Rg as a function of the total length of wire in the ground
system and frequency for several major VLF stations. The dashed curves are
calculated values
Fig. 15.11 shows measured and calculated values of Rg for some of the major
VLF stations in the world. It is clear that the ground loss resistance varies inversely
with the total length of wire in the ground system, and that hundreds of kilometres
of wire are needed to achieve values of Rg less than 0.1 £1.
The curve in Fig. 15.12 shows values of Rg at MF plotted against the number of
radial wires n in the ground system. In this presentation, Rg varies approximately as
l/n. While Rg is not critically dependent on the length of the radial wires, their
length was between 0.25 and 0.5 wavelength. Values for Rg at MF of 1-2 SI,
referenced to the current loop, are typically quoted in the literature as being
representative of good MF broadcast antenna systems.
Obviously, the length and number of radial wires should be as great as possible,
but practically more than 120 radial wires are seldom used, and for single tower
radiators the length of the radial wire should be at least 2-4 times the height of the
VLF, LF and MF antennas 573
radiator. The VLF antenna systems with very extensive top-loading, the radial wire
system should extend a tower height or so beyond the outer limit of the top-
loading. At MF where two or more towers are used to form a directional array,
each radiator must have its own radial ground system and adjoining systems should
not overlap but are terminated in a common bus. A grid wire system might be a
better arrangement in this case, but since grid wire systems are not used, there are
no examples for comparison.
/
> BELROSE AT 2 8 M H T
/
BROWN ET Al AT ; JMHz
/
/
in
4
y
c
y
y
Fig. 15.12 Ground loss resistance Rg as a function of number of radial wires in the ground
system over sandy soil (frequency about 3 MHz)
A recent study (Tippe,96) has shown that for radiator systems between 3X/8 and
5X/8, the conventional approach at MF of using 120 radials between 0.25X and
0.4X long may be over-dimensioned, and as few as 50 radials 0.2X long may be just
as effective. Monteath106 has also shown that the efficiency of an earth system is
not greatly reduced if only 72 radial wires are used.
Wire size apparently has a negligible effect on the effectiveness of the ground
system and, except for very high power stations, the wire size is chosen for
mechanical strength. Wire of size AWG no. 10 or larger is typically employed. The
wire is usually buried 10-15 cm, or much deeper to permit cultivation of the soil.
When such deep burial is required, the wire should descend to its required depth
on a smooth gentle incline from the tower base (Head40). Usually, bare solid copper
is employed, but insulated wire has also been used; this increases its mechanical
574 VLF, LF and MF antennas
strength and protects it from corrosion. In fact, insulated wire is claimed to provide
better stability with seasonal changes of earth conductivity (Sather,77). Alterna-
tively, a counterpoise is sometimes employed, at least over some radius about the
towers in a multi-tower array, for the pattern stability.
15.2.9 Effect of the finite conductivity of the earth on the vertical antenna
radiation pattern
The finite conductivity of the earth not only affects the input terminal impedance
of antennas, but the ground in front of the antenna has a major effect on the sky-
wave field strength launched at a particular angle of radiation above the horizon.
The vertical pattern shapes, which were sketched in Fig. 15.2, have been calculated
assuming a perfectly conducting flat earth.
In Section 15.2.2, we noted that the unattenuated field strength at 1 km in
front of the antenna for 1 kW of transmitter power was a figure of merit for a
particular antenna system, since this value could be compared with that expected
for an antenna with no losses. This field strength is easily measured, and it is
the field strength that would be used for purposes of predicting ground-wave
propagation. The radial ground system and the ground surrounding it, within a
radial distance of a wavelength or so, are effective in determining the terminal loss
resistance. This is not the situation for launching of skywaves. The reduction
in the skywave field strength because of the imperfectly conducting curved earth
is influenced by the ground well beyond the ground screen; in fact, for radiation
angles less than 5° above the horizon, the ground conductivity fifty or more wave-
lengths in front of the antenna is effective in determining the vertical radiation
pattern.
Thus, while the terminal-loss (ground-loss) resistance limits the total power
radiated, and hence the field strength at a particular radiation angle, the shape of
the vertical pattern is determined by the ground in front of the antenna far beyond
the limits of the antenna and its radial ground system.
While qualitatively the effect of the finitely conducting curved earth on the
vertical pattern of antennas is well known, and has been discussed by several
authors, quantitatively, the theoretical calculation of these patterns is not straight-
forward. Furthermore, the ability of an antenna to launch skywaves at low
elevation angles cannot be measured directly since, by employing a receiver in
an aircraft, the signal received is the total field comprising ground and skywaves.
While the vertical pattern for launch of skywaves can be rigourously calculated
employing a Fresnel reflection-coefficient model for the ground, or the SommerfeId-
Norton integrals (cf. Trueman and Kubina97), for propagation over a flat earth,
there is a discontinuity in the skywave field strength at grazing incidence (see
Fig. 15.13). That is, the skywave field decreases to zero as the radiation angle \p
approaches zero (angle of incidence 6 approaches 90°). Note that the field for
perfect ground does not decrease to zero, but this is only because the ground wave
was included in the calculation of the far field. For a perfect earth, the effect would
monopole r field
point
ZS Televation
ground screen
of 8 radials
I I i 1 1 I
! Fresnel Method X
en
CD 0.4 1 * I CD 0.4 S-N Method _
S- -- .—. Space wave pj U 8
S-
4-* / i^S-N Methoa +->
CO ground wave
(b) -a
0.3 0.3
X
\ (0
X
\
0.2 0.2 - \ _
:ive F- i e l d
CO S-N Method
CD
CD
x / s p a c e wave only
O.I O.I -
AJk 1 1 1 i i j
10 30 50 70 90 0 8 5 86 87 88 89 9 0 i
e degrees e degrees
Fig. 15.13 Calculated elevation pattern for a vertical monopole (H =201°) with eight radials at 28X distance for a ground conductivity of 20
mS/m at a frequency of 860 kHz; (a) Monopole with co-ordinate system; (b) Elevation pattern; (c) Detail of elevation pattern near
grazing incidence
576 VLF, LF and MF antennas
be the same, except that it would be more closely confined to angles near to grazing
incidence. In the real world situation, it is clear that the skywave field is not zero
when \jj = 0.
f=20kc/s
O.l- _PAST0RAL LAND.
< €=I5€ O
cr=5xl6 3 S/m
a = 4 / 3 x 6 3 6 0 km
V200
0.01-
15 10 5 0 -5 -10 -15 -20
ANGLES ABOVE HORIZON, * , (DEC)—•*
LJ
0.01
10 5 0 - 5 -10 -15 -20
ANGLES ABOVE HORIZON,»//, (DEG.) —
Fig. 15.14 Antenna factor F, as a function of frequency and radiation angle \p for: (a)
Pastoral land (o = 5XW-3 S/m); (b) Sandy soil (o = 1 X W'3 S/m)
The real earth is not only finitely conducting but it is curved, and account needs
to be taken of both these factors, especially for frequencies below MF and for
estimating radiation patterns at low angles above the horizon. Wait and Conda101
have considered skywaves incident on a finitely conducting earth; and evoking
reciprocity, their numerical analysis, which includes also atmospheric refraction,
VLF, LF and MF antennas 577
can be used to deduce the effect that the curved earth has on the effective antenna
radiation pattern. The curves in Fig. 15.14, which were deduced by Belrose11 from
their work (see also CCIR Report 265-4, 1978), can be used to take account of the
effect on the antenna pattern of a finitely conducting curved earth. Results are
shown for two ground conductivities, pastoral land (5 x 10~3 S/m) and sandy soil
(1 x 10~3 S/m). The factor F, when multiplied by the field strength that has been
calculated for a perfectly conducting flat earth at the corresponding radiation
angle, gives the expected field strength that would be radiated at that angle. Knight
(see CCIR Report 401-3, 1978) has made similar calculations for LF/MF broadcast
frequencies (0.2-1.5 MHz). The negative angles of radiation apply when waves
diffract around the curvature of the earth, and which are defined as the angular
distance between the receiving antenna and the tangent point for the downcoming
skywave that has been reflected by the ionosphere.
It is clear that the modification to the vertical radiation pattern is that of a
reduction in field strength, and this reduction is greatest for radiation angles near
grazing incidence and for land of poor conductivity. The reduction is greater at
MF than at VLF.
G = J-T (15.23)
K
c
7////////////////7/7/777 77 7 7/
(o) (b)
(
(c) (d)
(f)
Fig. 15.15 Method of feed for various grounded and ungrounded antennas of the T type
(a) to (d); and umbrella types (e) and (f)
Fig. 15.15 illustrates several basic types of top-loaded antennas and also dif-
ferent methods of feeding base-insulated and grounded tower radiators. The arrange-
ment in (a) is the one that is most commonly used. It is a base-loaded T-type
radiator; sometimes the inverted-L type is used. The horizontal arms of the T are
VLF. LF and MF antrnn** 579
suspended between two supporting towers. While the sketch illustrates it as a
single-wire antenna, in practice, the top-loading and the vertical radiator can take
the form of a cage of wires to increase their capacitance and reduce the base termi-
nal reactance of the antenna. Sometimes the top-loading is configured into the
form of a delta suspended between three towers; or it can take the form of a
diamond requiring four support towers.
The sketch in (b) illustrates an antenna type that is seldom used because of the
electrical and mechanical difficulty in installing and tuning a coil at the top of a
tower radiator. Since the coil is tuned so that the base reactance is zero, this arrange-
ment yields the greatest possible current area (for Ga < 90°) on the vertical radiator,
for an antenna of a given physical size, and hence a maximum value of the radiation
resistance. The improvement in radiation efficiency is, however, partly compromised
by the fact that a larger tuning inductance must be used compared with the con-
ventional base-loaded radiator.
The sketch in (c) has been said (Ray,72) to be electrically equivalent to (b), but
with the advantage that the tuning coil is at the base of the antenna. Using the
tower as a transmission line, the antenna is effectively fed at the top, between the
tower and the top-hat terminals. While this can result in an increase in the radiation
resistance over that for the simple base-tuned insulated tower, very heavy top-
loading must be employed. In practice, in the experience of the author, the effective
radiation resistance can be less than that for conventional base-loading. The require-
ment for 'heavy' top-loading can be seen, at least qualitatively, since in (c) if the
top-loading is reduced to zero, we no longer have a radiator; whereas in (a) with
no top-loading the antenna becomes a simple straight base-loaded vertical
monopole.
The arrangement in (d) is an alternative method of feeding a grounded tower; in
this case, the entire antenna is a DC ground. This is advantageous for two reasons:
no base insulator is required to support the tower and the entire antenna is grounded
for protection against lightning. While a capacitance match network is illustrated,
on the assumption that the input impedance is inductive, this is only the case for
very short antennas, as we shall see later when we consider the folded-monopole
antenna. The physical arrangement in (d) is to surround the tower by a cage of
wires, employing a conducting outrigger at the top of the tower and insulated
outriggers at suitable intervals along its length. The feed is between the skirted
end of the cage and ground.
The final two sketches illustrate two arrangements for feeding grounded tower
umbrella-type top-loaded radiators. In the umbrella arrangement, the top-loading
takes the form of a number of active guys, attached to the top of the tower and
insulated at their lower ends. These active guys are arranged symmetrically around
the tower, and while only two are shown, usually 4, 8, 12 or 24 active guys are
employed. The antenna type in (e) is a folded-umbrella-type antenna (cf. Nolan66);
in (J) there are no connections made to the supporting tower. A cage of vertical
wires, shielding the tower from currents flowing on the cage, connects the base-
feed terminal to the umbrella top-hat.
580 VLF, LF and MF antennas
153.2 T-type antenna
The cage T-type radiator, a model of a particular installation, shown in Fig. 15.16,
is a particularly simple antenna to analyse, because it is symmetric in shape and
Fig. 15.16 A model of a cage T-type antenna suspended between two self-supporting
grounded towers
form. In the installation shown, the entire system, see Fig. 15.17, is constructed
from four-wire cages of 6.55 mm-diameter wires spaced 1.22 m from each other.
The flat-top is suspended between two 91.4 m grounded self-supporting steel
VLF,LF and MF antennas 581
towers. The average height of the flat-top is about 82.3 m, the length of the radiator
and of each arm of the flat-top is 76.2 m. Measured values (at full scale) of base
reactance Xh and resistance Rb are shown in the Figure. The resistance values for
frequencies less than 100 kHz have been extrapolated. The curve for radiation
resistance Rr has been calculated, but the values have been verified by measure-
ments at model frequencies and at full scale.
-182.9 m
2000
O.I
70 I00 200 500
FREQUENCY kHz
Fig. 15.17 Base resistance Rb, reactance (—jX^), and radiation resistance (Rr) for a cage T
antenna having the dimensions shown, as a function of frequency
While this antenna has already been analysed by Laport,56 it has been studied by
the author at full scale and by experimental modelling; and so, in the following, we
will carry out a detailed analysis of it to illustrate the application of the various
equations given in Section 15.2. The frequency is 100 kHz and the wavelength is
therefore 3000 m. At 100 kHz, the vertical height Hv of the radiator and the length
of each arm of the flat-top Lh are 9.15°. Assuming 5% lengthening to account for
end effects:
582 VLF, LF and MF antennas
Gv = Lh = 9.15(1.05) = 9.6°
The characteristic impedance of the vertical radiator is
Zg = 60
where a is the effective radius of the element. For an effective radius of 0.72 m,
The characteristic impedance of the flat-top can be calculated from the equation
for the impedance of a conductor of effective radius a at an average height h above
the ground:
= Rr (100) = 0.7(100) = 3
Rr + Rg+Rc 0.7 + 1.0 + 0.53
The antenna Q-factor is
X> 533
Qa
Rt 2.23
The antenna bandwidth is therefore
Since the antenna system is matched to the output impedance of the transmitter,
the operating bandwidth is
BW (operating) = 2 (0.42) = 0.84 kHz
For a transmitter power of 25 kW
/25,000
#» J-T^T = 106 amperes
Now this reactance Xo is one-half of the reactance of one arm of the T at the
junction. Hence
.ZpCQtL,, _^ 0
/ - / z 0 tan Crv
Since the horizontal arm and the height of the radiator are each 76.2 m (adding
5% for end effects),
220ta
180
A = — sin 2.7 = 2.69 degree-amperes
n
and Rr = 0.01215(2.69)2 = 0.09 ohms
VLF, LF and MF antennas 585
Using the measured radiation efficiency of 1.08% for the modelled antenna, the
formula
r? = ^ (100)
gives Rb, which is equal to the sum of the radiation resistance and the ground
loss, as 0.8 ohms.
ZT h<< X
c) d)
Fig. 15.18 Various configurations for a transmission-line radiator (h < \/4). The sketch at (c)
shows a practical method of matching and feeding the radiator
Z o = 60 In — ohms
a
which for the antenna modelled equalled 293 £1.
The Q-factor for a quarter-wave transmission line is
2nf0Z0
Q =
For the antenna modelled Reff = 0.3 O/m, and the antenna BW(calc) at the model
frequency of 7.785 MHz is 48 kHz. The measured BW was 43 kHz.
The impedance of the open end of the transmission line is
2ZO2
Rs =
^~T ohms
This is an important parameter in antenna design, since the voltage at the open end
of the line can be estimated by assuming that the transmitter power is dissipated in
this resistance.
The effect of tapping along a line a distance $X (electrical degrees) is equivalent
to transforming the antenna impedance Rh to the line impedance Rx, and this
impedance step up can be calculated from the transmission-line equation
Rb sin2 (pX)eff
where (pX)eff is used in the calculation instead of the actual electrical distance
PX. This is because, for small PX, (fiX)eff is greater than PX due to coupling between
the driven and shorted (radiating) elements of the transmission line. The relation
between PX and (pX)eff is given in Fig. 15.19 for the modelled antenna. It should
be noted that for PX greater than 5°, (pX)eff is equal to PX.
The curves in Fig. 15.20 give calculated values of the radiation efficiency for a
low-loss (the one modelled), and for a very low-loss (essentially including con-
ductor losses only) transmission-line radiator versus the height of the radiator
(in electrical degrees).
1
DEGREE
«•- \
\
\
=8. \
in \
0.3 0.5 I 2 5 10
fix ( EXPERIMENTAL ) - DEGREES
Fig. 15.19 Empirical relation between fix' (the distance measured from the shorted end in
degrees) to pxefp the value used to calculate the impedance at the feed point,
for a transmission-line radiator of type shown in Fig. 15.18c
Rr = 0.14 ft
Rb = 0.165 ft
If the helix tuning coil factor is 1500, we can calculate the following:
Radiation efficiency = 61%
Antenna g-factor Qa = 273
588 VLF, LF and MF antennas
Antenna bandwidth BW = 55 Hz (Operating BW = 110 Hz)
Base current (for 2 MW input) = 3098 A
Base voltage = 177 kV
Particularly impressive are the reactive power of 624 MVA in air dielectric, and the
real power of 2 MVA delivered to a resistance of 0.21 £2.
20
j
10
/
/ f
/*>
y
A
Yy
AV
LI A
Fig. 15^20
1 I 50
RADIATION EFFICIENCY (%)
100
1000'
2000 •
1
i
I
KDO'
8" 4 -1000' TOWERS
(GROUNDED)
jic
i
* 2250' •j
°000
1000- \ 1
10
—Xr —t—'
X j. J
g 500- \ -5
°X \ R
b//
A
7 200- V /Rr
-2
LJ
O / /\ \
2,oo- ' V-
^—^— \
1 V
% 50 0.5
/ /
['
}
/
20- -0.2
/
10- 1 (__ -O.I
20 50
FREQUENCY kHz
Fig. 15.21 Base res/stance R^, reactance —JXb, and radiation resistance Rr for a diamond
antenna having the dimensions shown, as a function of frequency
Figs. 15.24a, b and c illustrate by the arrows the phasing of currents on umbrella-,
T- and I^type top-loaded antennas. It is clear that in (b) and (c) the currents on
the top-hat do not interfere with currents on the radiator, but this is not the case
for the umbrella antenna. In fact, if the number of umbrella wires is sufficient,
and/or if these wires are connected by a skirt wire which is equivalent to increasing
their number, the top-part of the radiator, under the hat, can be effectively screened,
and radiation is due only to currents flowing on the lower part of the tower (having
a height h — s).
590 VLF, LF and MF antennas
Thus the radiation resistance increases as the length of top-loading increases, but
only to a point, beyond which it decreases. The height for i? r (max) is s/h — 0.43.
Since the antenna reactance decreases continuously with increases in top-loading,
the maximum radiation efficiency can occur for top-loading somewhat longer than
that for the maximum radiation resistance.
15 20 25
FREQUENCY kHz
Fig. 15.22 Base resistance Rb, reactance —jX^ and radiation resistance Rr for the USN
VLF station at NWC, Australia having the dimensions shown, as a function
of frequency (after Patch, private communication)
While there are many parameters that can be varied, in practical design, d should
be as large as is structurally feasible. The design data to follow have been measured
by antenna modelling for djh = 1.4, which is considered to be a practical design.
The experimental modelling measured not only antenna resistance and reactance
for various top-loading arrangements, but transmitter power and field strength
were also measured, with reference to a quarter-wave monopole,and so the radiation
and ground loss resistances were deduced.
VLF, LF and MFantennas 591
Fig. 15.25 shows the amount of top-loading s/h0 needed to resonate an antenna
of height h0 at the operating frequency / 0 . The parameter n is the number of
umbrella wires. The curves show the relation between s/h0 and ho/Xo. For a given
antenna configuration, and s/h0 chosen, the curves in Fig. 15.26 show the mag-
nitude of the radiation resistance Rr at frequency/ 0 . For short antennas shorter
than hQ
where h = height of the antenna and h0 is the height required for resonance.
Fig. 15.23 Model of an umbrella-type antenna with 24 skirted radials (see Fig. 15.29 for
full-scale dimensions)
Naturally, if h is less than h0, the antenna reactance Xa is not zero. The curves
in Fig. 15.27 give values of the antenna reactance Xa for various values of h/hQ
(or /// 0 since ratios of frequencies and heights scale in proportion).
If the antenna is reactive, and if Xa is very much greater than Ra, then the
antenna Q-factor and bandwidth can be calculated as discussed above. However, if
the antenna is operated very near or on its self-resonant frequency, then the antenna
Q-factor cannot be simply calculated. Fig. 15.28 shows how the antenna ^-factor
depends on the parameter s/h0. The (2-factors shown are those for the modelled
antenna where Rg = 0.75 O; hence for an antenna at full scale having a different
loss resistance,
R a (model) \
Qa = Qmo,del
R a (full scale) J
Obviously for s/h0 = 0, the antenna must be quarter-wave resonant and its
radiation resistance Rr = 36 O (see Fig. 15.26), and its Q-factor will be low (see
Fig. 15.28). For s/ho>Q, the antenna now becomes electrically short and Rr
592 VLF, LF and MF antennas
//////////////////////////////////////y
(a) (b)
7777777/7777/7////
(O (d)
Fig. 15.24 Sketches illustrating phasing of currents on umbrella, T- and L-type radiators.
The sketch in (a) illustrates base insulated series feed; that in (d) a grounded
tower folded umbrella antenna
0.7-
0 6
\ \ \
\l6
V
\
1 \ \\ -—-
\\ \ \
2
3T SK RT ED
- s KIR TEC)
0 5
1 >
0 2—
0 I
1N \\
\
k
\
\
Fig. 15.25 Empirical relation between the amount of top-loading (measured by the parame-
ter s/h0 and number of radial wires (n)) required to resonate an umbrella antenna
of height ho/\
VLF, LF and MF antennas 593
decreases and the antenna Q-factor increases. Beyond a certain point it will no
longer be possible to resonate the antenna at the operating frequency and its input
impedance will be reactive as well as resistive.
3 5 7 10
RADIATION RESISTANCE Rr
Fig. 15.26 Empirical relation between the radiation resistance Rr and the amount of top-
loading (measured by the parameter s/hQ and the number of radial wires n)
required to resonate an umbrella antenna
The data given in Fig. 15.29 show measured values for base reactance, base
resistance and radiation resistance for a full-size umbrella antenna having the
dimensions shown. Since the antenna is short, the amount of top-loading has been
increased beyond that for maximum Rr.
To illustrate the use of the graphs suppose we design an antenna for a frequency
of 200 kHz:
/ 0 = 200 kHz
Xo = 1500 m
for s/h0 = 0.43, from Fig. 15.25
hlh0 = 0.1
or h0 = 0.1(1500) = 150 m
594 VLF, LF and MF antennas
According to Fig. 15.26 the radiation resistance is 6.75 £2. If the ground loss
resistance is 2.5 O, the radiation efficiency is
Rr(l00) 6.75(100)
= 73%
Rr+Rg+Rc 6.75 + 2.5
Note that Rc, which is equal to the coil tuning loss, is zero since the antenna is
resonant.
Fig. 15.27 Empirical relation between the base reactance —JXa and the height of the antenna
h/h0 and amount of top loading
The antenna g-factor is given in Fig. 15.28. The Q for the model was equal to
20, but since Rg for the model and Rg for the antenna under consideration are
different
BW = =
77^ 12.3 kHz (operating bandwidth = 24.7 kHz)
lb.2
To illustrate the use of the curves for h<h0, suppose that h = 83 m, then
VLF, LF and MF antennas 595
h 83
= 0.055
1500
for eight umbrella wires, and if s/h0 = 0.43, as before, then ho/Xo = 0.1.
Therefore
/
and according to Fig. 15.27
Xa = --/215 ft
'/
07
/ /
/
•
/
06
Cy
//
1v
04
0 3~'
o ?—
/
/
V
/ / /
n—
15 20 30 40 50 70
ANTENNA Q-FACTOR
Fig. 15.28 Empirical relation between the antenna Q-factor for a resonant umbrella top-
loaded antenna, and the amount of top loading (measured by s/h0 and number
of radial wires n)
Rr = ^] = 6.75(0.55)2 = 2 12
Naturally, we could design the antenna to reduce this reactance by increasing the
number of umbrella wires or their length.
The largest umbrella antenna in the world is that used for OMEGA transmissions
at Tsushima, Japan. The antenna employs a single insulated tubular mast, 450 m
596 VLF, LF and MF antennas
1
= I78' = 25O'
U—d = 35O'—-I
50 100
FREQUENCY kHz
Fig. 15.29 Base res/stance /?&, reactance —jXb and radiation resistance Rr as a function of
frequency for an umbrella antenna having the dimensions shown
top of the tower, but physically placed at ground level. This antenna element,
as described earlier, is effectively fed at the top between the tower and the top-
hat terminals. It is argued that this results in the maximum radiation resistance for
a radiator of a given height. Ma and Fitzgerrell31 have used an antenna of this type
on the roof of a van, and while the application was at HF, the results of their
measurements give further insight into this antenna.
Fig. 15.30 Method of feeding a multiple-tuned diamond antenna supported by four grounded
towers (after Ray,72 1972)
10,000
100 50 20 10 5 2
ANTENNA HEIGHT (H) DEGREES
Fig. 15.31 Base reactance —jXb and res/stance (R^ or RJ, as a function of height of the
monopole (degrees). The solid lines are measured values for a modelled antenna
over a wire-grid ground screen, the calculated reactance values (dots) show agree-
ment between theory and experiment (see text)
on the graph for most of the stations he used, to show the self-consistency of these
data. Since the differences between the two curves on each graph could be brought
into better agreement by shifting one of them by an amount dependent on H, the
differences are probably due to stray effects, such as base insulator capacitance,
capacitance of guy insulators, lighting transformers etc.
MORRISON
//* I
H
AND SMITH V
1
f N\
• 1
00-
' 1
\
Y
j *
r
10-
/
— i- _.
5-
40 60 80 100 120 140 160 180 200 220 240 260
PHYSICAL HEIGHT (H) DEGREES
Fig. 15.32 Measured base resistance R^ as a function of antenna height H (degrees) for
standard broadcast guyed towers of uniform cross-section
In Figs. 15.34 and 15.35 the measurements of Morrison and Smith64 are com-
pared with values calculated by two different formulations of the transmission-
line method. The curves calculated by the Schelkunoff81 method employ his
formula for the characteristic impedance of the tower, which differs from the one
adopted here. The curves labelled Belrose have been calculated by the author,
using the method given in Section 15.2.5. The relation between H and G was not
exactly the same as the one shown in Fig. 15.5, since the curves in this figure were
changed slightly after the calculations shown in the present figures were completed,
but the difference is small.
To illustrate the use of the equations given previously, let us calculate the input
impedance of a 122 m tower, having a square cross-section 2 m wide for 1300 kHz
h/X= 0.528 (H = 190°)
0 = 0.59(2) = 1.18 m
122
= 60 In -1 = 21812
1.18
VLF, LF and MF antennas 601
I000-
500-
6 200-
10-
Fig. 1 5 3 3 Measured base reactance ± jX^ as a function of antenna height H (degrees) for
standard broadcast guyed towers of uniform cross-section
IUUU-
EX PERU
co 5 0 0 - BEL R O S E ^ - f ^
O SCHELKUNOFF^ J / * ^
£
/5 '?\
V
en
us 1 0 0 -
/
\
y
CO J
A
a •*> A
LJ
CO
Fig. 15.34 Measured and calculated base resistance as a function of antenna height H
(degrees) for a standard broadcast tower of uniform cross-section
602 VLF, LF and MF antennas
D = 3.9°
G~ 1.2(190) = 228° (see Fig. 15.5)
From equation 15.16, the radiation resistance referred to the base of the mast is
sinh
218 ' " ~ \\ 218
?ix /
2 u2 /52.5\ 90 ft
cosh — I - (228)
Measured values for Rb (see Fig. 15.32) range from 95 to 140 ohms.
500 —
200-
Fig. 15.35 Measured and calculated base reactance as a function of antenna height H
(degrees) for a standard broadcast tower of uniform cross-section
ELEVATED FEED
Jf
"ELEVATED
FEED
BASE
SOURCE
Fig. 1536 Calculated current distribution on monopole antennas of various heights h/k for
base feed and elevated feed
90° 70"
30*
(o)
h = X/2
0.5
(c) BASE
48.6°
30° -10°
ZZ~~~" EED ^
ELEVWED FEED
0 Q5 10
Fig. 15.37 Calculated elevation patterns for monopole antennas of various height h/k for
base feed (dashed line) and elevated feed (continuous line)
or a parallel conductor type which is shown in Fig. 15.38. This latter type has been
analysed and described in the literature [Uda and Mushiake (see Tai,92), and King
and Harrison.49]
The input impedance of a folded monopole can be calculated from expressions
given by these authors. If the two conductors are the same diameter, then
VLFf LF and MFantennas 605
— RADIUS a
The analysis of Uda and Mushiake gives some insight into the parameters of the
expression for the input impedance of the folded monopole. They consider its
excitation as due to the superposition of two modes: a symmetrical mode charac-
terised by equal driving currents in the two conductors and an asymmetrical mode
characterised by equal and opposite currents in each arm. The former, the radiating
mode, is equivalent to a monopole of effective diameter d. The latter, which does
not lead to radiation, is equivalent to a shorted transmission line of length (h + s/2)
whose impedance is Zt.
When the electrical length of the radiator is approximately a quarter wavelength,
the input impedance of the asymmetrical mode is very large compared with the
monopole impedance;i.e.Zt > AZU and
Hence the input impedance of a A/4 folded monopole is about 144 £1.
When the electrical length h < X/4, the asymmetrical mode dominates, the
radiation resistance is very small and the antenna reactance is inductive. At
606 VLF, LF and MF antennas
frequencies between these limits, as the frequency increases to the value where the
symmetric mode becomes dominant, the reactance changes from inductive to
capacitive, and the input impedance becomes very high. Thus, the folded monopole
does not behave at all like a monopole for h < X/4.
The input impedance of a folded monopole for which s/X = 0.002, d/\ = 0.0005
has been calculated by using the computer antenna modelling code NEC described
in Section 15.2.3. The computed impedance versus h/X is given in Fig. 15.39.
These results show good agreement with values calculated from the expression given
above for ZQ. Thiele et al.9* have similarly compared calculations employing a
computer moment method (Richmond73*74) with those deduced from the trans-
mission-line model.
Fig. 15.39 Calculated input resistance and reactance of a folded monopole antenna as a
function of antenna height, for the case where sfk = 0.002, dfk = 0.0005 and
eA = 0.0025 (see Fig. 15.38)
15.5.1 Introduction
A great deal of what we know today about vertical tower radiators, about section-
alised antennas to better control the current on the radiator, about the use of two
or more towers in directional antenna arrays, and about optimum ground systems
has been developed as a result of diligent research and experiments that have been
verified by extensive application at hundreds of MF broadcast stations. In addition,
VLF, LF and MF antennas 607
there are special requirements for broadcasting in the LF band, where an adequate
bandwidth must be achieved without sacrificing antenna radiation efficiency. And,
with the increased development of the broadcast band, and the higher powers
currently employed, development of directional antenna arrays that have ever
increasing requirements on the depth of the null of the directional array, and on the
stability of the directional pattern, has necessarily led to investigations of re-
radiation from various conducting metal structures that are located nearby and
which could cause a distortion of the directional pattern. In spite of the fact that
the early development of broadcast antennas began in the mid-1920s (Ballantine,6'7)
and that our understanding of the modern broadcast tower radiator had evolved
to its present state by about 1934, research and development is still being carried
out to optimise the development of ever increasing broadcast services within the
limited frequency spectrum. Some of these aspects of antenna design will be
described in this Section.
Economics also influences antenna design. For example, the high cost of tall
towers has resulted in the requirement to diplex two broadcast stations operating
on adjacent frequencies into the same tower radiator. The high cost of land, since
land having good conductivity and flat terrain contours is usually the same land
that is ideal for farming, has resulted in the need to place more than one directional
array, operating on nearby frequencies, on the same antenna site. There are also
special requirements to change the pattern of a particular station from a day to
a night pattern to protect listeners in a nearby broadcast service area. Some of
these more technical details will be discussed in a later Section.
of side width 1.8 m surrounded by a cylindrical cage of radius 1.8 m. If the cage
consists of 32 wires each 0.625 cm diameter, Knight has calculated that 5 = 0.85;
hence the power radiated by the tower is 15 dB below the power radiated by the
cage. Belousov and Kouzmetsov8 used 18 wires in the cage antenna described by
them. Bern used even fewer wires in the cage, and reports a screening factor very
much higher than calculated by Knight. According to a summary of his experiments
(CCIR Doc. 1970-73, Doc. 10/62), he reported that, for a cage of nine wires,
the screening factor was 0.97. While this difference is not resolved here, it is clear
that a cage with relatively few wires will provide satisfactory performance, and
there is, therefore, no fundamental difficulty involved in driving a base-earthed
mast, or in sectionalising a mast without physically doing so. The cage of wires
must be physically bonded together at the top and bottom of the cage by a peripheral
wire. The cage type of radiator has seen greater application because sectionalising
the mast by means of an insulator involves serious mechanical/electrical difficulties
and expense.
A A/4 stub can also be used to sectionalise a mast, because it effectively reflects
a very high impedance into the mast at the place where the stub is open. A quarter-
wave stub can be used to match a A/2 radiator, and effectively isolate its driven end
from the ground beneath it and the transmission line feeding it (the J-antenna);
or it can be used to reduce the re-radiation from unused towers affecting the
directional pattern of broadcast antennas. Both these applications are further
discussed below.
fin Iff
1
Fig. 15.40 illustrates various arrangements for feeding and sectionalising grounded
towers. In (a) the cage is used to feed a A/2 grounded tower (Knight,50). In this cir-
cumstance, the cage (labelled 1) is bonded to the mast (labelled 2) at its top, and the
feed is between the bottom of the cage (3) and ground. The sketch also shows the
equivalent circuit. In the arrangement at (c), an integrated A/4 matching stub is
610 VLF, LF and MF an tennas
Fig. 15.41 Method of feeding and equivalent circuit for a grounded tower cage antenna,
sectionaUsed into three equal lengths
One of the most complex antennas so far utilised is that described by Belousov
and Kouznetsov,8 shown in Fig. 15.41. It is a cage type; the antenna is effectively
sectionalised into three sections, two of which are fed; and a variable reactance
VLF, LF and MF antennas 611
(physically lengths of coaxial cable) permits control of the current on the lower
cage. The upper and central antenna sections may be considered as a symmetrical
dipole with arms H/3 in length. The lower section of height H/3 is an asymmetrical
dipole. The mast height was 321 m. The authors describe the antenna and show
calculated and measured vertical patterns at various frequencies across the MF
broadcast band.
where EK is the magnitude of the field intensity at unit distance in the horizontal
plane produced by the kth tower, fK(6) is the vertical pattern of the kth tower,
and n is the number of towers. The phase factor $K is usually expressed as the sum
of a space phasing, which results because of the location of the kth tower with
respect to a vector reference axis; and a time phasing, which is the electrical phase
angle measured with respect to this same axis, due to the particular phase of the
current on the kth. tower.
The early antenna patterns for 2-tower arrays were hand calculated. Later, in
the 1940s a systemised approach was taken, and hundreds of antenna patterns for
2- and 3-tower arrays were tabulated (Smith,85). These calculations were carried
out on a simple electromechanical calculator.
Nowadays, with the advent of minicomputers, most broadcast engineers have
computer programs which will calculate the resulting field due to a general con-
figuration of any number of towers, each with its own time-phase vector relation,
since the vector sum given by the above relation is basically a very simple cal-
culation. However, in the design of an antenna array to meet certain specific
requirements in regard to its directional pattern, a synthesis approach is needed;
that is, the pattern requirement is known, and what the designer engineer must do
is design an antenna array that will produce the required pattern. Since the numeri-
cal methods for pattern synthesis have not so far been developed, and since it is
not possible by examination of the general vector equation given above to decide
how to proceed, it is at this point that design experience is required. Many direc-
tional antenna arrays have four towers at the corners of a parallelogram. The
chief reason for the popularity of this array is the relative ease with which the
pattern shape can be designed to meet complicated requirements. A 4-tower
parallelogram can be considered as two sets of 2-towers, the resulting pattern is
612 VLF, LF and MF antennas
\
40°
130 140 150 160 170 180 190 200 210 220 230 240 250
AZIMUTH • DEGREES (b)
Fig. 15.42 Measured (dashed line) and theoretical (continuous line) horizontal radiation
pattern, and part of the pattern on an expanded plot, for MF broadcasting station
CHFA (680 kHz), Edmonton, Alberta. The antenna system employs three towers,
equally spaced on a line bearing 10.5° (after CBC, 1976)
614 VLF, LF and MF antennas
Lakeshore
Notes:
Total effective area equivalent to 0.4.A radius
Minimum radial length to be O.25X
Dimensions are approximate
— Limit of ground system
Q 100' diameter elevated counterpoise
Fig. 15.43a Tower configuration and ground system layout of MF broadcasting station
CFTR (680 kHz) at its proposed new site near Grimsby, Ontario. The open
circles are the limits of the 30 m diameter elevated couterpoises
320°
330° 340° 350° (
) 10° 20° 30°
N
40°
<
/
X \
310° 50°
\
/ \
300°
> <
290° 70°
280°
t Ml
270° 90°
250° 240° 230° 220° 200° 180° 160° 140° 130° 120° 110°
35 -|
)H 15 -
u
S' MEDIAN
/ .RESULTANT
FIELD
Fig. 1543b Theoretical horizontal radiation pattern, and part of the pattern on an expanded
plot, for MF broadcasting station CFTR (680kHz), at its proposed new site,
near Frimsby, Ontario. (After Rogers Radio Broadcasting, 1980)
VLF, LF and MF antennas 617
physically large, such as an EHV (500 kV) power line, and is close to the trans-
mitting antenna, even the main beam can be adversely affected.
High-rise buildings, having cross-sectional dimensions of 30 x 30 m, and heights
of 60 m are resonant in the middle of the MF broadcast band. EHV power lines
are particularly troublesome. It just so happens that 500 kV power lines which
employ 54 m towers spaced 274 m, with a 'skywire' for purposes of 'grounding'
the power line for lightning protection connecting the tops of the towers, are
2-wavelength loop-resonant at 860 kHz. That is, the loop consisting of a tower,
the skywire to the adjacent tower, that tower and the image of this half-loop in
the ground has a perimeter of two wavelengths and is resonant. A power line
comprising a long row of evenly spaced towers acts like a series of over-coupled
resonant circuits, absorbing power from the broadcast antenna array and re-radiating
it.
A number of researchers have discussed re-radiation effects from tall TV towers
and other structures, e.g. Presthold,69 Alford and French,4 Flood and Field.32
More detailed attention has been given to measuring the magnitude of these effects
and to developing predictive techniques by Belrose et al.15 and a great deal has
been learned. The understanding of the re-radiation mechanism was first sought by
computation and model measurement of re-radiation by simple structures such as
thin towers and straight transmission lines with uniformly spaced towers. The
numerical analysis employed thin wire models for analysis by prepackaged computer
codes such as NEC (Burke24). The values of key parameters of the model were
formed by a combination of variation studies and theoretical guidelines, such that
agreement was obtained with the scale-model measurements on an outdoor ground-
level antenna pattern range. Both the azimuthal pattern and the frequency depen-
dence were studied (Trueman et al.9*). Also, the effect of detuning stubs on the
various structures to minimise the re-radiation has been studied (Belrose et al.16
Trueman et al.99 Royer76).
Figs. 15.44 and 15.45 show calculated values for the normalised scattering
cross-section for grounded and ungrounded towers of various heights and thickness.
The scattering cross-section (o) is a quantitative measure of the power density at
a given range from the re-radiator, in terms of the power density of the signal
incident upon it. The computer program NEC provides phase as well as amplitude
for the scattered field, but the amplitude gives limits on the magnitude of the
effect, which is sufficient for the purpose of deciding whether a re-radiator will
have a significant effect on the pattern of the broadcast array. Royer76 has shown
that these limits are:
,^V2 /G a (18O>(0)V /2 l 2
G(0)
1/2] 2
where
G a (0) = the gain of the array in the direction 0 when the tower
re-radiator is not present
618 VLF, LF and MF antennas
Fig. 15.44 Normalised scatter cross-section for a grounded tower as a function of tower
height and thickness
And
For the cardioid pattern shown in Fig. 15.46, which is produced by two X/4
monopoles spaced X/4 and with the current in monopole 1 leading that of mono-
pole 2 by 90°, Ga (0) is given by
VLF, LF and MF antennas 619
2
Ga(<p) = 6.56cos {45° [1 - c o s (0 - 150°)]}
Then if X = 300 m (i.e./= 1000 kHz)
h = 70.5 m; h/\ = 0.235
a = 2.4 m; a/X = 0.008
rt = 1500 m
it follows from Fig. 15.43 that
2
a /X = 0.85
Fig. 15.45 Normalised scatter cross-section for an insulated tower as a function of tower
height and thickness
Therefore,
o = 0.85(300)2 = 7.65 x 104 m2
and = ! 75S x
2
2
4TT(1500)
The total gain of the array, in the direction where previously there was a null,
is therefore
620 VLF, LF and MF antennas
t -TOWER
ARRAY WHEN NO
TOWER IS PRESENT
tower (approximately a 5X/8 radiator). The power line runs past the radiator, and
near to it, 1470 ft (448 m) about 1.2A distant. The results in Fig. 15.50 show com-
puted and measured azimuthal patterns (at a scale model frequency) for 515 and
860 kHz. The powerline is approximately resonant at these frequencies. In the
absence of the powerline, the azimuthal pattern would have been a perfect circle.
XY PLANE PERFECTLY
CONDUCTING
TO FAR-FIELD POINT
WHERE CTfl(0,(?)/A2
IS COMPUTED
INCIDENT PLANE
"WAVE
-•X
XL(OHMS)
-H !
— a o » WIRE RADIUS
(a) (b)
a o /A «0.001
*-«--f « t -*
T
0.1
r i
0.3
I
0.5
0.2 h/A
Fig. 15.48 Scatter cross-section for a wire grid building 0.1\ on a side as a function of the
height of the building
(a)
515 kHz
Comparison of
oonputations and
measurements.
(b)
860 kHz
Fig. 15.50 Theoretical and measured (continuous line) azimuthal patterns for an omnidi-
rectional antenna affected by a power line (comprising 9 towers): (a) 515 kHz and
(b) 860 kHz. The physical arrangement is illustrated in Fig. 15.49, and the fre-
quencies correspond to the 1\ and 2\ loop resonant frequencies of the power line.
^—tower
tower
Fig. 15.51 A method of detuning a power line by installation of a stub on the sky wire
624 VLF, LF and MF antennas
strongly, and this re-radiation can distort significantly the pattern of a nearby
broadcasting antenna array. The work to date (numerical and experimental modell-
ing) has considered the case of a perfectly conducting ground plane. The analysis
must be extended to include losses due a finitely conducting real earth. Also, while
initial studies using detuning stubs have shown that the re-radiation can be signif-
icantly reduced, the installation of such stubs on powerline skywires has not yet
been accepted as structurally satisfactory from the power utilities' point of view,
and the installation of detuning stubs on highrise buildings presents even greater
difficulty.
1 1 1 1
10 -
no detuning stub v \
8 IP* ~°~ $ -
measured — - S " " ^ >^
calculated-> / V
6 V -
/
J
/
V
4 / \
/ ybent detuning stub
2 _ 1/1/^-straight detuning stub _
i
750 800 850 900
FREQUENCY in kHz
Sather77 has discussed the use of tuning stubs on the broadcast radiators them-
selves to reduce currents on them due to nearby broadcast stations.
the vertically-polarised field component is a maximum along the axis of the antenna.
Since the efficiency of the radiator is inversely proportional to the ground con-
ductivity, the antenna is only suitable for use in locations of very poor ground con-
ductivity such as frozen arctic tundra, old lava beds and arctic or antarctic ice caps.
Long wire horizontal dipoles have been used in Hawaii (Seeley and Wiborg79) for
VLF frequencies 8-24 kHz, in the antarctic for frequencies 5-14 kHz (Raghuram
et al10) and for a portable LF communication system for use in the Canadian
arctic (Evans30). It is interesting to note that, although the antarctic antenna was
above an ice cap of very low conductivity, nevertheless it was necessary to mount it
on wooden poles to prevent it melting into the ice. For half-wave resonance, the
efficiency of the antenna at Siple, Antarctica, was estimated to be about 2% over
the frequency range 5-14 kHz.
7°
Fig. 1553 VLF/LF trailing wire antenna and counterpoise wire for aircraft application
Evans30 has applied a rather simple expression for the efficiency of dipole on
the ground, an expression given by Weeks and Fenwick,103 which, while not exact,
seems to give agreement with experiment:
144
V =
R. 1-F2 \ cos —
2
where T? = efficiency defined as the power density at the receiving location
when the dipole antenna is transmitting divided by the power den-
sity at the receiver location when a quarter-wave vertical is trans-
mitting
Rs = input series resistance
coe0
Accurate field strength versus distance measurements for a 610 m dipole showed
that o = 0.5 mS/m, the antenna was resonant at 200 kHz, and the measured effi-
ciency was —20.4 dB. The measured parameters to apply the above formula are:
Rs = 154 £1 and F= 0.61. Thus, the calculated theoretical efficiency is - 2 1 . 9 dB
(about 1%). These measurements were made for a dry sandy soil. Arctic tundra
has a similar conductivity (0.3 mS/m), while glacial ice has a conductivity an order
of magnitude smaller. Since antenna efficiency is inversely proportional to con-
ductivity, radiation effiencies of 10% would be expected for this situation.
llo
Fig. 15.54 Electrically short monopoles; (a) Base-loaded; (b) Inductively-loaded; (c) Con-
tinuously loaded
Hence if
Hx = H2 = 0.05(360)/2 = 9°
and d = G2 = 1.05(9) = 9.45°
the inductance Lo for base resonance is
XLQ = /1828 ohms
37
Hansen calculated XLQ — 1823 £1 (as read off graphs given in his paper). Thus
A 2 = 9.41 degree-amperes
Ax = 4.67 degree-amperes
and Rr = 0.01215(4! + A2f = 2.41 £1
If the coil Q-factor were 300, the radiation efficiency for Rg = 0 is calculated to
be 28% (Hansen calculated - 25%).
These operational parameters can be compared with conventional base loading
(Fig. 15.54a). Here
VLF, LF and MF antennas 629
Xa = -J'Z0 cot Ga = - / 9 1 4 ohms
180 / l - c o s G a \
A = 1 = 9.54 degree-amperes
7T \ s inG a /
Rr = 0.01215 A2 = 1.1 ohms
and the radiation efficiency for a coil Q-factor equal to 300 for Rg = 0 is calculated
to be 26%.
Let us now consider the helically loaded antenna, illustrated in Fig. 15.54c.
Kandoian and Sichak47 have analysed in some detail tuned helical antennas. If a
length of wire approximately a quarter-wavelength long is wound onto an insulated
flbreglass whip, the number of turns and the length of the wire can be adjusted
so that the input impedance of the whip is purely resistive. The antenna is then
'quarter-wave' resonant. Clearly, for a helical antenna the velocity of propagation
down it, v, is much less than the velocity of light, c. The radiation resistance can be
calculated approximately from the expression
Rr = l^\ Rr(\/4)
h/\ =
plane, because its physical size is a small fraction of a wavelength, and it is insulated
from the ground beneath by the rubber tyres of the vehicle. Track vehicles are
'grounded' but imperfectly through a sliding contact with the track, and the track
in some cases may be rubber. Metal ships of large physical size provide an adequate
ground plane to feed the antenna against, but wooden or fibreglass ships pose a
problem. Here, unless a counterpoise is employed, the connection to ground can be
through a rather long loop; through the coaxial feed cable and the battery supply
leads to the ship's batteries, which are usually grounded to the engine or to a
ground plate connecting to the water.
Fig. 15.55 shows the current return paths for an electrically short antenna on an
automobile. It is clear that only a part of the base current is due to directly induced
currents in the car; a part of the return current flows in the ground beneath the
automobile and returns to the base of the antenna through the capacitance of the
vehicle to ground. The effective ground plane is therefore larger than the physical
size of the frame of the vehicle.
^mrm77Jm///^^
Fig. 15.55 Return path for currents flowing in the base of a short monopole (whip) installed
on an automobile
There will certainly, however, be current flow in the wires connecting the
counterpoise to the motor or water ground. This author questions the need for a
counter-poise at all, since the antenna could be fed through a grounded metal
pipe, which itself should be inside PVC plastic pipe since there will be RF current
and voltage on it. In other words, the radiator becomes a monopole with elevated
feed. The feeder cable would be routed up the centre of this metal pipe to decouple
it from the radiator as shown in Fig. 15.57.
Fig. 15.56 Method of 'grounding' for a short monopole antenna (not shown) installed on
the roof of the cabin of a fibreglass ship. The sheath of the coaxial feed is con-
nected to the ground-mat counterpoise. The heavy wires labelled (1) are ground
wires connecting to the ship's ground
Fig. 15.57 Proposed method of feeding a mo no pole antenna (elevated feed) installed on a
fibreglass ship. The antenna tuning unit (ATU) would be located adjacent to
and connected to the ship's ground
BASE
INSULATOR
TUNE/MATCH I CAPACITY I ANTENNA
T
ll
rrtn
=nnnr>—=L_
rrm
Fig. 15.58 Equivalent circuit illustrating incorrect (a) and correct method (b) of tuning an
electrically short monopole, where a base insulator feedthrough to the inside of
a metal vehicle or an A TU is employed
VLF, LF and MF antennas 633
If the antenna is electrically short and base-tuned, it is absolutely essential that
the tuning coil be installed at the base of the antenna, at the top of the feedthrough
insulator, for metal vehicles. The reason for this is clear from Fig. 15.58. Since the
feed through capacitance can be 3-15 pF, an appreciable non-radiating current
can flow through it if the tuning coil is inside, whereas if the antenna is resonated
'outside' the metal vehicle, the capacitance of the feedthrough insulator becomes
unimportant.
Centre loading circumvents this problem, since the antenna is resonated, and
hence a low impedance. Also, centre loading offers other advantages. Since the
high-voltage part of the antenna is well above the vehicle (see Fig. 15.59), this
could result in reduced losses over base-loading.
Fig. 15.59 Current and voltage distribution on electrically short monopoles that are base-
loaded (left) and centre-loaded (right)
15.6.6 Ground effects when only a few radial wires are employed
Counterpoises of only a few radials, say less than four, are sometimes used for
transportable applications. The radials are laid on the ground. Fig. 15.12 showed
the effective ground loss resistance for radials from 1 to 120, for a frequency of
about 3 MHz. Counterpoises of three or more radials can be arranged symmetrically
about the electrically short antenna, which exhibits no directional effect. When the
number of radials is reduced to two, the field strength in the plane normal to these
radials is marginally stronger (by 1.5-2 dB at 3.8 MHz, employing radials about
A/4 long) than in the plane containing the radials. When only one radial was used,
its orientation relative to the direction of propagation, had a significant effect on
the transmitted signal strength. When the radial was oriented towards the direction
of propagation, the signal strength was found to be 10 dB higher than when the
radial was directed away from it. Similar directional effects have been observed
for vehicular antennas, where the antenna is placed on one of the 'corners' of the
body of the vehicle.
634 VLF, LF and MF antennas
20°
200° 340°
160*
180°
200°- 340°
320°
Fig. 15.60 Theoretical vertical and azimuthaf radiation patterns for a Beverage antenna at
125 kHz over poor (dry) soil. The length of the antenna is 2.4 km and its height
is 7.62 m
636 VLF, LF and MF antennas
X
130° ' \
140
X
/ L* 90° — J\ r
127°-"
180 v
kW'.V
150 I
r
160'/
i\
70'/ /
0 i
120
i...
100 80
j i_
60 ;\N
40 20 20 40 60 80 100 120
Fig. 15.61 Elevation patterns (measured on an antenna pattern range) for inverted-V dipole
antennas: (a) horizontal polarisation in the plane broadside to the antenna; and
(b) vertical polarisation in the plane of the antenna
Most of the antennas so far described were discussed in terms of radiation efficiencies
which are important for transmitting application. For receiving the requirements
VLF, LF and MF antennas 637
are quite different. Here signal/noise is to be maximised, where the noise is atmos-
pheric, man-made noise and receiver set noise. In the frequency band of interest,
MF and below, the atmospheric-noise field strengths increase with a decrease in
frequency. At VLF, since atmospheric noise levels are relatively high, there is no
need to have a receiving antenna that is physically larger than the optimum dimen-
sions which will provide a signal that is well above the atmospheric and receiver set
noise levels. Although directivity could help, directivity at MF and below is not
normally used for receiver application, other than the figure-of-eight directivity of a
small loop antenna.
Other considerations are the sensitivity of the antenna to man-made noise, which
in the MF band and below is dominantly electrostatic in nature, and the sensitivity
of the antenna to precipitation and snow static. For aircraft applications, precipi-
tation static can be the major limitation.
A small 1 m square loop is considered to be a very desirable antenna (Belrose
et a/.14), since it can be electrostatically screened, and so provide good immunity
to electrostatic noise fields and precipitation static. For point-to-point fixed fre-
quency operation, a remote preamplifier can be employed, designed to match the
tuned loop to a coaxial transmission line. Hence the loop can be considerably
remote from the main receiving equipment. For portable application, a ferrite loop
provides a considerable reduction in physical size for the same sensitivity (Belrose10).
638 VLF, LF and MF antennas
120'
30°
180°
10 20 30 40 50 60 70 80 90 100
(a)
(b)
Fig. 15.63 Radiation patterns (measured on an antenna pattern range) for a half-delta loop:
(a) Elevation patterns for vertical polarisation in the plane of the delta loop;
(b) Azimuthal patterns for vertical polarisation (the delta loop is in the 0-180°
axis direction}. The various curves are for the second, third and fourth harmonics
of the fundamental frequency f0 of the antenna
VLFf LF and MF antennas 639
Very short rod-type active antennas offer an attractive alternative to loops. The
main advantage of the active monopole antenna is that it is very broadband, but,
being broadband, the principal disadvantage is that of intermodulation in the
preamplifier, due to the strong signals in the broad receiving bandwidth.
A great deal of research has been carried out, for example, by Lindenmeier and
Landstorfer,58 Lindenmeier57 and Nordholt and von Willigen,67 to develop short
active receiving antennas that are extremely linear for high incident field strengths,
so as to reduce the problem of intermodulation.
If the active antenna of height ha is mounted on a short grounded metal mast
of height hm, the effective height, which determines signal pick-up, is greatly
increased. For a short active rod antenna, its effective height, when hm = 0, is
Kff (hm = K
If mounted on a mast
hm +
h
eff 2
(hm, = 0 ) KI2
If ha = 1 m,and hm = 2 m,
heff
= 5 (14 dB)
h
eff >=0)
The present state-of-the-art is such that, according to Lindenmeier and Land-
storfer,58 the basic sensitivities for a 21 cm antenna at 1000, 100 and 10 kHz
are about 22, 45 and 150 nV/m\/Hz. Therefore, in order to obtain a 10 dB SNR
in a receiver bandwidth of 3 kHz, the signal strength at 100 kHz must be 10 juV/m
for the 21 cm active rod antenna. For a i m antenna on a 2 m mast, this minimum
field strength is 0.4 /xV/m.
Unfortunately, the maximum tolerable field strength Et decreases in proportion
with the increase in effective height. That is, for a 21 cm active antenna (hm = 0 )
the maximum field strength ^ f = 55 V/m, but for the 1 m antenna on a 2 m mast,
Et = 2.3 V/m. While this is still a fairly high field strength, it must be remembered
that, in actual fact, the rms field strength of all strong signals in the bandwidth
of the active antenna may exceed this level, which is the reason why active antennas
must be used with due caution. Taking nonlinear effects such as intermodulation
and cross-modulation into account, present research has been directed toward the
development of a tunable short active receiving antenna. Hopf and Lindenmeier41
have presented design considerations for an extremely linear selective active antenna
with at least two tuned resonant circuits, tunable within the range 1-30 MHz
within milliseconds. For a 1.2 m active antenna, the second-order intermodulation
at the receive frequency fr is typically — 70 dB if undesired signals at 0.45/ r and
O.55/r, with field strengths of 100V/m each, excite the antenna. Considering
640 VLF, LF and MF antennas
The full potential of VLF and LF transmitting stations can only be attained if the
radiating system is designed to give maximum efficiency. The degree to which this
can be attained, however, is a function of many sometimes conflicting factors such
as land availability, permissible tower heights, bandwidth required, the power of
the transmitter and the capital outlay. The compromises necessary have to be
considered and all factors optimised. The electrical and mechanical structural
design of the antenna, once the antenna type and the site for it is chosen, normally
requires the combined effort of radio, civil and mechanical engineers. The practical
design of such antennas is a specialised field of engineering practised by a few
major companies that supply complete systems. If the antenna design has been
done, companies supplying towers can erect them and string the antenna wire. MF
antenna design is also a field of specialisation, particularly if, as already discussed, a
complicated directional array is required to meet licensing requirements. In this
case, the design and erection contracting, the field testing and the adjustment of the
array, is normally done by broadcast consultants specialising in this particular field.
Obviously, we cannot go into much detail here concerning practical antenna
design, except to make a few general remarks, and illustrate photographically some
of the elements of antennas, as well as of complete systems.
Fig. 15.64a A pair of conical base insulators with spark gap for lightning protection
others are a curved-sided cone and some have corrugated surfaces. High-voltage
and high-power installations are fitted with corona rings and most have rain shields.
The oil-filled types can be fitted with heaters. Base insulators for VLF and LF
antenna installations are manufactured to provide 9 x 103 to 4.5 x 106 kg working
loads for a safety factor of about 3. They are usually provided with a lightning
spark gap and an isolation transformer for tower lighting. Design working voltages
range from about 25 kV to 500 kV.
A typical installation for transmitting powers of 100 kW is shown in Fig. 15.65.
Note the oversized rain shield and lightning spark gap.
Is is clear that, especially for the larger installations, an antenna design which
utilises a grounded tower offers distinct advantages since no base insulator, lightn-
ing spark gap or lighting isolation transformer are required. Clearly, cages about
towers are needed, but strain insulators can be used. The mast can also be used to
support VHF and UHF antennas.
642 VLF, LF and MF antennas
<*tm#tffr' n
Fig. 15.64d Base insulator with rain shield and spark gap for lightning protection
Fig. 15i>4e Base insulator with a ring type isolation transformer for tower lighting
Fig. 15.65 Photograph of the base of a 100 kW LF umbrella type radiator (after Marconi
Communication Systems, Ltd)
and down the centre of the mast, or alongside it where they are attached to counter-
weights to optimise the loading and tension. This arrangement protects the top-
loading, since, under extreme icing conditions, the top-hat will be automatically
lowered. In major VLF installations, these counterweights can be tremendous, up
to 180 x 103 kg. The strain insulators holding the flat top are equally impressive,
646 VLF, LF and MF antennas
F ig. 15.66b Johnn y-ball types of guy strain insulator Fig. 15.66c Compression-type cone guy strain insulator
648 VLF, LF and MF antennas
Fig. 15.67 Static-drain resistors for guyed radio masts (after Lapp Insulator Division, Inter-
pace Corporation)
The configuration of the top-hat wires also requires some ingenuity in the case
of major installations. The arrangement is sometimes configured so that, by use of
strain insulators, it is broken up into insulated loops, each loop having its own
down-leads to provide sleet melting circuits.
Fig. 15.68 Giant antenna insulator for top-loading, manufactured for USN VLF transmitter,
Cutler, Maine (after Lapp Insulator Division, Interpace Corporation)
TUNE/MATCH ANTENNA
b) /TT77
Fig. 15.69 Typical method of tuning and matching VLF/LF antennas: (a) Equivalent circuit
of antenna, base tuning coil and L-section matching unit; (b) actual components
required
650 VLF, LF and MF antennas
Fig. 15.70a Helix tuning coil for 100kW LF radiator (after Marconi Communication Systems
Ltd)
Xx = Xa-XCm = Xa-y/Rt(R0-Rt)
and X2 - XLm - Ro
Ro —Rt
VLF, LF and MF antennas 651
Fig. 15.70b Helix tuning coil for 2000kW VLF radiator (after Continental Electronics
Manufacturing Co.)
The main tuning inductor is usually wound with litz wire. Typical installations
employ antenna tuning coils having diameters from about 1 to 6 m, and employ
litz wire having diameters from about 1 to 9 cm. Fig. 15.70 shows photographs
of 100 kW and 2000 kW helix coils.
To facilitate tuning, a variometer coil is generally inserted in series with the
tuning coil, on the low-voltage side (the end towards the transmitter). Fig. 15.71
shows variometer coils for the power levels noted above.
The shunt inductor is usually wound with copper tubing to facilitate tapping
and matching. It is sometimes silverplated. For power levels up to about 25 kW,
a ferrite transformer is sometimes used to match Ro and Rt, once the antenna has
been resonated.
Fig. 15.71b Variometer for a 2000kW VLF installation (part of the antenna tuning system
illustrated in Fig. 15.70b)
© 26 SWG COPPER
SHEET CLADDING
FEEDER ''
POWER SUPPLIES
IN
LOADING PAD
Fig. 15.72 A typical LF antenna tuning installation shown with the components inside
the tuning hut (after Jink ings,45)
Fig. 15.73 Antenna feedthrough or entrance insulators (after Lapp Insulator Division,
Interface Corporation)
VLF, LF and MF antennas 655
Zaf1 1S8H25
Zaf2 56-M37
Fig. 15.75 Photograph of a fiord mountain supported VLF antenna, Narvik, Norway (after
Continental Electronics Manufacturing Co.)
VLF, LF and MF antennas 657
16.1 Introduction to HF
16.1.1 HF propagation
The HF band as defined by the CCIR extends from 3 to 30 MHz 100 to 10 m
wavelength, but for practical purposes it is the band for which the ionospheric
skywave is the dominant mode of propagation. The useful extent of the HF band at
any given time depends upon the ionospheric conditions, and the limits may extend
beyond the CCIR decade from below 2 MHz to over 40 MHz. In order to maintain
communication over a given path as the ionosphere varies it is necessary to have
available a choice of frequencies spanning most of the band, and 4 to 28 MHz is a
typical practical system bandwidth requirement.
Since the use of ionospheric propagation has an important bearing on the design
of HF antennas, this chapter will begin with a discussion of the ionosphere. The
ionosphere is a sparse region which extends from around 50 km to around 400 km
in altitude. During daylight hours it is bombarded by solar radiation so that the
rarefied air becomes ionised by the ultra-violet and X-ray emissions. Different parts
of the solar spectrum are absorbed at different altitudes, so that several ionised
layers are formed. The upper layers reflect radio waves while the lower layers
attenuate the waves passing through them (Fig. 16.1).
Reflection can be explained by considering the radio refractive index of an
ionised layer. The refractive index JJL of a plasma of ionisation density TV electrons
per cubic metre is given approximately by
• - ! ? 06.1)
where / is the frequency in hertz. For any given density of ionisation the refractive
index decreases as the frequency decreases, and at a frequency/p called the plasma
frequency, where
/ p = SIN (16.2)
664 High Frequency Antennas
the refractive index (JL falls to zero. A radio wave at a frequency below/ p and norm-
ally incident on an ionised layer penetrates only as far as the point where JJL falls to
zero, and is then totally reflected. The plasma frequency at the peak of the layer is
called the critical frequency/ c . Normally-incident waves at frequencies above/ c
pass right through the layer.
v.h.f. transmitted
reflecting layer
At frequencies above the critical frequency, reflection may still occur if the wave
is obliquely incident. The phenomenon is similar to that of total internal reflection
of light in a water surface. The highest frequency which can be totally reflected at
a given angle of incidence 0 is given approximately by
/ = /csec0 (16.3)
and may be several times greater than the critical frequency. This formula is only
approximate because of the curvature of the earth.
The attenuation loss depends mainly on the product of the ionisation density
and the frequency of collisions between the free electrons and heavier particles.
When the electrons are set in motion by a radio wave they acquire oscillatory energy
from the wave, but when the collisions occur this energy is converted into heat.
Since the energy comes from the wave, the latter is attenuated. In the HF band the
rate of attenuation A is given approximately by
1.16 x 1(T6 Nv , , , ,
A = j dB/metre (16.4)
400
300
~ 200
100
0
12 18 24
local time , hrs
Fig. 16.2 Ionospheric layers
through the night at an altitude of around 250 km, During daylight hours the layer
intensifies, thickens and usually develops two distinct maxima, the ¥x layer at
around 200 km and the F 2 at around 320 km altitude. The electron density is
around 3 x 1011 /m 3 , but the collision frequency v is very low, less than 10 3 /s. The
critical frequency fc is thus around 5 MHz while the attenuation is negligible. The
F layer is the most important for long-range communication as it has the highest
critical frequency and provides the longest 'hops', The F layer remains usable dur-
ing the night although its critical frequency is lowered as the ionisation decays.
Fig. 16.3 shows in diagramatic form the reflection of an HF wave by an ionised
layer. For any given critical frequency and radio frequency there is a minimum
incidence angle given by eqn. 16.3, and this determines the minimum single-hop
sky wave range or 'skip distance'. For any given range, there is a maximum radio
frequency given by the same equation; this is called the 'maximum usable frequency'
or MUF.
skip distance
Fig. 16.4 shows a typical relationship between the single hop range and the angle
of elevation at which the waves leave and return to the earth. It is advantageous to
design the antenna to fire at the optimum elevation angle for the desired path, but
in practice there is no single precise optimum angle because of the varying heights
of the ionised layers. The design angle of fire is thus a compromise, but the eleva-
tion beamwidth is usually wide enough to cover the variations.
Over long paths the wave may be reflected several times from the ionosphere,
bouncing off the sea or land between each hop. If the frequency is well below the
MUF for the route, several different paths may be possible, having various numbers
of hops and using the E, ¥x or F 2 layers or perhaps a combination of reflections
from two different layers (Fig. 16.5). It is even possible for a superior reflection
from an E layer to occur between two F layer reflections; this is known as an M
reflection. Since multipath propagation introduces distortion due both to wave
interference and to differential path delay, it is usual to minimise these effects by
working on a frequency fairly close to the MUF so that only one mode of propa-
gation can exist. The use of the highest possible frequency also minimises attenu-
ation. To allow for unpredictable variations in the MUF itself the optimum working
frequency is usually taken to be 85% of the MUF forecast.
Fig. 16.6 shows a typical MUF prediction for a particular route. The curves take
into account a number of different effects:
High Frequency Antennas 667
30-
F2 320 km
F 250km
F1 200 km
.E 110 km
SUMMER WINTER
50
0 hrs 12 24 0 hrs
Over land which has poor or very poor conductivity, the ground wave is greatly
attenuated and dies away within a few tens of kilometres of the launching antenna.
There is then a silent zone which extends out to the first sky wave skip distance.
This section has shown that the propagation of HF is fundamentally different
from that of any other frequency band. Ionospheric physics is a major subject in
its own right, and the outline given is necessarily brief, being intended only as an
introduction to antenna design. A far more detailed approach may be required
if any unusual HF system is contemplated, and References 28 and 29 may prove
helpful.
High Frequency Antennas 669
Signals are coded and a parity check is used to identify distorted groups so that a
repeat may be requested automatically. Telegraphic codes have a second advantage
in that they make optimum use of the limited HF spectrum which is available at
any given time.
These techniques, used with precise frequency control, enable many simultaneous
telegraph signals to be packed into the spectral band which would be needed for a single
speech channel. Note that multipath distortion limits the speed of telegraphy at HF.
It will be seen that, although wideband antennas are needed to accommodate the
slow changes in the ionosphere, the instantaneous bandwidth of any individual
signal channel may be quite small. However, one antenna may be used to transmit
or receive on several frequencies simultaneously.
High-frequency systems differ from microwave systems in that they are almost
always limited by external rather than internal noise. The thermal noise in a perfect
receiver is — 204 dBW in a bandwidth of 1 Hz. In a hypothetical receiver of overall
noise figure 6 dB and bandwidth 6 kHz the thermal noise is then — 160 dBW. At
HF, the background noise picked up by the antenna is normally much higher than
the thermal noise, and might typically be —130 to —100 dBW in the 6 kHz receiver.
The external noise is made up of contributions from several different sources:
(i) Manmade noise from vehicle ignition systems, industrial machinery, power
lines and similar sources
(ii) Manmade interference from other radio services, including spurious trans-
missions outside nominal allocated spectra
(iii) Natural noise from thunderstorms
(iv) Extraterrestrial noise from the sun or from the galaxy.
Since noise as well as signals may propagate by sky waves, noise sources such as
tropical storms may make an important contribution to the total, although they
may be thousands of kilometres distant.
Because the HF spectrum is external-noise dominated, receiving antenna direc-
tivity is more important than absolute gain. A lossy but highly directive receiving
antenna may give a better signal/noise ratio than a lossless antenna of lower gain.
For transmitting, however, maximum power gain is required, and directivity is only
important from the point of view of reducing interference with other users.
Because of the changing quality of the ionospheric path, HF skywave losses may
vary greatly. Under favourable conditions the ionosphere is a near-perfect reflector,
and at the higher end of the band where the D layer loss is small, communications
may be possible over antipodal paths with only a few watts of transmitted power.
The downward wave from the ionosphere may sometimes be focused into a small
area by the curvature of the reflecting layers, and then an artificially high signal
may be received (Fig. 16.8). On the other hand, under unfavourable conditions, it
may be necessary to employ very high powers and high antenna gains in order to
maintain communication.
While the familiar concepts of antenna gain and inverse-square-law power decay
apply equally to HF as to other bands, it should be noted that the method of calcu-
lating system parameters in practical applications may be somewhat different.
High Frequency Antennas 671
For system design purposes, a standard transmitter is often assumed, having a
power of 1 kW and driving a short vertical monopole antenna of gain 4.7 dB over
isotropic. The signal strength S in volts/metre at a range r metres from the antenna
is then given approximately by
300
S = (16.6)
irregular
ionosphere
40-
20 MHz
20
15 MHz
T3
hrs 12 24
These forecasts also estimate the noise level at the distant site assuming an omni-
directional receiving antenna and a receiver of standard bandwidth, perhaps 6 kHz
or 1 kHz. The result is presented to the system designer in the form of a chart of
expected signal/noise ratio against time of day for various available frequencies. It
is then relatively easy to determine how much additional transmitter power, or
transmitting or receiving antenna gain, is needed to provide the quality of service and
fading resistance specified. An example of a chart of this kind is shown in Fig. 16.9.
672 High Frequency Antennas
Note that the communication engineer still occasionally uses a horizontal dipole
at optimum height as an alternative gain standard. Its gain is 8.15 dB with respect
to an isotropic antenna.
Because of the relative unimportance of small losses in an external-noise-limited
receiving system, the impedance match of receiving antennas is not critical, and
VSWRs up to 3.0 are usually acceptable. For transmission a good match is needed
to protect the amplifier, and while a transmitter can usually tune out mismatches
of up to 2.0, VSWRs of 1.5 or less are generally preferred, particularly when the
antenna and transmitter are separated by several wavelengths of feeder line.
high beam
At small incidence angles (large values of a) the reflection coefficient falls and the
loss may be correspondingly greater, up to 3dB; but even here the gain remains
some 3dB higher than it would be if the ground reinforcement were absent. For
practical purposes, therefore, and particularly for antennas having low angles of
fire, the effects of ground imperfections may often be ignored if the polarisation is
horizontal. The majority of large HF antennas use horizontal polarisation for this
reason.
horizontal polarisation
0-5-
high beam
low beam
At low angles of fire, the ground reflection tends to cancel the direct wave, just as
it does for horizontal polarisation. Thus the efficiency of a vertically polarised
antenna over natural ground is poor at low elevations and varies unpredictably with
ground moisture content. The use of vertically polarised HF antennas over unim-
proved ground is generally avoided when a low angle of fire is required.
L = (16.11)
8 ( 1 - c o s a)
where a is the angle of lire as before. In other directions improvement to a distance
of \/4 from the antenna is suggested (Fig. 16.13). The lateral extent of the improved
zone depends very much on the type of azimuth coverage required (see Wait26).
Ground improvement may be by parallel or radial wires when the ground-current
676 High Frequency Antennas
direction of the axis the field must be zero, since the radiation pattern of each
elementary portion includes a sin 6 term. When the angle becomes substantial, the
contributions received from all parts of the wire differ in phase by:
If the amplitude of the wave is assumed to be uniform along the wire, that is, if
the loss of energy by radiation is neglected, then a null occurs in the pattern when
this phase difference is equal to any integral multiple of 2n. A maximum occurs
approximately halfway between each null. The radiation pattern of a uniformly
energised wire antenna of length 3X is shown in Fig. 16.15. Note that the largest
lobe is nearest to the direction of wave propagation along the wire.
678 High Frequency Antennas
It can be shown that the radiation pattern of any uniformly energised termin-
ated long wire is given by
sin 9 \nL (1 — cos0)
(16.13)
1 - cos 0 sin 'L X
cos<9 = 1 - 0 . 3 7 — (16.14)
and this function is plotted in Fig. 16.16 for typical practical values ofX/L.
40
20-
The assumption that the excitation is uniform is not justifiable in practice, since
perhaps half the power may be radiated, but the effect of this approximation on
the position and shape of the pattern lobes is relatively small. Since the main lobe
of the pattern in three dimensions lies on a cone surrounding the axis of the wire,
the long wire antenna is of little use by itself. It becomes useful when the radiation
patterns of two or more identical long wires are combined to form a single beam,
as described in the following sections.
High Frequency Antennas 679
16.3.2 Vee antennas
The inverted-vee antenna described by Bruce5 is a simple practical configuration
based on the long wire. It is inexpensive since it requires only one mast (which must
be non-conducting) and one wire radiator terminated in a resistive load at the far
end. Fig. 16.17 shows a typical design. The angle of inclination, 30°, is chosen to
make the maximum radiation from each section approximately horizontal (Fig.
16.16). This also ensures that the phase reversal brought about by the fact that one
half of the wire goes up and the other half down is compensated by the phase
reversal produced by the distance between the centres of the wires, approximately
0.5X less than the length of the current path.
3OO/i
2-5*
must be earthed, and the antenna is generally used only for reception because of
the difficulty of making satisfactory high-power earthed terminations. Like the
Bruce, the sloping-vee antenna has an octave bandwidth, substantial sidelobes, and
a low efficiency. The antenna has a high input impedance, around 800 £2, and is
conveniently fed by a tapered line transformer from a balanced twin transmission
line.
direction
load
The antenna is balanced and radiates horizontal polarisation, and like other
travelling wave antennas it requires a matched load at the far end. If the load is
not correctly matched a backlobe is formed by the reflected wave and the input
impedance is degraded.
Referring to Fig. 16.20, each of the four radiators generates a conical principal
lobe at an angle of about 29° to its axis. If the rhombus vertex angle were 58°,
these four lobes would all combine along the longitudinal axis of the antenna, the
half-wave fore-shortening of the path between the forward and rearward portions
High Frequency Antennas 681
compensating for the phase reversal due to the opposite inclination angles. An
improvement in gain can be obtained by choosing a slightly smaller vertex angle,
say 55° in the above example. This causes the single beam to break up into two
lobes corresponding to the intersection of slightly overlapping cones. One beam is
at around 15° elevation and is precisely in the wanted direction, while the second
beam at — 15° reinforces the first after ground reflection, the height of the rhombic
above ground being calculated from eqn. 16.9, so that the direct and reflected
beams are in phase.
10
side length, I
Fig. 16.21 Vertex angles and gains for rhombic antennas of optimum design
The large rhombics used for long-range communications have small vertex
angles and their resultant pattern lobes, corresponding to the direct and ground
reflection beams, cannot lie far from the horizontal plane. It is not possible to
design large rhombics to produce an optimum performance on high-angle short-hop
paths. The largest rhombics, of around 10A side, are at their best at elevation angles
of fire around 5°, for which masts of 3X height are necessary. A 5X rhombic is
suitable for 10°, a 3A rhombic for 15°. For each of these elevation angles, there is
little increase in gain if the rhombic is made larger than the suggested size.
one-third of the way up rather than at ground level the break-up of the antenna
currents into higher-order modes can be delayed and the elevation coverage can
be preserved.
insufator
transformer
feeder
The antenna consists of two concentric tubes, the larger forming the lower part.
At the feed point, a small transformer may be used to give the best compromise
match into the coaxial-cable feeder, which passes up through the lower tube. In
practice, elevated-feed monopoles may be used over bandwidths of up to 4:1 so
that two monopoles are sufficient to cover the entire HF band.
When impedance bandwidth is the limiting factor, an improvement can be
obtained by increasing the monopole diameter. If a wire-cage rather than a solid-
tube construction is employed, the monopole diameter may be made very large and
a working bandwidth exceeding 2:1 is readily obtained. In one practical design
described by Mason16 a double cone monopole is used to obtain a bandwidth of
2.6:1 with a VSWR not exceeding 1.6 into a 50 Q feeder. Fig. 16.23 shows a typi-
cal practical double-cone monopole antenna. When the limiting factor is physical
size, as it may be at the lower end of the band where a quarter wavelength may be
25 m or more, other variations may be used. Fig. 16.24 shows the inverted-L and T
antennas, in which a horizontal wire provides top capacitance and enables the
height to be reduced by an amount roughly equal to the length of the horizontal
portion. Since most of the current flow is in the vertical portion, the radiation
pattern remains substantially vertically polarised. Although the height is reduced,
two supports are now required so that there is little cost saving unless suitable
supporting structures are already available.
Top capacitance may also be added in the form of an umbrella consisting of the
upper portions of a circle of guy wires, Fig. 16.25.
Series inductance loading is convenient in practice because it is easily incorpor-
ated in self-supporting radiators. The position of the inductance or inductances
affects both the impedance and radiation pattern behaviour particularly at the
upper end of the working band where break-up is approaching.
High Frequency Antennas 685
\
unbalanced feed
feed feed
inverted-L T-antenna
top cap
mast
For mobile applications, loaded flexible 'whips' may be used. These are dis-
cussed in Section 16.8.1.
Since external rather than internal noise predominates in HF communications,
the power gain of receiving antennas is often unimportant provided that the direc-
tivity and linearity characteristics are acceptable. Thus it is becoming increasingly
common to use short low-gain monopoles with integral low-noise amplifiers in
place of full-size monopoles. These 'active' monopoles occupy less space and may
be considerably cheaper to install. Note that the active monopole is not merely a
short monopole plus an amplifier, since the amplifier circuit may be designed so
that it also acts as an impedance transformer, matching into the very low impedance
of the short monopole and compensating for its frequency-dependent reactance.
matching
section
limited by variations in the angle of fire, but for transmission the impedance is also
a critical parameter. The impedance bandwidth may be improved by reducing
the inductance/capacitance ratio in the same way as in the monopole, and designs
such as the 'bow-tie' (Fig. 16.27) and the cage dipole (Fig. 16.28) are often used
in practice in preference to single wires.
While it is possible to operate dipoles in full-wave as well as half-wave modes,
thus extending their coverage to a second frequency band, it should be noted that
their radiation patterns are then quite different. The use of harmonically related
frequency bands by amateur radio enthusiasts encourages this type of operation,
and it is even possible to operate on three bands with appropriate matching circuits.
Moxon18 describes one such design, but its application is limited to the special case
of harmonically related bands, and it does not meet the broader requirements of
commercial HF operation.
High Frequency Antennas 687
Slots are complementary to dipoles, and therefore in principle offer an alternative
family of antenna configurations. In practice slots are far less convenient at HF
because they need a large conducting sheet or mesh screen instead of a half-wave-
length conductor. Even when a conducting sheet is available, as for example the
side of a ship or the wing of an aircraft, it is not usually structurally permissible to
cut a slot large enough to resonate at HF. Nevertheless slot antennas remain an
interesting possibility for unusual applications.
feeder
Fig. 16.27 Bow-tie dipole
spreaders
feeder
Fig. 16.28 Cage dipo/e
in a small loop (small in terms of wavelength, that is) is proportional to the area
multiplied by the number of turns.
The radiation pattern is thus the same as that of a short magnetic dipole, or a
short electric dipole with the electric and magnetic fields interchanged. It is given
by sin 0, where 6 is the angle between the direction of the incident field and the
axis of the loop.
incident
wave
Loops may be untuned, but they are far more efficient if tuned to resonance.
The inductance of a practical multiple-turn loop at HF is sufficient to make it self-
resonant, and a single-turn loop tuned by a small capacitance is more convenient
and more efficient. Its area may be considerably larger than that of a multiple-turn
loop if self-resonance is the limiting factor on loop area.
earthed
screen
balanced feeder
Apart from their use in domestic receivers, loops are mainly employed in direction-
finding (DF) applications. The design of DF antennas is a higly specialised subject
which cannot adequately be covered in this general chapter. However, it should be
noted that the reduction of errors in DF systems depends principally upon the
High Frequency Antennas 689
elimination of pick-up due to unwanted modes. The use of shielded or balanced
loops (Fig. 16.30) helps to reduce the sensitivity to unwanted (unbalanced vertical)
signals.
feeder load
At resonance, however, the dipole impedance is much lower than the impedance
of the transmission line. The first dipole thus extracts a large proportion of the
total power in the line. The second dipole extracts a large proportion of the power
remaining after the first, so that the resultant amplitude distribution along the
array is approximately exponential. The far end of the line is terminated in a
matched load, but little power remains in the line at the far end and the contribution
of the last few elements is minimal.
As the frequency is varied from resonance, the impedance of the dipoles rises
and the power extracted by the first few elements is reduced. Since more power
remains in the line, the contribution of the elements at the far end is increased, so
that the antenna directivity gain is increased. At the same time, the power wasted
in the end load is increased, so that the overall gain rises to maxima at the upper
690 High Frequency Antennas
and lower ends of the band where all dipoles are contributing and then falls away
steeply as the end load loss becomes dominant. The useful radiation pattern band-
width of the antenna is about an octave, but the impedance match is never particu-
larly good because of the overwhelming effect of the first dipole. In consequence,
the antenna is used for reception where impedance match is not critical.
An interesting variant of the HAD is the RCA Fishbone, (Fig. 16.32). This uses
ceramic insulators to connect the dipoles to the feeder, so that the capacitance of
the insulators controls the coupling. Less power is extracted by the first few dipoles,
enabling a more uniform distribution to be achieved with a consequent higher gain.
Since the dipoles all present a capacitive reactance to the line, the line has a reduced
phase velocity and the dipole fields no longer add in phase in the direction of fire.
The phase lag actually increases the gain slightly (see Hansen and Woodyard10). The
antenna is usable for reception over a band of up to two octaves.
IIUUUUUL
feeder
TTTTT/T T T T T TTT
coupling capacitors
load
Both HAD and Fishbone antennas are slung horizontally between four masts or
towers. The elevation angle of fire is given by eqn. 16.9 in Section 16.2.1.
Vertically polarised endfire arrays are also possible but few have ever been used
in practice because of the need for extensive ground improvement if the potential
low angle of fire is to be realised.
T
4~
feeder
_L load
One example was the series phase array described by Ladner.15 This was an end-
fire array of quarter-wave folded monopoles at quarter-wave spacing, Fig. 16.33.
The array was fed against the ground plane, and as in previous endfire arrays a
resistive load was used to terminate the far end, typically 300 O. A gain of around
12dB was claimed for an array four wavelengths long, but the impedance band-
width proved poor and the array was unable to handle high powers.
The series phase array required only two supports, and since these were parallel
High Frequency Antennas 691
to the radiators it was necessary to use wooden poles or towers rather than metal
masts.
All uniform endfire arrays can be made to fire on an opposite bearing by inter-
changing the feeder and the matched load, and this feature is useful in HF com-
munications when the same antenna is used at different times of the day for dif-
ferent circuits.
Thus the number of rows of dipoles depends on the distance of the service area
(Knight13) while the number of columns depends on the angle which the service
area subtends at the transmitter. The gain of a typical array is 22dB, relative to
t
2
to transmitter
that of an isotropic source in free space (22dBi). Fig. 16.35 shows gains of arrays of
type HR 4/m/h and Fig. 16.36 shows the elevation angle of the main lobe.
The type of curtain array described above can only be used for one broadcasting
band. Consequently large numbers of arrays had to be provided to cater for all
High Frequency Antennas 693
broadcasting bands and for all required directions of propagation. Modern curtain
arrays, however, can be used for up to four adjacent broadcasting bands, covering
a frequency range of nearly 2 : 1 . This is achieved by the use of centre-fed wide-
band dipoles, such as those shown in Fig. 16.28, and by branch feeding the rows of
dipoles, as shown in Fig. 16.37.
22 -
20
c
&
16
10
0.5 10 1.5 20
height of lowest row of dipoles above ground, A
The reflector curtain consists of a grid of horizontal wires spaced about 0.03A
apart. Althouth the array cannot be reversed, this restriction is not a serious limi-
tation because service areas seldom lie on reciprocal bearings. The number of
vertical feeders has been doubled, but this is an advantage because much wider
slew angles can be achieved by progressive phasing. Slew angles of up to 30° are
possible without a serious reduction of gain or the generation of strong sidelobes.
Horizontal and vertical radiation patterns of wide-band arrays, covering a 2:1
frequency range, are shown in Figs. 16.38 and 16.39.
Whereas at microwave frequencies it is usual to optimise phased arrays for some
relatively narrow bandwidth such as 10%, at HF the design is compromised to
permit operation over as wide a band as possible. When the frequency is raised
above the grating-lobe threshold, the gain reaches a plateau as the grating-lobe
loss compensates for the narrowing main beam. Provided that the element remains
workable, the array continues to provide gain in the wanted direction, but un-
wanted signals are now transmitted and received at wide angles rather than close
694 High Frequency Antennas
to the wanted direction of radiation. Thus, the array can be used successfully at
these higher frequencies provided the grating lobes can be tolerated, (see Fig.
16.40.)
50r
40
;30
o
£20
c
o
g
I
10
At lower frequencies also, the degradation can be tolerated. The gain is deter-
mined by the overall aperture available regardless of whether the aperture is filled
or over-filled by elements. In receiving arrays the elements do not need to collect
all the power available, but merely to sample the field with sufficient element
gain to make external noise mask receiver noise; thus the mutual effects between
elements may be unimportant even with close element spacing.
Simple omnidirectional receiving elements such as elevated-feed monopoles
have the disadvantage of radiating equally on both sides of the array so that a
backlobe of equal magnitude to the wanted lobe is generated. Use of a reflecting
system enables the backlobe to be cancelled at the frequency for which the reflector
is optimised, but passive reflectors do not behave well over wide bands. Not only
does a reflector distort the forward primary pattern; it may even turn into a director
and reverse the pattern at some other frequency.
High Frequency Antennas 695
One solution is the use of doublet elements, fed in delayed antiphase so as to
give perfect cancellation in the backward direction. The elements are closely spaced
(Fig. 16.41), so that in the forward direction they do not add in phase except per-
haps at the highest frequency in the operating band. The resulting primary pattern
is a cardioid or quasi-cardioid, and considerable suppression of the backlobe is
obtained.
A
/2
(a) unslewed
(b) slewed
Fig. 16.38 Horizontal radiation patterns of wide-band HRS 4/m/h curtain array
f0 = design frequency
\30°
elevation
angle
20°
others meant that the maximum gain was less than that of a broadside array having
the same total number of elements; thus the average gain improvement was only
about lldB as compared with a single isolated element. The sidelobe and backlobe
High Frequency Antennas 697
performance was not particularly good, the sidelobes being more easily predicted
by statistics than by rigorous analysis.
Broadside arrays of more complex elements are possible, giving higher gains
but with limitations in scan coverage. Thus a broadside array of rhombics can
provide a very high gain, but is limited to a phase scan sector of a few degrees.
broadside beam
unwanted
grating
!obe
array elements
phase
reversal
fc feeders
goniometer
elements
The difficulty of the Wullenweber arises from the need to cover a wide band. A
single receiving array covering the whole HF band requires too many compromises,
but two arrays covering perhaps 2-8 and 8-32 MHz, respectively, are feasible.
The elements must therefore maintain an acceptable primary pattern over a 4:1
band, and must do this in the presence of the other elements. Possible problems
include the degradation of the primary pattern due to the reflector becoming a
director, or due to travelling waves running around the array. Fortunately these
effects can easily be checked by scale modelling before the full-size antenna is
constructed.
low band
high band
If high- and low-band arrays are constructed concentrically to save space, and if
the high-band antenna is mounted outside the low-band circle to avoid obscuration,
there will be a wide difference between the gains at the upper and lower ends, and
High Frequency Antennas 699
the high-band array will have at least four times as many elements as the low-band
array (Fig. 16.43). If the high-band array is mounted in the centre of the low-band
array (Fig. 16.44), the number of elements can be the same for each. Obscuration
effects are limited to low arrival angles, and the same phasing system can now be
used for both bands.
! |
I I!| III '
high band
low band
The circular array can only be correctly focused at one arrival angle. At higher or
lower elevations the pattern is defocused and there is a loss of gain accompanied by
a rise in near sidelobe levels. The effects are minimised by setting the phasing
system to give optimum focusing at medium arrival angles rather than at zero
elevation. The defocusing errors are then acceptably small for all but the highest
elevation signals.
The goniometer may be a mechanically rotating capacitive switch, or it may be
an all-electronic system. Future systems are likely to be of the latter type to over-
come the sampling-rate limitations of mechanical switches.
large enough to radiate all the power at the longest operating wavelength, any
required bandwidth was theoretically possible.
The impedance of the early log-periodic antennas showed a marked periodic
variation, but generally remained within a VSWR of 2 : 1 ; thus they could be used
for transmission as well as reception. Free-space gains of around 6dB were readily
obtained, giving around lOdB with ground reflection reinforcement. Azimuth
beamwidths of around 60° were obtainable, with a backlobe perhaps 15dB below
the forward radiation. In comparison with the rhombic, the log-periodic was
inferior at the design optimum frequency, but superior when considered over a
wide band. The log-periodic did not need a matched termination and did not waste
any of the transmitted power.
balanced feed
Fig. 16.45 Log-periodic antenna
However, the early log-periodic had some disadvantages. From the designer's
point of view, it was not easy to control the antenna characteristics precisely
because the same periodic structure was acting both as transmission line and as
radiating element. A further major step forward was taken when the two functions
were separated, as they already were in the HAD and in the curtain arrays, resulting
in the logarithmic dipole antenna described in the following Section.
elements behind, and so tends to produce a directional primary pattern in the same
way as a Yagi antenna. This effect also contributes to the directivity and helps to
produce the very low backlobe and sidelobe level.
dipoles
feeder
The active elements, those at or just below resonance, are spaced at less than a
quarter-wave apart. The capacitive loading effects of the below-resonant elements
increase the effective spacing in the active region to a value closer to the quarter-
wave which theoretically provides the best impedance match. Furthermore, the
sharing of the power between several dipole elements in the active region and the
resultant tapered amplitude distribution improves both the impedance match and
the radiation pattern.
The theory of the logarithmic dipole antenna was first published by Carrel.7 His
calculated antenna characteristics showed good agreement with experimental results
obtained elsewhere.
For use at HF, the logarithmic dipole antenna is hung from a pair of masts with
its vertex end supported by low wooden poles so that the mathematical apex of
the dipole curtain lies in the ground plane (Fig. 16.47). This ensures that the
elevation patterns as well as the azimuth patterns remain independent of frequency.
For any given antenna and given angle of fire there is an optimum tilt angle. If the
masts are high, the ground reflection generates a shoulder lobe above the main
beam (Fig. 16.48). If too low, the main beam itself is at too high an elevation for
long-range working as in Fig. 16.49.
Vertically polarised logarithmic dipole antennas may also be used. Whereas the
azimuth beamwidth of a horizontally polarised antenna is typically 50° to 70°, that
of a vertically polarised antenna may be 100° or more; thus two antennas may be
combined in a vee formation to reduce the azimuth coverage (Fig. 16.50). If a dual
polarised antenna is required, it is possible to hang both horizontally and vertically
polarised antennas from the same pair of masts.
An interesting variant of the dual polarised logarithmic dipole antenna is shown
in Fig. 16.51. Two identical antennas are hung from a single mast and a single
common vertex pole, with inclinations of ± 45°. The two antennas are fed from a
hybrid transformer which gives either in-phase or antiphase excitation according
to which input terminals are used. The same antenna system can thus generate
702 High Frequency Antennas
30°tilt
10°tilt
feeders
feeders
Fig. 16.51 Dual-polarised log-periodic dipole array
horizontal
polarisation
vertical
polarisation
One interesting variant was the broad dipole array invented by E. W. Woloszczuk
and described by Radford and Woloszczuk.20 In this antenna the dipoles were
broadened until they almost touched and were then replaced by rectangular wire
loops (Fig. 16.53). The action of the antenna was more complicated than the
simple explanation of element broad-banding would suggest, as was demonstrated
by current measurements on a full-size antenna, but the aperture efficiency was
some 2dB better than that of a simple dipole antenna. The improved impedance
match and very low sidelobes were additional advantages. However, the broad
dipole antenna was more difficult to rig than the simple dipole version, and it
required more wire and more insulators; thus it had no advantages in terms of cost-
effectiveness. As far as is known, the logarithmic broad dipole antenna has never
been investigated theoretically.
704 High Frequency Antennas
Loaded dipole elements have been used in logarithmic antennas. In rotatable
antennas used for communications with mobile units and other applications in
which mechanically steerable beams are desirable, inductive loading or capacitive
tip loading can be used to reduce the span of the longest dipoles (Fig. 16.54).
wire loop
dipoles
Dn ]DEL feeder
Series capacitance loading can also be used. This has the opposite effect, increas-
ing the resonant length of the dipole but at the same time increasing the element
directivity in the plane of the dipoles. This increase is obtained without any accom-
panying increase in directivity in the other plane. By contrast, increasing the length
of a simple dipole array increases the directivity in both planes. The importance
of this will be seen in the next Section where arrays of logarithmic antennas are
considered.
In designing practical logarithmic antennas, it is first necessary to choose the
configuration appropriate to the specification. For short-range sky wave circuits,
requiring high elevation angle coverage, short horizontally-polarised antennas of
low taper ratio are appropriate. For longer ranges where higher gains and lower
angles of fire are needed a higher taper ratio is needed. The taper ratio t is defined as
t = (16.15)
High Frequency Antennas 705
where /„ and dn are the lengths and spacings of the nth dipole, respectively (see
Fig. 16.55).
It is also convenient to define the spacing ratio s:
s = |=- (16.16)
Lln
Many early designs used taper ratios such as 0.7943 (antilog — 0.1), 0.8913 (antilog
— 0.05) and 0.9441 (antilog —0.025), because this enabled lengths and spacings
to be read directly from antilogarithm tables.
10 • 100°
H-plane
E -plane
5- 50°
in .c
-D
T3
C
03 -
O°
0-8 0-9 1-0
Fig. 16.56 shows the variation of free-space antenna parameters with taper ratio
for a simple logarithmic dipole array, and Fig. 16.57 shows approximate optimum
ratios for given taper ratios. Ground reinforcement adds some 4dB rather than
706 High Frequency Antennas
the 6dB obtained with horizontal dipoles because the elevation pattern maxima
of the antenna and its ground image do not correspond.
In designing vertically-polarised logarithmic antennas, the usual problems of
ground conductivity arise, and extensive ground improvement is necessary unless
the antenna is on a marshy or coastal site. However, the design of the elements
follows the same rules as before.
0-2
0-8 t 0-9 10
Fig. 16.57 Log-periodic dipole antenna: Relationship between spacing and taper ratios
In a simple logarithmic dipole antenna the longest dipole should be half a wave-
length long at the lowest frequency of operation. It is usually convenient to short-
circuit the feed at this end; thus the last element is never driven but always acts as
a reflector. No terminating load is needed as little unradiated power remains at this
point. The shortest element should be resonant at about three times the highest
operating frequency of the antenna.
The transmission-line impedance is nominally 400-600 12, but since it is reduced
to 200-300 £1 by the capacitive loading of the dipoles on the vertex side of the
resonant region, it may be necessary to add a lumped capacitance at the point
where the feeder is connected to the antenna. This compensates for the discon-
tinuity in the capacitive loading; its value is best adjusted experimentally. With
correct compensation, an impedance match of 1.25 VSWR can be obtained with a
broad dipole configuration, or around 1.5 VSWR with simple dipoles.
to
phase
controlled
feeder system
These wide coverages are only obtained with short antennas of low taper ratio, and
result in high angles of fire in the horizontal case. If long-range low-angle coverage
patterns are needed, it may be necessary to resort to special logarithmic antenna
designs having parameters optimised for this application. Series capacitance-loaded
elements may be used to improve the elevation directivity of vertically-polarised
arrays without loss of azimuth scan coverage.
The array spacing can only be optimum at one frequency. At lower frequencies,
there is insufficient room for horizontally-polarised antennas while vertically-
polarised antennas are severely degraded by mutual effects. At higher frequencies,
the angle over which a unique beam is generated is reduced, until at frequencies
for which the spacing exceeds one wavelength two or more beams will always be
present. The amplitudes of the widely spaced grating lobes are, of course, reduced
by the shaping of the primary patterns, but grating lobes remain a potential source
of interference with other circuits in transmission and of unwanted noise pickup
and directional ambiguity in reception.
To avoid this problem, the idea of inserting small logarithmic antennas between
the larger antennas has been examined (Fig. 16.59), but the system becomes
unattractive when its detailed design and performance are considered. In particular,
there is an awkward frequency band just below the frequency at which the smaller
antennas can begin to contribute. Mutual effects cannot be neglected in this arrange-
ment; indeed they continue to degrade the array element performance out to
spacings of little over one wavelength.
708 High Frequency Antennas
direction
of fire
This system works well for arrays with few elements, but if a higher gain is required
the principle cannot be extended as the outer elements in the array would be firing
in the wrong direction. Fig. 16.61 shows a complete circular array using this prin-
ciple. It gives a full 360° coverage, but only combines those elements facing in the
right direction in any given beam position. Obviously a part-circular array can be
used if coverage is only required over a limited arc of scan.
The horn is fed from a transmission line via a giant ridge waveguide section in
the throat. It is horizontally polarised.
The rhombic section results in some curvature of the field in the mouth of the
horn which tapers the excitation and so improves the sidelobe performance.
The bandwidth is limited by the matching of the feeder to about 3.5:1 for a
VSWR of 4. There is a considerable variation in the beamwidth over the frequency
range, but the sidelobes remain reasonably good, better than 20dB down at the
upper end of the band.
One unusual property of the TAHA is that the elevation angle of fire is less
frequency-dependent than that of conventional antennas having a fixed height.
This is because the phase centre height varies with frequency.
A practical full-size antenna covering 6.5-23 MHz measured some 257 m long,
154 m wide and 77 m high. Its gain varied from about 20dB at the lower end to
28dB at the upper end of the band.
The TAHA is, compared with a rhombic, expensive to construct and maintain,
and it has not proved as cost effective as the rhombic or the logarithmic antenna.
flare-
feed in throat
Fig. 16.64 HF wire grid lens antenna
From symmetry considerations, the lens can also be made by placing a single
grid over a perfectly conducting plane and making the spacing # instead of 2a.
Consider now the practical design of an HF antenna using the wire-grid lens
principle. The upper mesh can be suspended from a ring of non-conducting poles,
High Frequency Antennas 711
with spacers to separate it from the lower grid which is supported on stakes. A
conical flare surrounds the lens to shape the elevation pattern (Fig. 16.64). A
practical antenna covering 3-30 MHz had a diameter of 260 m and a flare height
of 27.5 m (see Sharp22).
The wire-grid lens antenna is expensive to construct and maintain, and can only
be justified if its multiple beam capability is fully exploited.
matching by raising the input impedance from 1 £1 to perhaps 15-20 £2, which is
easily transformed to 50 £2 for feeder connection. The calculation of loading coil
inductance follows the basic theory of Chapter 1 (see also Chapter 15 and Belrose2).
E
JC
O
An elevated loading coil can be used to improve the antenna current distribution
and thus raise the radiation resistance. The optimum coil position varies with the
earth loss, which is itself variable and unpredictable, but halfway up the whip is
a good practical compromise. The elevated loading coil gives a small but worthwhile
improvement in efficiency as compared with the base loading coil position, but
there are two attendant disadvantages:
(i) The inductance is no longer easily accessible for tuning when the frequency is
changed.
(ii) The loading inductance complicates the mechanical design of the whip, requir-
ing a stronger lower portion to support its weight and increasing its vulner-
ability to accidental damage.
If a wide tuning capability is required, the use of an elevated loading coil is thus
hardly worthwhile. If fixed-frequency or narrow-band operation is sufficient and
if the operational environment is not too difficult, a fixed elevated loading coil
may be used. Note that even if the coil inductance is calculated to make the input
impedance purely resistive on any one site, additional tuning reactance will still be
needed to compensate for changes in site characteristics.
The shorter the whip and the greater the inductance needed to tune it to reso-
nance, the narrower will be the instantaneous bandwidth. The instantaneous band-
width of a typical whip antenna is only around 1%, which is adequate for voice
or teleprinter operation but leaves little scope for frequency changing without
High Frequency A n tennas 713
I. - ()
where Zs is the equivalent slot impedance. This method cannot be perfectly accu-
rate as it uses only a 2-dimensional projection of the aircraft fuselage, but it gives
the designer an initial insight into which modes are present. On a large aircraft,
some part of the structure such as the wing or the tail may become quarter-wave
resonant, and will present a low impedance and so draw most of the power. It will
then dominate the radiation pattern, and may result in little or no power being
radiated in some directions.
The earliest HF aircraft antenna was the 'clothes-line', a wire antenna supported
by a short mast at the forward end and the tail fin at the aft end. The clothes line
actually forms a slot of sub resonant size, and external reactance is needed to tune
it to resonance. The clothes line was satisfactory for slow piston-engined aircraft,
but its drag was not acceptable on high-performance aircraft.
To meet the needs of the aerodynamicist, completely suppressed antennas were
714 High Frequency Antennas
devised, that is, antennas buried in the aircraft structure. Examples include notches
(Fig. 16.67), tail and wing caps insulated from the main structure, slots, and con-
ducting rails buried in insulated fairings. Small probes having minimal drag have
also been used on large aircraft.
coaxial feed
image plane
conducting
sheet
notch in /
fin fairing^/
Of these various possibilities the notch first proposed by Johnson12 is the most
efficient. Unfortunately, the ideal electrical position for the notch is usually a wing
root or at the base of the tail fin, both highly-stressed regions where the inter-
ruption of a spar by an insulated joint is mechanically undesirable. As a result, the
notch cannot be made as large as the antenna designer would like, and its impe-
dance is normally inductive. Thus, it has to be tuned to resonance by a capacitative
tuner.
The impedance of the tuned notch may be very different from the 50 £2 impe-
dance of the feeder cable; thus a matching transformer is also required. Since the
slot impedance varies with frequency, a simple transformer is not enough and an
adjustable reactive matching network must be used. The tuning and matching func-
tions may be separated, or they may be combined in a single unit. Fig. 16.68 shows
the arrangement of a typical notch and its associated tuning and matching equip-
ment.
For efficient operation, the notch must be coupled as tightly as possible into the
High Frequency A ntennas 715
currents flowing on the aircraft surface. In practice, even when the notch is opti-
mally sited at a high-current point like a wing root its radiation resistance is only
around 0.1 ^2. Low-loss reactances must be used for tuning and matching, and the
tuned notch may have a useful bandwidth of as little as 10 kHz.
The low antenna efficiency plus the possibility of noise generated by other
electronic equipment and static discharges means that the receiving signal/noise
problems cannot be forgotten. Losses in reception must be kept low, all inter-
ference from other equipments must be carefully suppressed, and antennas must
be chosen to minimise static noise. The notch is one of the best antennas in this
respect and this is a further reason for its popularity.
The provision of efficient HF antennas in aircraft will become more and more
difficult as the constraints of aerodynamics, space and weight become more severe.
However, the availability of satellite communications will relegate HF to emergency
back-up status, whereas until recently HF provided the only means by which an
aircraft in mid-ocean could communicate with land.
At least two classes of scale model may be used. Small-scale models having scale
factors of order 1/100 are suitable for initial investigations of new designs and for
measurement of radiation patterns. Large-scale models having scale factors of 1/2
to 1/5 are suitable for impedance measurements and for initial trials. Since their
construction can be similar to that of the full-size antenna they also provide useful
experience for the mechanical designers and riggers.
In choosing the best scale for any model, the size of the antenna test site, the
available equipment, the constructional facilities and the nature and accuracy of
the information required must be taken into account. True radiation patterns can
only be measured in the far field, which is usually taken to be the region beyond
the critical range R given by
ID2
R = — (16.19)
where D is the antenna aperture and X the wavelength. For precise measurement of
very low sidelobes, a somewhat longer range should be used. The site thus deter-
mines the lowest frequency which can be used for antenna pattern measurements.
The highest frequency which can be used is determined by the amount of detail
to be reproduced in the model and the frequency and bandwidth of the measuring
equipment available.
Impedances may be measured accurately in coaxial lines over a frequency range
extending from HF to microwaves, but accurate measurement in balanced open
lines becomes difficult above the VHF band. Balance-to-unbalance transformers
covering wide bands are not easy to construct with sufficient accuracy; thus for
precise impedance measurements on balanced antennas large-scale models should
be used.
The construction of small-scale models cannot easily reproduce all the features
of full-size antennas. Wire cannot be precisely scaled, or it becomes extremely
fragile and its ohmic resistance is too high. It is better to use copper wire of suf-
ficient diameter to stand up to normal careful handling, and recognise that impe-
dances and losses may not be perfectly represented in the model. Again, rigging
cannot be precisely scaled in a 1/100 model but unless there is some question
of obscuration or reradiation by the rigging, the radiation patterns will not be much
affected. Satisfactory models of dipole arrays, curtains and logarithmic antennas can
be made by supporting the radiating wires on light perspex frameworks.
Models of aircraft, vehicles and ships can be used to optimise antenna positions
and predict radiation patterns when the structures are too complex for theoretical
modelling. However, the models must be reasonably detailed, and must include
any other antennas or other structures which are likely to be added. With careful
modelling, approximate prediction of antenna impedance becomes possible. The
cost of a detailed model is usually amply justified by the time saved when the full-
size antenna is installed.
High Frequency Antennas 719
tn = — tan0 n ) (16.20)
where <j)n is the inclination of the catenary between the (n — l)th and nth dipole
and T is the tension of the section whose inclination is zero. This formula assumes
that the weight of the catenary itself is negligible compared with the dipole tensions.
v t
The catenary tensions in large antennas may be several thousand kilograms. Mech-
anical design involves finding a compromise between the catenary tension and the
permissible curvature, since the lower the catenary tension the greater must be the
total span of the catenaries at the end of the array. In the case of a vertical array,
the lower the catenary tension the greater must be the height of the masts above
the top of the radiating elements.
The tensions in an ice-loaded antenna under wind-loaded conditions will be
much in excess of the normal working loads. A wire catenary which is only slightly
elastic suffers a greater excess load than a synthetic rope catenary which stretches
under load and so eases the dipole tensions. However, synthetic ropes are less stable
than wires and are generally only used on smaller antennas. The antenna should
always be designed to fail safe, so that under intolerable loads it is an antenna wire
or an insulator which gives way and not a mast or a main guy wire. For antenna
design in the United Kindgom, an antenna designed to withstand a 15 m/s wind
with a 5 mm radius ice loading represents a reasonable compromise between surviv-
ability and cost.
High Frequency Antennas 721
structure should be electrically bonded, and that imperfect joints which could cause
crossmodulation or generate noise should be eliminated. It is also important that
any ignition systems or other electrical equipment on the vehicle should be care-
fully screened and decoupled.
Transmission lines for HF antennas must have a good impedance match, low
losses and adequate power-handling capability. Older transmitting antennas usually
had open-wire balanced feeders of impedance 600 £2 (two-wire) or 300 Q (four-
wire). Attenuation in open feeders is low, typically 1-2 dB/km, and power handling
is high, tens or even hundreds of kilowatts with suitable spacing and insulation.
Open feeders require careful design at bends or twists, as any unbalance can
result in leakage of energy as well as degradation of the impedance match. If open
feeders are allowed to pass too close to each other, crosstalk can occur. Open
feeders are subject to wind and ice loading, and their mechanical design must take
these factors inco account.
Coaxial feeders are now preferred for most applications. They overcome leakage
and crosstalk problems and are immune from ice and wind loading, but their
initial installation costs are higher. The attenuation of large-diameter concentric
lines is comparable with that of open-wire feeders, and power levels of up to 100
kW can be handled by coaxial lines of around 10 cm diameter. Impedances of
50 £2 are usual for transmission, this being the best compromise between low
attenuation and high-power-handling requirements. For reception 75 £2 is some-
times used on account of its marginally lower attentuation, but recent designs tend
to standardise on 50 £2 for both transmission and reception.
High-power concentric line feeders must be sealed and are usually filled with
dry air or nitrogen at slightly above atmospheric pressure to prevent ingress of
moisture.
On large HF communication sites there may be a switching centre which enables
any transmitter to be coupled to any antenna.
Complaints from users about antenna performance can often be traced to faults
in feeders and switching centres. In particular, crosstalk and leakage can be inter-
preted as sidelobe pickup. Broken or out-of-balance feeders can radiate and receive
sufficient power to maintain communication under good ionospheric conditions,
and this can lead to complaints of low signal strength rather than absence of signals.
services. However, there must be reasonable site access for construction, and reason-
able domestic and other facilities for operators.
The site must be protected from vandalism and pilfering, copper being par-
ticularly attractive to thieves. It is also necessary to protect innocent members of
the public from high voltages and from intense radiation fields.
Having found a site, the individual antennas must be carefully placed to mini-
mise mutual obscuration and interference. It should be noted that wideband
antennas can cause more obscuration problems than highly resonant antennas, since
they are active over a wider band. If the site is an existing station with antennas
already present, the possibility of resiting some of the existing antennas should be
considered. In determining which positions give least mutual disturbance, a scale-
modelling or theoretical-modelling exercise may be justifiable.
The risk of accidental ignition of inflammable vapour at nearby petrochemical
plants and gas terminals by high-power HF transmitters should also be checked
when any new site is assessed.
Environmental acceptability can be improved by using low ground, where con-
ductivity may also be better, and by using slender masts rather than heavy towers.
In undulating country, antennas on the slopes of hills do not break the skyline and
are far less noticeable. A site which slopes gently downwards in the direction
of fire also improves the low elevation performance of the antenna.
The antenna engineer may protest that environmental considerations are outside
his terms of reference. However, he should remember that planning authorities
and the general public at large understand little of electromagnetic theory, and that
he is better able to optimise the environmental impact of his work than anyone
else. The acceptance of this responsibility is the inescapable duty of the professional
antenna engineer.
16.10 References
10 HANSEN, W. W., and WOODYARD, J. R.: 'A new principle in directional antenna design',
Proc.IRE, 26,1938, p. 333
11 ISBELL, D. E.: 'Log periodic dipole arrays', IRE Trans., AP8,1960, p. 260 (also Univ. of
Illinois Ant. Lab. Tech. Report 39, June 1959).
12 JOHNSON, W. A.: "The notch aerial and some applications to aircraft radio installations',
Proc. IEE, 102B, 1955, p. 211
13 KNIGHT, P.: 'The selection of the optimum curtain array for a high-frequency broadcast-
ing service', Proc. IEE, 109, Pt. B, 1962, p. 91
14 KRAVIS, A., and RADFORD, M. F.: British Patent No. 884889. 1959
15 LADNER, A. W.: 'The series phase array', Marconi Rev., 67,1937, p. 1
16 MASON, H. P.: 'Some factors influencing the design of broad band HF monopole aerials',
IEE HF comm. Conv. ED4,1963, p. 114
17 MORRIS, D. W., MITCHELL, G., MAY, E. J. P., HUGHS, C. J., and DALGLEISH, D. I.:
'An experimental multiple-direction universally steerable aerial system for HF reception',
IEE HF Comm. Conv. ED4,1963, p. 121
18 MOXON, L. A.: 'High frequency antennae and propagation modes in relation to the
amateur service', IEE Ant. and Prop. Conf., Pt. 1, 1978, p. 83
19 NORMAN, F. J., and WARD, J. F.: 'Rhombic aerials, design charts for high frequencies',
Electronic & Radio Engr., 34,1957, p. 398
20 RADFORD, M. F., and WOLOSZCZUK, E. W.: 'New HF logarithmic aerials', IEE HF
Comm. Conv. ED4,1963, p. 141
21 RUMSEY, V. H.: 'Frequency-independent antennas' IRE Nat. Conv. Rec. 1, 1957, p. 114
22 SHARP, E. D.: 'Electromagnetic theory of wire grid lens HF antenna', IRE Trans., AP13,
1965,p.703
23 SINCLAIR, G.: 'Theory of models of electromagnetic systems', Proc. IRE, 36, 1948,
p.1364
24 TANNER, R. L.: 'New ideas in HF antennas', IRE PGAP Meeting, Stanford Univ., Jan.
1960
25 TANNER, R. L., and ANDREASEN, M. G.: 'A wire grid lens of wide application', IRE
Trans., AP10,1962, p. 408
26 WAIT, J. R., and WATERS, L. C: 'Influence of a sector ground screen on the field of a
vertical antenna', NBS Sci. Report no. 22. Oct. 1962
27 WATT-CARTER, D. E., and YOUNG, S. G.: 'A survey of aerials and aerial distribution
techniques in the HF fixed service', IEE HF Comm. Conv., ED4, 1963, p. 67
28 BENNINGTON, T. W.: 'Short-wave radio and the ionosphere' Iliffe and Sons Ltd, London,
1950.
29 DA VIES, K.: 'Ionospheric radio propagation' National Bureau of Standards Monograph 80,
Washington, 1965
Chapter 17
17.1 Introduction
17.1.2 Applications
The range of wavelength permits the use of antennas whose dimensions are of the
order of a half or one wavelength. The antennas are physically convenient and
mechanically simple, and are one of the reasons for the widespread use of these
frequencies.
The predominantly line-of-sight propagation characteristics of VHF and UHF
allow repeated use of the same frequencies within comparatively small distances,
and so make these frequencies well suited for area coverage systems. At the same
time, it is practical to construct antennas having moderate gain and directivity so
that these bands are also used for fixed point-to-point links. Typically, a complete
communication or broadcasting network, in an area where telephone lines are
726 VHF and UHF antennas
A/2
H
A/4
Whilst the simple construction of Fig. 17.1a will be suitable for temporary
measurement purposes, the dipole for long-term exposure is more likely to take the
form of Fig. 17.1Z?. The balun and centre of the dipole are encased in metal, thus
providing mechanical support at the base of the antenna, protection of the terminals
from the weather, and the facility for supporting the antenna by the balun. The
antenna thus formed is termed a sleeve dipole. For overall lengths of the order of a
halfwavelength the vertical radiation pattern can be regarded as identical with that
of a plain cylindrical dipole.
The equivalent circuit for the impedance of a sleeve dipole is shown in Fig. 17.2
in which ZA represents the impedance of each half of the dipole at the top of the
sleeve, XF is the shunting reactance due to the discontinuity at this point (normally
capacitive) and Z o and B are characteristics of the coaxial line formed by the antenna
element and the sleeve. When the length of each antenna element is approximately
a quarter wavelength the resonant impedance at A is RA =R0/cos2d, where ^ 0 is
the basic impedance of the antenna element alone. Fig. 17.3 shows measured
impedances at the top of the sleeve for a range of sleeve monopoles. In general, the
sleeve increases the radiation resistance and the capacitive reactance.
728 VHF and UHF antennas
The dipole in Fig. 17.1 employs a balun to permit feeding from coaxial line.
Below about 150MHz, quarterwave baluns are inconveniently long and it is prefer-
able to use either a lumped circuit type or a capacitance-loaded line. The latter,
though physically attractive, has a reduced bandwidth compared with the full
quarter-wave balun. Between 150 MHz and 1 GHz the folded balun or Pawsey
stub is generally the most suitable, (Fig. 17.4). At the highest frequencies the series
inductance of the line AB may be sufficient to unbalance the antenna and the
slotted balun is to be preferred. It can be seen that the balun provides a reactance
shunting the dipole impedance ;Shnitkin and Levy78 demonstrate means of improv-
ing impedance bandwidth by suitable choice of balun and dipole.
xf zOte z in z o ,e x f zA
Dipoles within the frequency range 30-1000 MHz will normally be self support-
ing and the length/diameter ratio will be smaller than the range plotted by
Schelkunoff for the cylindrical antenna. Resource has then to be made to measure-
ments on 'fat' dipoles such as those described by Brown and Woodward.10 Even
fatter dipoles take the form of cages such as Fig. 17.5 taken from Brown and
Stanier.9 These are 'full-wave' dipoles and it will be noted that at midband the
antenna is only 0.62 of a wavelength. Considerable experimental work on broad-
band dipoles has been done in Germany, for example by Lamberts and Pungs.58
Another variation of the broadband dipole is the bicone (Fig. 17.6). Practical
antennas have cone angles of between 30° and 70°, the larger angles corresponding
to lower characteristic impedances. It should be said here that the main limitation
on the use of such antennas is the band width of the balun used with them. There
are in fact very few applications for such wideband balanced dipoles apart from
antenna test ranges and EMC measurement facilities. Papas and King70 gives
impedance data for conical monopoles.
The folded dipole (Fig. 17.7a) is rarely used as a single element but is commonly
used in stacks for UHF land-mobile base stations, and as the driven element in Yagi
antennas. Apart from its electrical virtues in providing a wide range of input
impedances, the folded dipole has a low DC impedance to ground, which provides
lightning protection and permits the addition of electrical de-icing elements.
For dipoles close to A/2 the impedance Z = (1 4- a)2Zr, where a is a function of
VHF and UHF antennas 729
the radii px and p 2 of the driven and short circuited elements and the spacing d
between centres. When d is large compared with the radii, a is approximately
(Guertler30)
a = lOg(£// )
P2
Z r is strictly the impedance of the equivalent circular dipole, but near resonance the
0=0.0266 L
d=0.375D
—D
• 30i-
,0.266
o-
o
o
-40-
0.222
-80 100
50
resistance,!!
resistance will not differ significantly from that of the driven element and a first
approximation can be obtained by assuming Zr to refer to the driven element.
When the radiators are of equal radius, the radius R of the equivalent circular radi-
ator is approximately R = yfpd. For unequal radii, the equivalent radius is
l/ f/
R = pv J rjv .
where
P1/P2 = r?
When the length of the dipole is not near X/2, a more generalised approach must be
ZA'B -Z 0i tan9 Zo
9 rods
V4in(6mm)dia
on 8V2in (216mm)
circle
adopted. The equivalent circuit shown in Fig. 17.1b shows the basic dipole
impedance stepped up by the ratio (1 + a)2 and shunted by an impedance 2Zf,
where Zf is the input impedance of the short-circuited line formed by each half of
the dipole.
Zf = jZ0 tanJC/, where / is the length of the dipole arm, K = 2n/\ and Z o is
the characteristic impedance of the half dipole as a two-wire line. The resulting
impedance for the folded dipole is
2(1 +afZrZf
{l-¥afZr + 2Zf
Two points should be noted about the form of the impedance curve, of which Fig.
17.8 is typical.
(a) The reactance is positive for / < 0.15X, in contrast to the open dipole.
(b) The very high impedance for short element lengths makes the folded dipole of
little practical importance except for near-halfwavelength dimensions or very
narrow-band applications.
Coaxial dipole: Fig. 17.9 shows the basic coaxial dipole. The radiator A is con-
nected to the inner conductor of the coaxial transmission line, the outer of which is
!"•
o
o
1 +a:1
2000
1500 1
1000
c| 500
/
0
| / x
1//
-500
-1000 -
0.1 0.2 0.3 0.A 0.5
h/A
Fig. 17.8 Impedance of typical folded monopole
current flow on the latter. It is impracticable to obtain a very high Z o for the outer
coaxial line thus formed, so the bandwidth of the choke is limited to the order of
10%. The feed-point impedance of this form of dipole will approximate to that of
the centre-fed dipole of the same diameter, but will be modified by fringing
capacitance at the base of radiator A, especially if a dielectric weather seal is added.
X/A
A/A
Coaxial dipole with tuned decoupling: For frequencies below about 60 MHz, the
outer sleeve may be inconveniently long. Since the aim is to provide a high
impedance at BB, one solution is to shunt a shortened sleeve with a capacitor, the
combination being tuned to resonance at each frequency.
An alternative arrangement is shown in Fig. 17.10. This form is used on military
vehicles in the tactical VHF band 30-76 MHz, where the height of a full dipole
would be impractical but where it is desired to reduce the effect of the vehicle by
minimising current flow on its metal surface. A special triaxial cable is used, coiled
to reduce its height and capacitively tuned between the inner and outer braids
(Brueckmann11)
Discone: Broadband decoupling of the transmission line from the lower half
of the dipole is achieved by making the latter conical. The height of the assembly is
reduced by converting the upper half to a flat disc. The resulting antenna (Fig.
17.11) has an impedance dependent on cone angle and a substantially constant
VSWR above a low-frequency cut-off frequency determined by cone length. Fig.
17.12 shows variation of VSWR with frequency for several cone angles (Nail68).
Coaxial bicone: The broad-band decoupling described above has also been
734 VHF and UHF antennas
applied to an antenna widely used for the military UHF communications band on
ships and at military airfields. The lower section begins as a cone at the feed point,
but is completed by a cylinder to reduce its width. The upper section is usually
conical from the feed point, surmounted by an inverted cone. Both solid and cage
forms exist. A useful reference to such an antenna is Kitchen49 (see also Fig. 17.81).
• skin of
vehicle
25° discone
250
250
125 125
100
125
r 11
1 1— 1 1
1 1 1 1
1i tJ
1 1 .1
125
i 1 1
\c J 7
/ / / / / / / / / / / / / /
Fig. 17.18 Printed X/8 monopole
L - 150
500 dia.
500 dia.
short vertical radiator with horizontal top loading. In its simplest form for single-
frequency operation no capacitor is necessary. In this configuration the gain can be
comparable to the normal A/4 monopole. Experience shows that for maximum
efficiency large-diameter elements are necessary and good grounding of the vertical
element is of prime importance because of the very low radiation resistance of the
basic radiator. This will become apparent from the following analysis.
ill
= cot where X
2irfC
and
Z o = 138 log 10
matching pins
70
For transmission lines of this form the phase velocity can be considered equal to
the speed of light, so that the physical length I is given by
Bx - 360I/X
The vertical radiator, when it is short compared with the wavelength, can be con-
sidered as a constant-current element of radiation resistanceRo ~ I515(h/X)2 (see
Schelkunoff and Friis77 p. 309), where h is the height of the radiator. This is the radi-
ation resistance for zero end capacitance; as the capacitance is increased the current
742 VHF and UHF antennas
T V
-* = cos (90° -62) = sin cot"1 - *
140
(b) Annular slot: Two forms of annular slot are used (Figs. 17.20 and 17.21).
They differ in the position of the slot, and hence in application. Whilst the flush
mounted type (Fig. 17.21) may be more familiar, there are many applications for
which the pill box (Fig. 17.20) is more suited since it does not require a large cut-
out in its mounting surface.
250
\\vv\u
Fig. 17.28 Notched plate
f V\\
This discussion applies only to small annular slots having diameters up to 0.6X.
Within this range the radiation patterns are virtually those of a vertical monopole.
Johnson and Rothe39 have considered this antenna as the slot analogue of a
circular loop, the slot being backed by a radial transmission line short-circuited at
its centre. The slot is loop coupled into a coaxial line. Using the nomenclature of
Fig. 17.21, the slot impedance is seen as a resistance Rs shunted by capacitance C,
the latter given approximately by
C = 0.17816D loge Djh picofarads, D and h in centimetres
Rs is derived from the equivalent loop resistance
(1207T)2
Re by the equation Rs =
2/ve
Fig. 17.29 gives values of Rs against slot diameter D
The slot width has little effect on Rs
The impedance Z ( r ) of the short-circuited radial transmission line is in parallel
with the slot capacitance and is given by
744 VHF and UHF antennas
= _ h J0(Kr)N0(Kr0) - J0(Kr0)N0(Kr)
(r) }
r Jx(Kr)N0(Kr0)-J0(Kr0)Nx(Kr)
where / 0 , Jx and A^o, Nx are Bessel functions of the first and second kind respect-
ively, of orders 0 and 1. Z ( r ) is chosen to resonate with the slot capacitance. At
resonance the equivalent series resistance, R's of the slot becomes
1000
500
100
50
10
'0 0.5 D 10 1.5
X
Fig. 17.29 Radiation resistance of an annular slot radiating to one side only
Considering now the loop formed by the inner radius of the cavity, the feed line
and the top and bottom surfaces of the cavity, the mutual impedance of this loop is
given by
N0(Kr0)J0(Kl)-J0(Kr0)N0(Kl)
M =
N0(Kr0) Jx (Kr) - J0(Kr0)Nx (Kr)
where JU is the permeability, normally unity. The input impedance R at the feed
point is given by RR'S = (coM)2 hence
VHF and UHF antennas 745
N0(Kr0)J0(Kl) -J0(Kr0)N0(Kl)
R — R
S
[N0(Kr0) J0(Kr) -
This neglects the inductance of the feed wire for which capacitance compensation
may be required.
Cumming and Cormier22 give some useful design curves.
Whilst the design data given above refers to circular antennas, a square slot has
similar radiation characteristics and may be easier to construct.
(c) Notch antenna: The short, capacitance-loaded notch has been described in
Chapter 16. For frequencies above about 100 MHz, the use of self-resonant notches
provides wide bandwidth in convenient mechanical form. Fig. 17.28 shows a notch
on a printed-circuit board, the notch being coiled up to avoid a plate width greater
than X/4 which would introduce directivity in the azimuth plane. Experience shows
that the coupling between adjacent turns is negligible provided that the spacing is of
the order of four times the slot width.
feed point
L2
mouth
Two alternative methods of considering the general notch antenna, (Fig. 17.30)
will be discussed. Johnson40 considered it as the magnetic analogue of a monopole
on the edge of a metal sheet. Using Babinet's principle, the mouth or open-end
746 VHF and UHF antennas
admittance
.at open end of notch
shunt
capacitance -§
required
Several loops may be connected in parallel in the horizontal plane (Fig. 17.33),
giving an arrangement with substantially in-phase currents and good omnidirectional
coverage. Normally four loops are used, but three are possible at some sacrifice of
circularity.
One variation on the loop is the short notch cut into a horizontal disc (Fig.
17.34). The same constraints on diameter apply as for the loop. Since the notch is
short, it requires capacitive loading at its mouth and has a narrow (1%) instan-
taneous bandwidth although capable of tuning over at least 10% band. If the disc is
748 VHF and UHF antennas
feed point
< A/3
between their input currents whose amplitudes are equal. Whilst it would be possible
to feed from a common transmission line with an additional X/4 to one dipole this
leads to unequal power sharing since the junction impedances are unequal, unless
these impedances are conjugate. Such an arrangement results in a narrow bandwidth,
and a better system results from connecting the dipoles by equal transmission lines
VHF and UHF antennas 749
to a 3 dB coupler. In this form of the antenna, the pattern circularity is determined
by the VSWR at the coupler input ports.
Slotted cylinder: This antenna consists of an axial slot cut in a hollow conduc-
tive cylinder, the slot being fed from a parallel or coaxial line. The antenna may be
compared with a vertical stack of close-spaced horizontal loops connected to a
2-wire transmission line short-circuited at top and bottom.
A/A
to phasing
network
support each dipole on a horizontal yard arm projecting from a vertical mast, the
spacing being chosen to give the least perturbation. Carter16 gives methods of calcu-
lating the resultant field. This may result in an inconveniently large structure, and
one alternative is to use fat dipoles wrapped around the support mast. Fig. 17.35
shows one such arrangement using X/4 elements. It is only necessary to feed the end
pairs. Three pairs of elements, outer pairs fed, give 4- 3 dB gain on a A/2 dipole,
whilst 14 pairs, with only the outer pairs fed, give + 6 dB gain.
that a centre-fed stack with vertical symmetry will give maximum horizontal plane
radiation, whilst a progressively fed system can give a beam tilt which will be fre-
quency-dependent.
A vertical dipole close to a vertical mast will have a broad unidirectional pattern.
Two or three elements spaced round the mast will create an approximately omni-
directional pattern. To reduce their mutual coupling the elements may be staggered
in the vertical plane as in Fig. 17.37. If series, in-phase, feeding is used, the dipoles
will normally be folded to give a satisfactory match to a 50 £2 line.
A/2
pattern and so avoid gaps in the service area (Platts71). Main station antennas are
nearly always omnidirectional,but smaller antennas used for low-power relay stations
often have directional horizontal radiation patterns to satisfy particular service
requirements (Millard65).
0.24 A
dipole
0.8
V
4 I
0.6
0.4
*Cv
0.2 -
0
30° 60° 90° 120° 150° 180°
Fig. 17.41a Theoretical and measured horizontal radiation patterns (HRPs) of a vertical
dipole spaced 0.25\ from a triangular mast 1.0\ wide.
HRP calculated by Carter's method
HRP calculated by the infinite-plane method
— - — HRP calculated by the induced-current method
HRP calculated by the diffraction method
o o o o Measured points (averaged over 360°)
Table 17.1 Gain of dipole and finite, square, flat sheet reflector
Square of side L/A Spacing of A/2 dipole, S/X Gain w.r.t. dipole, dB
1 0.2 5.2
Large 0.215 5.8
1.5 0.775 7.0
or image dipole having a phase of 180° referred to the first dipole. The radiation
patterns, impedance and gain can be computed for the dipole pair, taking into
account the mutual impedance between them. If the conducting surface is finite,
diffraction from the edges of the sheet must also be taken into account. The edges
parallel to the dipole have the greater effect, that of the perpendicular edges being
VHF and UHF antennas 755
negligible if their spacing is slightly greater than the length of the dipole. Fig.
17.42 shows variation of radiation pattern with sheet width and indicates that a
comparatively small sheet can be an effective reflector. Fig. 17.43 shows variation
of impedance and gain with spacing for an infinite sheet and Table 17.1 gives some
typical gain figures for finite sheets. It will be noted that, if forward gain is the only
criterion, there is little advantage in increasing the sheet size beyond a certain
0.8
-* 0.6
LU
X\
\
0.A
0.2
0
30° 60° 90° 120° 150° 180°
0
Fig. 17.41b Theoretical and measured horizontal radiation patterns (HRPs) of a horizontal
dipole spaced 0.25\ from a triangular mast 1.0\ wide
HRP calculated by Carter's method
HRP calculated by the infinite-plane method
HRP calculated by the induced-current method
HRP calculated by the diffraction method
o o o o Measured points (averaged over 360°)
minimum dependent on spacing. The front-to-back ratio for a dipole at X/4 in front
of a sheet of width 2x is given in Fig. 17.44 and here, too, there is little sense in
using sheets wider than 2X.
Corner antennas: The gain of the dipole and single sheet is clearly limited to the
order of 6 dB. Higher gains can be achieved by folding the sheet to form a corner. A
756 VHF and UHF antennas
90° corner can have a gain of up to 13 dB whilst with a 60° angle a gain of 14.5 dB
is possible. As Fig. 17.45 indicates, these gain figures require a larger apex-antenna
spacing than is normally acceptable physically. As the corner angle is decreased the
length of the side has to be increased to give a good approximation to the
theoretical performance of infinite-length sides. For any reasonable bandwidth a
90° corner with IX aperture and spacing 0.35X represents the best engineering
solution. It may well be physically more desirable to stack two 90° corners one
above the other than to build a 60° corner with 1.25X spacing to achieve the same
gain.
axis of
symmetry
0.25 A
D/X
-2.215
field strength
axis of
Hsymmetry
Fig. 17.42 Effect of sheet width on pattern of dipole and finite sheet
7 = (2dl\)\oge(dl2ua)
Front-to-back ratio is much more sensitive than is forward gain to the wire spacing
and diameter. In some instances some backward radiation is desirable; a striking
example is the 24-dipole corner array used as the localiser antenna in high-accuracy
ir2
0.2 0.4 0.6 0.8 1.0
P/A
50
A0
CD 30
20
2 3
2x /A
Fig. 17.44 Front/back ratio for dipole and finite sheet
Short backfire antenna: Fig. 17.47 shows the essential features of the short
backfire antenna which was first described by H. W. Ehrenspeck.26 A good
theoretical explanation has not been forthcoming, but it seems probable that the
mechanism is as follows:
Consider initially a plane sheet and dipole feed. This belongs to the single-sheet
class of antenna discussed above and its gain cannot significantly exceed 6 dB. It
can be considered as a dipole pair plus the vector addition of radiation from the
sheet. A similar analysis would apply if the feed were a dipole-dish combination
but the illumination of the sheet would then be altered.
The addition of a rim has two effects. It alters the radiation due to the sheet by
modifying the edge condition and superimposes another radiation component at a
different point in space. It is not clear whether the edge or the surfaces of the rim
are the more significant; if the former, a loop similarly placed should have a similar
effect. There is no evidence from the literature that this has been tried.
The large increases in gain obtained by the addition of rims are only possible
provided that there is adequate illumination of the reflector; so this type of antenna
is limited to reflectors of the order of 2-2.5X diameter maximum. Narrow-band
gains of the order of 18 dBi are reported, but since these require optimisation of the
rim at specific frequencies, the broadband gain is probably of the order of 12dB.
Larger reflectors, with appropriate increase in gain, can be used with multiple feeds.
For circular polarisation crossed dipole- or tripole- and disc combinations may
be used.
VHF and UHF antennas 759
> 0.6 A
metal or dielectric
support
dipole
A recent paper suggests that the gain and sidelobe level can be improved by the
use of rim corrugations Fig. 17.48. Fig. 17.49 shows measured E-plane patterns.
D
3A A/2
D
JUI
A/4i A/4
60 50 40 30 20 10 0 10 20 30 40 50 60
The illumination of the dish has to be tapered to reduce sidelobes and loss of
gain due to spillover. Typically an edge illumination lOdB below peak would be
used if maximum gain is the main criterion, but — 20 dB would be preferred for
minimum sidelobe levels. The distance F from the focus to the dish increases with
angle off-axis, so that energy at an angle 6 is attenuated by 20 log (sec2 0/2) dB,
compared with the energy on axis. This has to be considered in the design of the
feed radiation pattern, and it will be found that, for small F/D, it is difficult to
obtain an appropriate feed pattern with a dipole/reflector combination. In conse-
quence, the gain will rarely be more than 65% of that obtainable from a uniformly
illuminated dish of the same size.
Energy from the feed is reflected back into the feed from the area adjacent to
the vertex of the dish. The reflection coefficient due to the dish is F = gXjAnF,
where g is the axial gain of the feed. ^ _ _
A small circular plate radius yFX/3 placed A/24 in front of the vertex will
reduce the mismatch, but is rarely used now because of its narrow bandwidth.
The slotted feed (Fig. 17.50) which is often used for dipole/reflector combi-
nations provides an impedance transformation whose effect has not always been
762 VHF and UHF antennas
recognised. If the inner of the coaxial line is continued into the slotted section, it
is obvious that its effect is modified by the short circuit to one side of the slotted
section. In effect the dipole impedance is transformed by a line of impedance 2Z 0
shunted by a short circuited section also of 2Z 0 as shown in Fig. 17.51a and b. An
assumption has been made here that the slot is X/2 long and itself has no effect on
the dipole impedance. This is true so long as the slot is not covered by a dielectric
which would modify its effective length.
^reflector
dipole
e
/ / \
/ N
2Z 0
Zo 2Z 0
02 B
ZA
Ao-
Gains compared with a half-wave dipole of 5.5 dB for a director and 5.2 dB for
a reflector can be achieved. High gain at close spacing results in very low input
resistance to the driven element, of the order of 10-15 0 at 0.1X spacing. A folded
dipole (see Section 17.2) is often used to improve the match to the transmission
line, but small bandwidths are usual with such small spacings. If operation over
bandwidths up to 20% is required, a dipole and reflector combination at 0.15 X to
0.2X is to be preferred.
764 VHF and UHF antennas
The use of full-wave dipoles and tuned reflectors is uncommon, but was the basis
of a wide range of antenna systems developed in Germany for such diverse uses as
500MHz airborne radar systems and 41-68 MHz point-to-point links. The basic
element of the latter system is shown in Fig. 17.53. It consists of a 'fat' (D — 0.07 A)
driven element, 0.7X long with a rod reflector at 0.3X spacing. The use of full-
wave dipoles permits metallic supports at the centre point of each arm, which is
electrically dead. The reflector and dipole spacings were adjusted to give compro-
mise performance over the band 41-68 MHz. A VSWR of 1.3 at 60 £2 was achieved
with a gain of 12 dB over the band.
Because of the need for a metal surrounding surface it is not possible to use slots
and notches in end-fire as is done with dipoles. They can only be used in broadside
array to skew the beam from the normal broadside position. Such parasitic slot
VHF and UHF antennas 765
systems have been used on aircraft to provide unidirectional radiation, the tuning of
the parasitic slot being achieved by a combination of its dimensions and those of
the backing cavity.
Similarly, whilst it is possible to have an end-fire parasitic array of horizontal
loops, there is no advantage in using parasitic vertical loops — a dipole system
would be smaller and more directional.
JL JL 7 3A
0.625A
3 reflectors
omitted for clarity
0.7A
Yagi-Uda array. The designs of Prof. Uda of Japan were first described in
English by Mr. Yagi, so this type of parasitic array should bear their joint names. It
is a logical extension of the use of parasitic dipoles with one driven element, in end-
fire array.
As Walkinshaw88 demonstrates, the most effective arrays use a number of
directors but only a single reflector. Whilst in theory the forward gain can be
increased indefinitely by increasing the number of directors, in practice there is
little to be gained by using more than 12. If higher gains are required it is better to
use arrays of Yagi-Uda arrays; this has the double advantages of:
(a) a structurally better system
(b) independent control of sidelobe levels by adjustment of array spacing. The
diminishing returns from multi-director Yagi-Udas is to be expected since successive
elements have smaller amplitude currents induced in them.
In general, the addition of parasitic elements reduces the radiation resistance of
the driven element so that a folded dipole is necessary to match the impedance to
normal transmission lines. There is one exception in the combination of one driven
766 VHF and UHF antennas
element and four directors each spaced 0.2X where the input impedance is approxi-
mately 45 O and no folding is required.
For mechanical convenience the parasitic elements are normally mounted on a
central metal tube which, since it lies along the neutral axis of the system, does not
perturb the performance apart from a slight modification of the effective length of
each element.
Normally Yagi-Uda antennas have narrow bandwidths, of the order of 2% for
optimum gain arrangements. Some considerable improvement in bandwidth can be
obtained at the sacrifice of some gain. One method of doing this is to design the
directors for the upper end of the desired band and the reflector for the lower end
of the band. Two five-element Yagis (one reflector, three directors and one driven
element) of comparable size, one optimised and one broad band, had gains respect-
ively of 8 dB over a 9% band and 4 dB over a 60% band, referred to a half-wave
dipole.
Since the effect of a parasitic element is a function of its length and diameter,
any change to these will affect its performance. One natural hazard is the formation
of ice, which modifies the effective diameter of the element and hence its reactance.
Experiments in Germany showed that it was possible to reverse the direction of the
main beam by ice loading, the effect being more severe on the thinner elements of a
500MHz Yagi than on those of a 220 MHz design. It seems unlikely, therefore, that
icing would be a severe problem below the UHF band.
The dielectric constant of ice in the VHF and UHF bands is approximately 3. A
fair simulation of the effect of ice can be achieved by putting glass fibre tubes
having a diameter equal to the anticipated ice, over each element of the array.
Fig. 17.54 Quad antenna derived from pair of dipoles and reflectors
Cubical quad: This antenna is little known outside the amateur ranks but has
some practical advantages over the simple Yagi. The derivation of the antenna from
a stack of two horizontal Yagis is shown in Fig. 17.54. Fig. 17.55a shows a typical
3-element array with sides of approximately X/4. In order to tune the reflector and
director a short-circuited and open-circuited stub, respectively, can be inserted in
the arms parallel to the feed point Fig. 17.55Z?. The gain of the 'quarter-wave' quad
VHF and UHF antennas 767
with three elements is 7-7.5 dB, but a 'half wave' quad (sides approximately A/2)
can have a gain of 12.5 dB.
The quarter-wave quad is attractive in its reduced width compared with a Yagi,
and is said to possess wide bandwidth and to be dimensionally non-critical.
readily explained by consideration of the log periodic dipole antenna shown in Fig.
17.56. This is only one of many forms of log periodic antenna but is one of the
more suitable physically in the VHF and UHF bands.
The antenna consists of a number of centre-fed dipole elements connected to a
twin transmission line so that the phase is reversed at each connection relative to
the adjacent elements. The element lengths and spacings are related as shown. The
antenna is fed at its high-frequency end and the twin transmission line is short-
circuited behind the longest element. For convenience, a coaxial cable can be taken
through one of the hollow twin lines.
Consider the frequency for which the «th element is half-wave resonant. This
element will have the lowest impedance of the array and will therefore abstract
more energy than any other element. It will be obvious that the smaller (higher-
frequency) elements will present a higher impedance and the energy abstracted will
progressively increase to the nth element and decrease beyond it. The active
part of the antenna is restricted to a few elements either side of the resonant
one. The possibility of effects due to 3X/2 elements can be excluded since very
little energy travels along the transmission line beyond the half-wave resonant
element. Increasing the frequency merely moves the active section towards the
high-frequency end of the antenna. The antenna can, therefore, be considered as a
series of short antennas optimised at the resonant frequencies of successive
elements. The degradation of performance between optimised frequencies depends
on the periodicity of the antenna and has to be weighed against an excessive
number of elements. The lowest operating frequency will be determined by the
length of the largest element, the highest frequency by the smallest. Energy travel-
ling beyond the longest element will be reflected back by the short circuit and the
antenna will not exhibit the single-lobed radiation pattern which is characteristic of
the operating frequency band. In the range 30-1000 MHz practical arrangements
of elements do not limit the upper frequency performance; a typical arrangement
is shown in Fig. 17.57.
The significant factors are r, the ratio between adjacent element lengths and
spacings, and a, the spacing in wavelengths between one element and its next,
smaller neighbour. Currel15 derives design curves relating r and o to gain and has a
useful series of nomographs giving structure bandwidth, structure length, and num-
ber of elements in terms of these two parameters. The useful range of r is shown to
be between 0.76 and 0.98 for values of o between 0.06 and 0.20. Gains are between
8 and 12dB, the latter only being achieved for values of r near the upper limit,
where clearly the active region includes more elements.
The mean resistance Ro of the antenna is shown to be a function of the
transmission-line characteristic impedance Z o which has little effect on the direc-
tivity. As might be expected there is a reduction in directivity with thinner
elements since their impedance change is more rapid with frequency. The VSWR
referred to Ro tends to rise withZ 0 and with decreasing o. Typically VSWR values
of 1.4 are achievable for a wide range of parameters. The operating bandwidth is
given approximately by
VHF and UHF antennas 769
B = BjBa
where B8 is the structure bandwidth and B^. is the bandwidth of the active region.
£9 = T1-N where N is the number of elements and Bar lies in the range 1.2 to 30
being lowest for large r.
Further design details can be found in Smith.84
The details given above refer to structures with half-wave straight dipoles. Mayes
and Carrel63 have shown that vee-dipoles operating in the 3X/2 and higher modes
can provide higher gains; straight dipoles would have undesirable multilobed pat-
terns and cannot be used. The arms of the vee are sloped towards the high-frequency
end of the structure. Gains of 12 dB and 18 dB, respectively, have been reported for
3X/2 and 5X/2 vee dipoles but the increased disparity between /f-plane and//-plane
patterns may be a disadvantage.
Such increased length elements are clearly impracticable for frequencies as low
as 30MHz, and attempts have been made to reduce the element lengths by end
loading. Difonzo34 gives a number of examples, the most significant of which are
shown in Fig. 17.58. It is also possible to add series loading in the form of a coaxial
line as shown in Fig. 17.59 and at lower frequencies a series coil has been used.
Although the discussion to date has been entirely on planar types, these are by
no means the only type of log periodic. The more general pyramidal form consists,
as Fig. 17.60 shows, of two sets of elements in inclined planes. By variation of the
angle between the planes, the E- and H-plane radiation patterns can be made more
nearly equal than with a single plane array. More variety in elements is possible and
the three most used forms are shown in Fig. 17.61. These are known as the
trapezoidal, toothed and zig-zag types.
770 VHF and UHF antennas
disc tee
in
I °
plane of
short circuit
b
Fig. 17.59 Series loading o f log-periodic element
(a) Internal (b) External
For directive antennas only the axial mode is of importance. This arises when
the circumference C of the helix is approximately one wavelength. As a general
guide, C lies within the range 0.75A< C< 1.33X for the axial mode. The on-axis
radiation is essentially circularly polarised provided that the helix has more than a
few turns. Degradation with the helix of few turns arises because of the wave
VHP and UHF antennas 771
reflected from the free end which produces opposite-hand circular polarisation on
reflection at the ground plane.
The characteristics of the axial mode helix are summarized below:
_ / . 90
Radiation pattern E = sin cos 0
n ]J sin 0///2)
Practically the helical antenna requires a dielectric support structure which must be
minimised to prevent modification of the phase velocity along the helix which
would alter the electrical characteristics. Two possible arrangements are shown in
Fig. 17.63. The ground plane should be at least X/2 in diameter. The diameter of
the conductor does not appear to be critical between 0.05 X and 0.006X. The
arrangement of the feed portion from the ground plane appears to affect both
impedance and axial ratio. If the feed point at the ground plane is on the helix axis,
the feed wire should have the same angle as the pitch angle (Fig. 17.64).
A full analysis of the axial-mode single-conductor helical antenna is given in
Kraus53 Chapter 7, and further details are in Chapter 7 of Vol. I.
zigzag
toothed
beams and better front-to-back ratios. Two methods have been applied to the pylon
slot antenna. In the first, two parallel axial slots are cut in the cylinder and fed in
phase. A spacing of 0.4X in a cylinder of D/X = 0.6 produced a 3 dB beam of 70°
width and a front-to-back ratio of 12 dB.
C -TTD
dielectric supports
(typical )
For smaller diameter cylinders the addition of dipole elements to the edges of a
single slot has proved effective. The proportion of power in slot and dipole is
altered by adjusting the length and position of the dipole elements. An optimum
arrangement in a cylinder of D/X = 0.1 gives a beamwidth of 110° and a front-to-
back ratio of 10 dB. Operation over the frequency band 87-100MHz with a VSWR
of 1.3 has been reported (Bosse7).
B
33.
The elements themselves can be any of the directional antennas described above.
Directional elements are preferred to omnidirectional ones because they are less
affected by support structures and the mutual coupling between adjacent elements
can be neglected. In calculating the pattern of an array of elements around a mast,
for example, it is essential to know the phase centre of each element. It may be
VHF and UHF antennas 775
necessary to measure this for elements such as Yagis or log periodic antennas for
which it is not readily calculated. Fig. 17.65 shows patterns for arrays of full-wave
dipoles plus plane reflectors on two sides of a mast, for various phase shifts between
the arrays.
Many communication and navigation systems used on vehicles operate within the
VHF or UHF ranges since the propagation characteristics permit the use of low-gain
omnidirectional antennas physically well suited to the vehicle constraints. The term
'vehicle' here applies to any means of transportation and includes the following:
Aircraft
Wheeled and tracked vehicles — iand mobile'
Marine vessels
Railway trains
Guided missiles
Rockets
Space vehicles and satellites
Hovercraft
Pedestrians and animals
In all instances the performance of the antenna will be significantly modified by
the vehicle on which it is mounted. Whilst the antenna itself may be simple in elec-
trical design, the combination of antenna and vehicle constitutes an electrically
complex structure. An insight into the influence of the vehicle can be obtained
by consideration of antennas mounted on or adjacent to certain regularly-shaped
conductors — flat sheets, cylinders and cubes are the most relevant.
Whilst analytical methods exist for some of the cases - the circular sheet, and
the circular cylinder — recourse has to be made to numerical methods such as wire
grid modelling or GTD for less regular shapes.
It follows, therefore, that only in those systems where the electrical require-
ments are least demanding will it be possible to provide a single antenna design
suitable for a wide range of vehicles. The choice of antenna may have an impact on
the structure or the performance of the vehicle, and it is therefore highly desirable
to predict the antenna performance at an early stage of the design both of the
antenna and the vehicle.
The general procedure should be as follows:
(i) Definition of shape of vehicle
Definition of radio installations required ('radio fit')
Definition of operational constraints
(ii) Formulation of required antenna coverage from radio system and operational
requirements
(iii) Study of possible antennas to satisfy (ii).
(iv) Construction of antenna installation diagram showing preferred arrangement
of antennas
(v) Submission to vehicle constructor or user for approval
(vi) Modification of (iv) until agreement reached with user
(vii) (a) Numerical analysis of antenna performance or (b) scale modelling
(viii) Engineering of chosen antennas if no suitable existing types
(ix) Full-scale trials with engineered antennas
It will be noted that up to step (vii) the basic antenna characteristics are assumed
and it is only after this stage has been completed that any necessary engineering is
undertaken. There will, of course, be exceptions when more basic investigations are
necessary because existing knowledge is inadequate.
It should be noted that the use of this technique for calculating radiation
resistance does require care, particularly with short antennas. Small errors in the
phase of currents close to the feed point can lead to large errors of magnitude in
the computed radiation resistance.
340
30
280
260° ioo°
200° 160°
180°
for the wave nature of light. These rays emanate from points at infinite curvature
(edges, tips etc.), when present on the scatterer. Other diffracted waves (creeping
waves) are waves which travel along smoothly curved surfaces shedding diffracted
rays tangentially into the surrounding medium. This theory thus enables predictions
to be made for the partially illuminated region behind an obstacle.
GTD can be used to construct the distribution of field intensity, polarisation and
phase throughout space for an antenna mounted on a structure, by tracing rays of
energy from the antenna incident on the scattering structure and then to a distant
field point. Each ray is assigned a field value, the phase of the field of the ray is
assumed proportional to the optical length of the ray from a datum and the
amplitude is assumed to vary in accordance with the energy-conservation principle.
Since GTD is based on an asymptotic high-frequency solution of Maxwell's
equations and the waves incident, reflected and diffracted on or from a scatterer
are assumed plane, this ray-tracing technique is applicable to, and has been used on,
structures which are large compared to the wavelength of excitation.
Fig. 17.67 shows a selection of the rays which have to be considered on a com-
plex structure such as an aircraft. It should be noted that these modelling techniques
still have some difficulty in the axial direction of cylinders. This arises from the
importance of the end conditions and the increased ray complexity required for
accurate modelling.
VHF and UHF antennas 781
-5
d
8 -1°
c
o
u
S -15
l-ao
-25-
Fig. 17.68 Impedance of 0.224X monopole 0.006X diameter on circular ground plane of
radius a
• • • • Meier and Summers results
15 20 25 30 35 40 45 50 60
9=0°
yaw
plane
pitch
plane
0=180°
roll plane
The practice of designing antennas on flat ground planes when they are intended
for use on three-dimensional bodies has little to recommend it except cost. It makes
much more sense to use a metal cylinder or cube of size appropriate to the ultimate
vehicle. As indicated above, if the cross-section is sufficiently large the length of the
body need not exceed one wavelength. This choice of a proper ground plane for
antenna development can be important in the determination of performance
characteristics. For example, a broad-band VHF aircraft antenna gave VSWR figures
of 4 at the band-ends against a specification of 2.5, but investigation showed that
the impedance had been measured on a 1.8 m square ground plane whilst the
antenna had been developed on a 2.4 m-diameter aircraft fuselage, on which the
specification figures were consistently met.
Further emphasis to the variation of impedance on a finite body is given by Fig.
17.69 which shows base impedance of a VHF monopole for seven positions on a
saloon car.
VHF and UHF antennas 785
bottom
Fig. 17.71 Roll-plane pattern of centre roof VHF monopole on car. Free-space pattern
Pattern dependence is not limited to the end directions as Fig. 17.72 indicates.
This is a pitch plane pattern of a longitudinal horizontal dipole below a cylinder 4 A
long and 1.0X diameter, for different positions of the dipole. The strong radiation
scale in DB.
below
antenna position
— 1.15 A from nose
1.91 A "
2.67 A "
• AA
nose i
""*—*~ below
in the nose and tail directions (end-on to the dipole axis) should be noted; this
would not occur if the dipole were mounted below a plane reflector and is a conse-
quence of skew currents on the cylinder. If the antenna is mounted on the cylinder
axis at one end of it, the combined structure may be considered as an asymmetric
VHF and UHF antennas 787
dipole whose patterns (see Jasik38 p. 3.16) depend on the relative lengths of the
two arms. The exact geometry of the adjacent end will have a marked effect if the
cylinder diameter is electrically large.
(a) Physical: These impose limitations on the height and size of antennas for
any of the following reasons:
Ground clearance, e.g. underneath aircraft
Overhead clearance: road vehicles, trains, helicopters
Hazards to vehicle operators and passengers
Interference with other vehicle functions
Launch-vehicle constraints: satellites, missiles
Concealment of vehicle
Disguise: plain vehicles for law enforcement
(b) Mechanical
Aerodynamic drag
Wind loading
Ice loading
Wave force: ships, submarines
Damage by cleaning equipment: cars, trains, public-transport vehicles
Weight: aerospace vehicles
(c) Environmental
Vibration
Shock, e.g. satellite launch, gunfire
High temperature
Low temperature
Effect of ice on electrical performance
Low pressure causing flashover and corona
Contamination by fuels and lubricants
Some of these constraints will be obvious on reflection, but others arise from the
possibility of consequential damage. For example, an antenna may be made
788 VHF and UHF antennas
sufficiently strong to withstand ice build-up in flight on an aircraft, but ice breaking
off the antenna may cause damage to engines and other parts of the structure.
Similarly the breakage of an antenna through vibration not only detracts from the
radio-system performance but the broken element may lodge in a moving control
surface and hazard the aircraft safety. It may also cause damage on the ground if it
falls clear of the aircraft.
The effect of low pressure is clearly seen from Fig. 17.73 in which the power
handling of a specific antenna is related to air pressure. The very low powers which
cannot be exceeded without breakdown represent a very real constraint on system
performance of high-altitude vehicles.
100
50
a
% 20
c 10
O
1
k
-L. J- -L. J
0.01 0.02 0.05 0.1 0.2 0.5 1 2 5 10
air pressure,mmHg
Fig. 17.73 Typical power breakdown as a function of air pressure. XIA bent monopole
(a) Cars and vans: In Section 17.4.2 it was shown that the roll plane pattern of
an antenna on a saloon car, measured in free space, indicated that contributions
from the whole vehicle had to be included. It must be concluded that, when the
vehicle is on the ground, currents will flow into the ground through the wheels, or
VHF and UHF antennas 789
tracks in the case of a tracked vehicle. The impedance to ground through these
paths will, therefore, be important in determining the radiation patterns of an
antenna on a land-based vehicle, and in some circumstances may also affect the
feed-point impedance of the antenna. A simplified model of the system might be
Fig. 17.74 which suggests that the vehicle might be considered as a sleeve mono-
pole or, if the base impedances are high, as an asymmetric dipole.
Clearly the influence of the ground will decrease as the vehicle becomes elec-
trically large. Webster90 indicates that in the 30-76 MHz band currents throughout
the body and the wheels of military vehicles were significant. Some workers have
claimed that on motor cars effects can still be seen at 150 MHz with antennas on
the upper surface, but there is little published data. At frequencies of this order,
the capacitance to ground of the lower surface of a car must result in a low
impedance between vehicle and ground, so that the model is simplified further by
short-circuiting Z in Fig. 17.74. For practical purposes the effect of varying ground
constants can be neglected above about 100 MHz.
For terrestrial communications systems only a narrow range of angles above the
horizon is significant. Davidson23 shows that in a heavily built-up area propagation
into streets from an elevated base station may involve elevation angles up to 20°,
the dominant mode being over the roofs and by diffraction into the streets below.
Since most mobile antennas have broad elevation patterns, this effect only becomes
of importance in higher UHF bands where some elevation gain would be possible
through vertical stacking.
Again, it has been shown by Lee59 and Stidham86 that severe fading in cities can
be reduced by the use of antennas with horizontal directivity, provided that the
beam maximum is pointed towards the base station. For a mobile whose direction
is constantly changing this suggests that some form of adaptive antenna would be
desirable, but, except for special cases such as the need to maintain continuous
communication with high data rates, the complexity of such systems is rarely
warranted. These results do suggest, however, that for purely city use an insistence
on very good omnidirectional coverage is unnecessary. This does not apply in more
open terrain; field trials in town and country with good and poor horizontal
coverage antennas clearly show the need for good omnidirectional coverage for
rural use.
Measurements on cars in the 80MHz, 150MHz, 450 MHz and 900 MHz bands
show that best coverage is obtained with a centre roof antenna. A typical pattern is
shown in Fig. 17.75 which should be contrasted with Figs. 17.76 and 17.77 for
front and rear roof positions. It should be noted that the front position has maxi-
mum signal to the rear and vice versa; pitch plane patterns such as Fig. 17.78 show
considerable distortion and help to explain this paradox.
Wing mounting gives more distorted patterns as might be anticipated (Fig.
17.79), and the mean gain is about 3 dB below that for roof positions.
Measurements on delivery vans show less variation in pattern with roof position.
The radiation patterns for roof-mounted antennas on cars do not correspond to
those on flat ground planes of the same size. Strong currents can be measured in the
790 VHF and UHF antennas
roof pillars, and these clearly influence the patterns and explain why the patterns
on metal-sided vans are less sensitive to antenna position.
Antenna types: Antennas for use on cars and vans are usually X/4 whips of very
thin (1.5 mm diameter) spring steel. 5X/8 or 3 X/4 whips can be used in the UHF
band; these have a loading coil at about X/4 above base to give cophased currents
along the whip. As noted by Davidson and Turner,23 the full gain of such antennas
may not be realisable in built-up areas.
front
Whilst the centre-roof position is clearly the best for electrical performance
some mobile operators prefer not to drill holes in car roofs, and fit antennas to the
wings instead. Some use has been made of wing-mirror antennas which combine
two functions. Essentially these are short monopoles, insulated from the car body,
and incorporating a preset matching section in their bases.
Although their very low cost makes whip antennas attractive for the civil user,
they are vulnerable to damage by several causes:
Vandalism
Low overhead clearances: bridges, garages
Automatic car washes
VHF and UHF antennas 791
Alternative low-height antennas that can be used are described in Section 17.2.1
under the heading 'Vertically polarised monopoles'. The 'low profile' (Fig. 17.19)
and annular slot (Fig. 17.20) have been widely used for public-service vehicles.
front
(b) Other wheeled civil vehicles: Vehicles which do not have continuous metal
upper surfaces present special problems. Examples include fire appliances, flat-bed
lorries, earth-moving machinery, tractors and articulated vehicles. Each of these has
to be considered separately. The increased height of some gives a height-gain which
compensates for the irregular radiation patterns. Measurements on a whip antenna
on a fire appliance at about 80 MHz showed a mean gain of at least 6 dB over a
similar roof-mounted antenna on a saloon car.
On some vehicles it may be more appropriate to use some form of dipole in the
absence of a well-defined groundplane. The coaxial dipole Fig. 17.9 has commonly
been used in such applications.
Antennas are normally X/4 whips with spring-loaded bases to minimise damage
from obstacles. The need to match the antenna at any frequency within the band,
typically 30-76 MHz, demands the use of complex tuning units which can accom-
modate the wide range of impedances encountered on a variety of vehicles, none of
which provide a large ground plane. Coaxial dipoles with tuned decoupling (Fig.
17.10) are commonly used; their impedances and radiation patterns are less
sensitive to the shape of the vehicle.
Since it is common to fit two or more radio systems to the same vehicle, the
provision of additional selectivity, either in the tuning unit or by separate tunable
filters, is common practice to reduce the frequency separation between equipments.
Whip antennas, typically 1.8 or 2.5 m long, have a number of disadvantages:
Vulnerability to damage by obstacles, gunfire or explosives. Visibility when the
remainder of the vehicle is concealed.
Various low-profile antennas have been considered. These are generally of the
top-loaded monopole form in which the top-loading is a metal plate on a short
vertical stub, the whole antenna being supported by a reinforced plastic fairing.
Such antennas have very low radiation resistances, and since they are essentially
capacitive their tuning-unit losses are high and the antenna efficiency may be at
least lOdB below that of a tuned monopole. One variant of Fig. 17.19 which
VHFand UHF antennas 793
appears to combine the merits of robustness, low height and good efficiency is
shown in Fig. 17.80. The conventional form can be tuned over the whole frequency
band but only at the expense of low azimuth gain at low frequencies where a large
capacity loading is necessary. By shorting out parts of the top loading coil the range
of capacitance is reduced and the azimuth gain does not fall below — 5 dB com-
pared with a tuned monopole. The whole structure is supported within a thick
GRP cover having an overall height of less than 0.5 m. The relatively high Q of this
antenna enables two separate equipments on the same vehicle to be operated with
frequency separations of no more than 12%.
\
\
front
bottom
Fig. 17.78 Pitch-plane pattern of VHF monopole on front roof of car. Free-space pattern
typical form, the shape of the lower cone being modified to minimise currents on
the transmission line. A particular design feature is the use of large insulators not
only for strength but to reduce the effect of encrusted salt and funnel effluents on
the performance of the antenna.
Fig. 17.79 Yaw-plane patterns of VHF monopole on front and rear wings of car
17.4.6 Aircraft
This section encompasses antennas for helicopters and fixed-wing aircraft. The
antennas will be considered under the following headings relating to the radio sys-
tems to which they belong:
Communications
Identification and air-traffic control
Telemetry
Navigation
Homing
antenna element
variable capacitor
For the majority of aircraft, therefore, the coverage requirements will be limited
to the azimuth plane and to elevation angles limited by normal aircraft attitude
changes. Angles of 4- 5° to — 30° with respect to the aircraft horizontal axes should
be adequate for most non-aerobatic aircraft. Clearly larger angles must be considered
for more manoeuvrable aircraft and the specification modified to suit individual
requirements.
796 VHF and UHF antennas
upper cone
insulator
-lower cone
above
300 MHz
The wings can be considered as an extended ground plane and their effects will
be a combination of reflection and diffraction, the former dominating if the
antenna is at significant height above or below the wing surface. To minimise these
effects fuselage positions forward or aft of the wings are preferable.
For upper fuselage positions, shadowing by the tail fin has to be considered. Fig.
17.84 due to Cary (unpublished) shows the effect of an elliptic cylinder for parallel
798 VHF and UHF antennas
above
nose 4—tail
300 MHz
Fig. 17.83 Pitch plane patterns of vertical monopole under aircraft fuselage
Under forward fuselage
Under rear fuselage
the limitation of currents down the leading and trailing edges of the tail fin which
will normally give rise to interference lobes. For VHF communications in the
100 MHz band this is a very good position. For higher frequencies the cable attenu-
ation may outweigh the advantages.
source of vertical
polarisation radiation thickness
vibrating parallel to
streamlined cylinder
Fig. 17.84 Effect of a streamlined obstacle on radiation pattern of a dipole parallel to its
own axis
(a) Horizontal polar diagram with streamlined cylinder of 0.1 X thickness, 1.0X
chord
(b) Horizontal polar diagram with streamlined cylinder of 1.OX thickness, 10.0X
chord
r = 10\
Frequencies between 30 and 100 MHz are mainly used for military communi-
cations with mobile ground forces, so tend to be required only for helicopters and
small aircraft. A maximum length of under 20 m means that at the lower frequen-
cies the fuselage is never more than 2A and may be significantly less. When, as is
often the case, the fuselage cross-section is non-uniform, unequal division of current
on the fuselage forward and behind the antenna results in distorted and very
frequency-dependent radiation patterns. Fig. 17.86 illustrates the rapid change in
yaw plane patterns on a centrally mounted helicopter antenna, described in more
detail by Burberry and Kelly.13
Apart from this problem of siting, there is a further difficulty in this frequency
band since wavelengths are too long to permit the general use of self-resonant
antennas. The alternatives are short, inductively-loaded monopoles such as that
described by Cooper19 and shown in Fig. 17.87, bent-sleeve monopoles with tuning
800 VHF and UHF antennas
units, or variants of the transmission line antenna (Fig. 17.19). Inevitably, gain at
the lower end of the frequency range is low compared with a A/4 resonant mono-
pole, probably about — 10 dB for the inductively loaded types and — 6dB for
capacitive loading. At the upper end of the band efficiencies approach that of the
monopole.
Quarter-wave whips have been used on the slower helicopters (speeds less than
about 150km/h) but have to be mounted at the tip of the tail fin or underneath the
fuselage with some means of retraction. Simple switched matching systems are
possible and gains are comparable with resonant monopoles.
starboard
nose 0
port
antenna
Another possibility is to use the tail fin as the antenna, treating it as a folded
monopole or shunt antenna (Fig. 17.88). The radiation patterns are very much a
matter of chance and each installation must be treated on its merits. It seems prob-
able that successful operation will only be possible where (a) there is no rudder, (b)
the chord of the fin is less than A/4, and (c) the fin height is less than A/2. Some
form of tuning unit will be required.
The large frequency band, together with the use of antennas very much physic-
ally constrained, necessitates tuning systems. In the simplest type band-switching
may be possible, but in others tuning at each frequency is essential. If frequency
VHF and UHF antennas 801
information is available from the radio equipment some pre-setting of the tuning
elements can be achieved whilst the transmitter is tuning, final trimming if necessary
requiring RF power from the transmitter. This method does permit silent tuning for
reception. In the absence of frequency information, antenna tuning must have RF
power from the transmitter and this results in longer tuning times.
\
\
port \/\starboard
antenna
Satellite communication: Whilst VHF and UHF systems for air-to-ground com-
munications via satellites do not require high antenna gains the need for circularly-
polarised hemispherical coverage cannot readily be achieved on an aircraft with a
single antenna because of physical constraints. Most systems have in fact been a
compromise between high-angle circular polarisation and low angle vertical
polarisation. A typical arrangement (Fig. 17.89) consists of a pair of crossed slots
with shallow cavities for the high-angle cover linked with a vertical monopole for
the lower angles.
802 VHF and UHF antennas
dielectric fairing
however, still be met by monopole antennas, the small wavelengths permitting the
use of highly efficient tuned monopoles and of annular slots (Fig. 17.21), where
flush antennas are necessary. In this respect it should be noted that the weight of an
annular slot is of the order of 1 kg compared with perhaps 0.20 kg for a monopole
of good aerodynamic design, and there will rarely be an advantage in using the flush
antenna.
upper surface
of fuselage
port starboard
tail
Fig. 17.90 Typical yaw-plane pattern of 1 GHz vertical monopole centrally mounted under
large aircraft fuselage
DME and ATC systems essentially require coverage below the azimuth plane,
and a single antenna on the underside of the fuselage will normally suffice (Fig.
17.90). IFF requires some upward coverage for air-to-air interrogation as well as to
804 VHF and UHF antennas
allow for aircraft manoeuvre. The addition of a second antenna above the fuselage
is not adequate since the tail fin will give severe rearward shadowing. A better
arrangement is to have one antenna on the upper surface of the nose or even in the
feed point of
driven notch
Fig. 17.92 Triple notch in tapered section
P: Parasitic or stop notch
nose, and a second one under the tail (Fig. 17.91). The positions of each have to be
chosen to make good the deficiencies of the other. If there are no suitable surfaces
for vertical monopoles, horizontal slots or notches can be used. Fig. 17.92 shows
VHF and UHF antennas 805
12
a notch antenna for mounting in the trailing edge of a tail fin (Burberry ). Parasitic
elements above and below the driven notch reduce current flow on the fin and give
azimuth radiation patterns similar to those of a vertical dipole and director.
Telemetry and command: Telemetry systems on aircraft are mainly used in the
flight test phase and command antennas are restricted to pilotless aircraft which
may be targets for weapon trials or RPV (remotely piloted vehicles). A variety of
frequencies have been used, but in general the coverage requirements have to take
account of large aircraft manoeuvres and will be similar to those for IFF. Most of
the systems are narrow-band, and can therefore use small antennas of monopole
types, the transmission-line antenna being frequently used in the 400MHz band.
ILS localiser and glidepath: The localiser and glidepath equipments in the air-
craft compare the depth of modulation of two signals radiated from the ground
station. The equisignal path is defined as that on which equal modulation depths
are received, and hence the accuracy with which this path can be followed depends
on the ability of the aircraft antennas to reject false information. The signal
received at a point in space, neglecting the effects of the aircraft, may arrive from
several directions as well as directly from the ground station. The additional paths
occur from reflections from objects on the ground or from other aircraft. Since the
radiation from the ground station has a difference in depth of modulation (DDM)
which varies with angle, the additional signals will each have a different DDM and
random phase. The combination of these unwanted signals with the wanted signal
produces the so-called 'bends' in the localiser and glidepath beams. For a given
ground station the severity of the bends as seen by the aircraft will be a function of
the aircraft antenna. The polarisation of the unwanted signals may also be random,
so that the response of the aircraft antenna to the unwanted vertically polarised
component may also be significant.
A further source of error arises from the signals received at the antenna by
re-radiation from parts of the aircraft. The vector sum of the aircraft contributions
is a function of aircraft attitude, and as this alters, the apparent position of the air-
craft also moves with reference to the equisignal line. Moving parts of the aircraft —
propellers, helicopter rotors and radar scanners — may also impose further modu-
lations on the received signal. Whilst many instances have been recorded of the
806 VHF and UHF antennas
serious effect of rotors and propellers on both localiser and glidepath reception,
radar scanners have been observed only to affect glidepath signals because their
dimensions are small in terms of wavelength at localiser frequencies. Studies of
phase-centre shift of ILS localiser antennas have been carried out extensively by
the Royal Aircraft Establishment (RAE) in England (Jones41).
Gain requirments for localiser and glidepath antennas are modest, the usually
acceptable figures being — 19dB for localiser and — 15dB for glidepath, in the
forward direction, referred to a half-wave dipole. Coverage requirements vary
with the flight procedures adopted to acquire the beams, but in general these
minima should apply up to 45° in azimuth from the line of flight. Where localiser
antennas are also used for VOR reception, omnidirectional coverage will be
required and a rather higher gain because of the longer ranges involved.
The basic requirements are therefore:
(i) Adequate gain
(ii) Minimum phase centre shift with normal aircraft manoeuvre
(iii) Good polarisation discrimination
(iv) Directivity
ramshorn
nose dipole
pair of suppressed
or external aerials
pair of
suppressed aerials
Localiser antennas: Fig. 17.93 shows typical positions for localiser antennas.
Four main classes can be distinguished:
(i) Ramshorn
(ii) Pair of monopoles or flush elements
(iii) Wing tip or tail-plane tip antenna
(iv) Nose dipole
VHFand UHF antennas 807
Pair of elements: To improve azimuth gain the vee dipole was mounted through
the tail fin. This is unsatisfactory on thick fins and a pair of monopoles, raked 45°
and fed in anti-phase, is preferred. Half-loops and flush-mounted elements with
backing cavities are also used (Thomas and Johnson87). For high-performance air-
craft a notch-fed plate, similar to Fig. 17.28 and enclosed in a dielectric fairing, has
proved satisfactory.
This type of antenna on a fin should be mounted at least 0.5X below the top of
the fin metalwork to minimise edge currents. Because the system is electrically
balanced, vertically polarised radiation from the fin leading and trailing edges is
a minimum in the forward and rearward directions. If, however, the balance is
upset, there will be a significant increase in such radiation received by the antenna.
Fin-mounted antennas receive signals from the forward direction with strong
reflected components from the fuselage and wings so aircraft attitude causes
phase-centre shifts, which will be unacceptable for automatic landing systems
although normally not a problem for general ILS use.
radome'
balance unit
Elements may be mounted in each side of the aircraft nose if their separation is
not more than about 0.9 m. This gives good horizontal plane gain and has proved
satisfactory for ILS and VOR in spite of the inevitable grating lobes Fig. 17.94.
Space for flush antenna cavities will normally be limited and, since the dimensions
808 VHF and UHF antennas
need to be about 0.75 m long, 0.25 m high and 0.2 m deep for a system VSWR not
worse than 7:1 referred to 5012, there will only be limited application for this type
of antenna. Any of the external elements used for fin mounting can be used if
aero dynamic ally acceptable, types totally enclosed in dielectric covers being most
suitable and least prone to damage in ground handling.
Wing-tip antennas: Structurally the wing tip will often be suitable for a flush
monopole or capacitance-loaded notch. A single element is used. The leading edge
of the wing will have a strong current flow and this may lead to significant pick-up
of re-radiated signals from the fuselage, making the antenna response very
susceptible to aircraft attitude. Phase-centre shift is very marked, and this antenna
position is not suitable for automatic landing systems.
nose
port starboard
2 A in scanner
glidepath aerial
Nose dipole: A dipole mounted in the nose radome is naturally shielded from
the remainder of the aircraft, and hence the dependence on aircraft attitude is a
minimum. Such an antenna has proved very satisfactory for automatic landing and
VHF and UHF antennas 809
for aircraft used for inspection and calibration of ILS ground stations. The antenna
is physically and electrically symmetrical, has adequate gain and good directivity.
It should be noted that correct disposition of lightning strips on the radome is
important to minimise pick-up of vertically polarised radiation and failures of such
strips may degrade the antenna performance.
It can be seen from the examples given that for the highest-quality service the
antennas must be mounted on the aircraft centre line and must be electrically and
physically symmetrical, with minimal reception of signals re-radiated by the aircraft.
Glidepath antennas: The criteria for localiser antennas apply equally to glide-
path antennas. Only limited forward and downward coverage is required and the
most suitable position is towards the nose of the aircraft. The movement of weather
radar scanners can seriously modulate the received signal as well as reduce the
signal level, as Fig. 17.95 demonstrates.
Since the bandwidth required is small, short antennas are possible. Variations on
the transmission-line antenna are commonly used (Fig. 17.96), the shape and load-
ing being adjusted to give maximum forward signal. This type can be mounted on
the aircraft nose, in a radome, or on the metal behind the radome.
As a flush-mounted antenna the half-wave slot with backing cavity is very suit-
able. Direct or probe-feeding may be used, the latter permitting two isolated feeds
810 VHF and UHF antennas
in the same cavity for multiple receiver installations. Typical cavity dimensions are
0.44 A long, 0.16A wide, 0.13 A deep, the slot with dielectric cover being only
0.40A. If the aircraft nose is occupied by a scanner, the slot can be cut in the metal
skin immediately aft of the radome, as shown in Fig. 17.97. With the latter
arrangement there may be significant reception from the rear, and propeller
modulation may be a problem.
20* minimum
Ideally the glidepath antenna should be on the pitch axis of the aircraft so that
pitch changes produce minimum change of antenna position. For this reason, tail-
fin mounting should not be used where the aircraft has a large pitch change in the
approach and landing phase.
On small and medium propeller-driven aircraft it can be difficult to find a
position free from modulation for the conventional antennas. Some success has
been achieved with a short Yagi array of director, driver and reflector in monopole
form mounted on the side of the aircraft nose (Fig. 17.98).
Marker antennas: Marker position-indication occurs when the signal at the air-
borne receiver exceeds a specified and preset level. For a given aircraft height the
received signal is dependent on the transmitter power and the product of the
ground and airborne antenna patterns. It can be shown that, because transmitter
powers and antenna patterns vary considerably, it is impracticable to satisfy the
specification in detail.
The ICAO standard antenna is a half-wave dipole spaced 0.038X below the air-
craft fuselage and with axis fore-and-aft. The position of such an antenna has a
significant effect on its radiation pattern as Fig. 17.99 indicates.
Because of its drag the standard antenna is rarely used, the alternatives being an
external or flush mounted transmission-line antenna (Fig. 17.19). In this appli-
cation, high capacitance-loading is necessary to obtain good radiation from the
VHF and UHF antennas 811
port starboard
director
Homing: The majority of airborne homing systems utilise ground stations used
for normal communications and are therefore vertically polarised. The principles
would apply equally to horizontally polarised systems, but these will not be
described here.
rtose
Fig. 17.99 Pitch-plane patterns of ILS marker antenna under aircraft fuselage
Length of fuselage: 2.5\
0.2\ from nose
1.25\ from nose
The aim of a homing system is to enable the aircraft to be flown towards the
transmitting station. If the apparent direction of the station given by the homing
system differs from the aircraft fore-and-aft axis the system is said to have a squint.
Homing to the station may still be possible but the flight path will then be curved.
The system must be free from ambiguities; if the ground station is to the left of the
aircraft, the homing system must show this. In summary the basic requirements are:
(i) Minimum squint
(ii) Good sensitivity on-course
(iii) Good angular discrimination
(iv) Freedom from ambiguities
VHF and UHF antennas 813
aircraft
centre line
identical cables
phasing link
Phase homing: The essential components of a phase system are shown in Fig.
17.100. They consist of a pair of identical antennas symmetrically disposed about
the aircraft fore-and-aft centre line and connected by a phasing link. The take-off
point can be switched so that one antenna leads or lags on the other. Assuming for
the moment that the two antennas would individually receive identical voltage
amplitudes in the absence of the other, and that the mutual impedance between
them does not alter their relative phase, then the phase between them is
(2TTJ/X) sin 0, where 6 is the angle between the axis of the aircraft and the direction
of the beacon.
If the phase shift in the feed cable is 0, then the total shift is
\jj = (2nd/X) sin (6 + 0).
If Et and E2 are the voltages at each antenna and we write E2 = nE\ the resultant
voltage ER = Ety/l + n2 4- 2n cos \p.
As the take-off point is switched from A to B, \p will alter from (2TT^/X) sin (d + 0)
to (2nd'/X) sin (6 — 0). At any angle 0 the ratio ERA/ERB is the DF ratio. lfEx =E2i
ER - 2EX cos \p/2 and the DF ratio becomes (cos i//l/2)/(cos
814 VHF and UHF antennas
The basic equation for ER permits calculation of DF ratios for different values
of 0 using measured radiation patterns for Ex and^ 2 -
Ideally the DF ratio should increase monotonically as 6 increases from 0° to 90°.
Maximum ratio at 6 = 90° is achieved by making the value of \p at this angle equal
to it, a condition which has, in theory, to be met at the highest operating frequency
to avoid ambiguity. In practice the system can be arranged to limit at DF ratios
exceeding some set value and the restriction can be eased, provided a good site can
be obtained. Difficulties arise when the radiation patterns of the two antennas
differ to the extent that, say, the port antenna has a minimum on the port side
which is not matched by a similar minimum by the starboard antenna at the same
angle on the port side, As a general guide the spacing between antennas should be
A/4 at the highest frequency and <j> should be 90°. As the frequency is decreased
the phase shift decreases and the DF ratio changes more slowly, to the point that
the equisignal path is not clearly defined. It is normally reckoned that the DF ratio
at 10° off course should lie between 1.4 and 2.0 for satisfactory homing. Higher
ratios make it difficult to hold a steady course and are generally accompanied by
reduced signal level on the equisignal line — in the extreme case the on-course range
may be very much less than the off-course, making it necessary for the aircraft to
be yawed to re-acquire signal.
To achieve good homing accuracy, identity of impedance and of radiation pat-
tern are both important. It is not sufficient to have identity of VSWR; very close
control of impedance is essential and low VSWR will make this control easier. It
follows, therefore, that high homing accuracy over wide frequency bands is difficult.
Normal communications antennas, even of broad-band design, may not have suf-
ficiently identical impedance to give good homing. There is some evidence that
resistance loading may be advantageous. Since the left-right symmetry of radiation
pattern of an antenna mounted on an aircraft fuselage decreases with the displace-
ment of the antenna from the centre-line, it becomes increasingly difficult to obtain
good homing at low frequencies with this type of antenna system.
Antennas used for this type of homing will normally be A/4 monopoles or
possibly transmission-line antennas. The latter are useful only for single-frequency
applications where height is limited, as they are difficult to optimise. For operation
in the 30-76 MHz band, dipoles in front of the fuselage are the only form of
antenna suitable because of ground-clearance limitations and because the spacing of
approximately 1 m would give very asymmetric patterns for monopoles on the
average fuselage. It is usual to load the dipoles with resistance since their length can
only be about A/2 at the top end of the frequency band, thus giving a low radiation
resistance at 30 MHz.
One additional facility from a phased antenna system is that it can also provide a
substantially omnidirectional pattern for communications. The two antennas have
to be fed in phase and some matching may be necessary (broadband antenna
elements are essential for broad frequency bands such as UHF), but satisfactory
systems have been achieved for the whole UHF band on a number of aircraft on
which the number of antenna sites was limited.
VHF and UHF antennas 815
The effect of aircraft attitude changes, particularly roll, has to be considered. In
general, monopole systems on a finite ground plane are more susceptible since they
have a higher proportion of cross-polarisation.
line of
flight
\
\
port j starboard
port antenna
aircraft
skin
director
sleeve dipole
(i) Short Yagi: Two-element or three-element Yagis will produce the desired
cardioid pattern. They will normally be mounted near wing-tips to minimise pattern
modification by the aircraft. Dipole or monopole systems can be used, the latter
being mounted on the wing surface and being rather more sensitive to position. A
frequency range of 1.36:1 centred on 200 MHz has been achieved with three mono-
pole elements. In this instance the director is tuned for the high end of the band
and the reflector for the low end.
(ii) Dipole-director pair on side of fuselage: Fig. 17.102 shows typical patterns for
a dipole-director pair mounted parallel to the aircraft centre line and spaced about
0.2X from the fuselage side. The necessary skewed patterns depend on diffraction
around the nose and the distance from the nose has to be determined by measure-
ment on each installation. 10% bandwidths have been achieved, the limitation being
the sharpening of the DF ratio and the loss of on-course signal at high frequencies
and the converse at low frequencies.
Fig. 17.103 (a) Terminated loop antenna. (b) Azimuth pattern of vertical loop as {a)
(iii) Slot in side of fuselage: A horizontal slot in the side of a metal nose can pro-
vide a low-drag solution with good radiation-pattern shape. On a parallel-sided nose
with radome the slot should be immediately behind the forward edge of the metal,
whilst for an all-metal nose the slot central normal should be skewed 45° from the
line of flight. Cary17 gives some examples of the use of this form of antenna. For
most aircraft a lower frequency limit of about 300 MHz will apply since typical
cavity dimensions will be 0.5X long, 0.22X high, 0.1X deep for a 10% bandwidth.
Decrease of cavity dimensions brings severe restrictions in impedance bandwidth.
Slots may be probe or direct fed, off-centre feeding being used in the latter
arrangement to improve matching.
(iv) Terminated loop: A vertical loop such as Fig. 17.103 will produce in the plane
of symmetry a cardioid radiation pattern if the correct terminating impedance is
used. Typical dimensions for a full loop would be 0.03X high and 0.07X wide, but
VHF and UHF antennas 817
the antenna is more often used against a horizontal metal surface in which case the
height is halved. Experiments have shown that the shape of the antenna is not
critical, full loops in which the height and width were transposed yielding excellent
results over a frequency range of 2.5 :1.
These loops may also be used in pairs mounted on either side of a fuselage; in
this form the antenna is bisected vertically. It has been found better to use a double
loop, which might perhaps be better described as a terminated dipole, to reduce
cross-polarisation pick-up.
The half-loop under the fuselage is usually constructed to be fed at each end in
turn, the terminating resistance being switched at the same time. In this way two
mirror-imaged radiation patterns can be obtained. Some criticism of the half loop
has been based on its susceptibility to cross-polarised radiation, and it is to be
expected that this will be true on the small ground planes available in the 30-
76 MHz band where it has been mainly used. Its application to higher bands (it was,
in fact, used between 200 and 600 MHz in the Second World War) with larger loops
should be much less susceptible.
parasite
starboard
tail
Fig. 17.104 Effect of parasitic notch on VHF communications antenna patterns
100 MHz
120 MHz
140MHz
Discontinuities: Control surfaces in wings and tail necessarily imply gaps between
one metal surface and another. Such gaps, which are bridged by low-impedance
818 VHF and UHF antennas
[starboard
If, instead of monopoles, linear or crossed slots are used, the relative signal level
in the axial and radial directions will be altered because of the different element pat-
terns. If slots are physically impracticable, heavily capacitance-loaded transmission-
line antennas (Fig. 17.19) will provide similar patterns and are very simple to
mount.
The roll-plane ripple can be reduced to any desired amount by increasing the
number of elements. Table 17.2, from Pugh, Barker and Thomas,73 gives results for
roll-plane gain and ripple for cylinders of four different diameters with arrays of
circumferential slots.
Although frequencies in the 2 GHz band are being increasingly used for TTC,
there is still considerable usage of 137 and 148 MHz, and these systems will be
described. Telemetry of the 'housekeeping' data may be required from launch, so
the antenna system requires good spherical coverage.
Widespread use is made of a four-element monopole array, loosely described as
a turnstile. It may be a true turnstile as in Fig. 17.107 or the elements may be
physically separated and set round the periphery of one face of the satellite body
(Fig. 17.108). In each case there is a progressive phase shift of 90° between adjacent
822 VHF and UHF antennas
elements. It will be recognised that this arrangement, with one-cycle phase pro-
gression, follows the scheme described in Section 17.4.7.
Much use can be made of numerical methods for determining the radiation pat-
terns. Because the satellites are normally small in terms of wavelength, magnetic-
and electric-field integral equations, MFIE and EFIE, are the most appropriate. Use
may also be made of wire grid modelling as described in Section 17.4.1. A good
summary of the application of these methods is given by Albertsen, Hansen and
Jensen.2
Because the TTC antennas have to operate from launch they have to radiate
through the shroud which protects the satellite in the launch phase. In most cases
this shroud will be of glassfibre and will not, therefore, have significant effect at
VHF. Some launchers, notably uprated Thor-Delta and Ariane, have a metal shroud
with dielectric windows. Clearly radiation through these will depend on the relative
position of the satellite antennas to the apertures. The shroud can be considered as
a cavity resonator in which a complex field will be excited, particularly if the
VHF and UHF antennas 823
antennas lie around the periphery of the satellite and close to the inside of the
shroud. Determination of radiation for a given satellite is best achieved by scale
modelling.
It may be necessary to fold or retract antennas whilst the satellite is within the
shroud. The need for operating in this condition may constrain the amount of
retraction which is acceptable.
-5
dB
-10
30 40 50 60 70 80
MHz
Fig. 17.110 Effect of body on impedance of antenna on UHF (460 MHz) portable radio
equipment
From the various studies a number of general points arise; these should be con-
sidered in designing a radio system to be carried by a man:
(i) The radio equipment has to be considered as part of the antenna system, this
may also be true for the earpiece and microphone leads unless inductive loop coup
ling is used.
(ii) If operational constraints permit, the equipment should be held as far away
from the body as possible.
826 VHF and UHF antennas
(iii) For UHF and the higher VHF bands the advantage to be gained by having the
equipment as high up the body as possible is greater than simple height gain would
predict.
(iv) When the antenna has to be in close proximity to the body, the impedance will
vary rapidly with position. The basic antenna should have as broad a bandwidth as
possible. Short, fat, helical antennas radiating in the side-fire mode have been
shown by several writers to be most effective. Simple tuning means should be
provided unless the loss due to mismatch can be tolerated. Providing adequate
tuning requires a knowledge of the antenna impedance. Measurement requires con-
siderable care to ensure that the measuring cable does not become part of the
antenna, Krupka55 and King and Wong48 show similar choke systems to minimise
this effect, although Kuecken,57 discussing HF measurement, states that chokes
were ineffective, probably because of the very high impedances involved. In any
case the system must be designed 'on the body' and not in isolation.
front
left right
Fig. 17.111 Azimuth patterns of a vertical antenna on the front of a standing man
82MHz 155MHz
245 MHz 800 MHz
Scales are relative only. No relation between different frequencies
Antennas for radio tracking: For ease of installation antennas fitted to mammals
usually consist of a loop fitted round the neck and carrying the transmitter and bat-
teries. These are arranged to lie at the lowest point so that the antenna is fed as a
vertical loop. Evidence from field trials is that on the animal the antenna is
VHF and UHF antennas 827
essentially vertically polarized in the horizontal plane and has a directional response
as a free-space loop. This indicates that coupling to the animal is small. This type of
antenna has been used with transmitters in the 27, 101 and 170 MHz frequency
bands, the higher frequencies normally with the smaller animals.
Radiation patterns: Wherever possible use should be made of scale models, but,
particularly with land-based vehicles, the complexity of the model may make it
cheaper to buy, for example, a used motor car which will also be valuable for
impedance measurements.
Measurements on land-based vehicles can be most rapidly done by rotating the
vehicle on a flat-bed turntable. Even the type of turntable used in car showrooms
will, if equipped with a position synchro, give good service out of doors for years.
In setting up the antenna test site the normal criteria for ground ranges (see
Chapter 8) should be observed. The antenna aperture D is to be taken as the
largest dimension of the vehicle. Some reduction in range below the far field
distance 2D2/X can be permitted at the expense of definition of depths of nulls.
This may be unimportant for most communications antennas.
In the absence of a turntable the technique is to keep the test vehicle fixed
and to take a measurement antenna round it at a fixed distance. Ideally, a con-
tinuous plot of signal level around the cricle is required. This can be done if the
measurement antenna is fixed to a vehicle driven round the circle, means being
provided to convert the vehicular motion to angular position with respect to the
test vehicle. A simpler system, adequate for most purposes in this frequency range,
is to define successive 10° points round the circle and to measure at each in turn.
If in addition significant maxima or minima are noted between the 10° points, a
good representation of the radiation pattern can be obtained. With practice it is
unnecessary to halt at each 10° marker and the vehicle can be driven continuously
round the circle. It should be possible to measure one pattern in 10 minutes, very
much more quickly than trying to turn the test vehicle to successive headings.
The same technique may be used for aircraft provided that the antenna is not
mounted on a lower surface of the aircraft. As an indication of the extent to which
the range can be reduced, Fig. 17.112 compares a 100ft aircraft measured at a
radius of 200 ft centred on the antenna, in the aircraft nose, with the far-field
scale-model pattern.
Flight polar diagrams on aircraft should be used only when scale-model or
ground measurements are impractical. Patterns can be measured either by flying
828 VHF and UHF antennas
an orbital or a daisy-pattern flight path. In each case the object is to present diffe-
rent attitudes of the aircraft to a ground station at which the field strength is
recorded. Fig. 17.113 illustrates the two techniques. The daisy-pattern usually con-
sists of 36 runs at 10° intervals over the pinpoint; this is time-consuming but has
the advantage that the aircraft is flying straight and level over the pinpoint and the
measuring range remains constant. In the orbit the object is to fly as flat a turn as
possible; this may mean a radius of several kilometres and hence the range has to be
increased to minimise the correction for the radius of the aircraft turning circle.
nose
port starboard
tail
The ground station will normally be elevated to give adequate range so that the
graph of signal strength against range is of the form shown in Fig. 17.114. The aim
should be to choose a pinpoint at P so that range errors have the least effect on
signal strength. For a detailed treatment, see Bothe.8
Impedance and VSWR: Vehicles rarely consist of isolated flat sheets of metal,
so measurements on flat ground planes are not representative of antenna perfor-
mance on a vehicle. It will usually be necessary to use either the vehicle or a
mock-up of it for impedance measurement. To what extent need the whole vehicle
be simulated? Consider a monopole mounted at the extremity of an aircraft tail fin
- clearly the antenna and the remainder of the aircraft constitute an asymmetric
VHF and UHF antennas 829
dipole. From Section 17.2, the impedance of such a dipole is approximately the
sum of the impedances of its two components, one of which is the antenna itself
and the other is the remainder of the aircraft. Non-circular sections can be trans-
formed to equivalent circular cross-sections and the resulting model of the aircraft
is of a series of cylinders and cones of varying diameter. It will be seen that, for
most frequencies, the diameter exceeds 0.05X and reference to measured
impedances of such fat dipoles, for example Brown and Woodward,10 shows that,
for arms greater than A/2, the impedance of the dipole is practically invariant. It
therefore follows that, provided the length of the structure exceeds A/2 from the
base of the antenna, the measured impedance will be a good approximation to
that on the whole aircraft.
ground
station
start
ground
station
Gain: Gain for vehicular antennas is usually referred to some agreed standard.
For a motor-car it might be a X/4 monopole at the centre of the roof, a halfwave
dipole or a ground-plane monopole. It may not be important to know absolute gain.
-80
-90
useful range
£ -100 of P
~° -110
I -120
a
•o -130
-150 20 40 60 80
range km
Fig. 17.114 Typical range-power curve for constant antenna heights and frequency
This Section describes antenna systems used either to determine the direction of
an incoming signal or to lay down defined patterns along which a vehicle,
particularly an aircraft, can be steered.
120
-o TOO
j? 80
$ 60
o
.c
a
20
-180 azimuth bearing
0 90 • 180
20
Fig. 17.115 Measured phase shift between pair of antennas showing influence of mounting
surface
Isolated ground plane
Vehicle roof of same size
Elevated H-Adcock array: The next system in terms of simplicity is the elevated
H-Adcock array, Fig. 17.116. It consists of a pair of balanced dipoles coupled
832 VHF and UHF antennas
together to produce a pattern with two nulls, plus a centre element whose output
can be appropriately combined with the pair to produce a cardioid pattern to
resolve the ambiguity. The array is raised above ground as high as possible, the limit
being determined by the need to avoid a null in the elevation plane in the angular
range of interest. This null would arise from the combination of the antenna and its
image in the ground. If h is the height of the antenna system above ground and 0
the angle of elevation, the first null occurs at sin 0 = X/4/z.
D D
signal level , V
20= 1.0\
The small base line of both forms of Adcock means that a good reflection-free
site is required to eliminate bearing errors since all the antennas are equally
affected. Earp and Cooper-Jones25 pioneered the use of the commutated antenna
system in what is now known as CADF (commutated antenna direction-finder) or
DDF (Doppler direction-finder) equipment.
Tracking in two planes: The antennas described above are used essentially for
tracking in one plane. The need for VHF and UHF tracking in two planes originated
with the radio astronomers and has its major application with satellites and ballistic
missiles. Two separate approaches have to be considered:
(i) Orthogonal arrays of omnidirectional elements whose phase is varied continu-
ously to sweep the beam.
(ii) Circular arrays of comparatively narrow-beam antennas combined to give a
conical scan.
The basis of the first approach is the interferometer.
VHF and UHF antennas 835
The interferometer: Consider a pair of simple antennas spaced S wavelengths
apart and fed in phase, Fig. 17.120. Their radiation pattern in the plane through the
pair consists of a number of narrow lobes which are the product of the individual
element patterns and the grating pattern given by two spaced sources. Around the
central lobe the width between the first nulls is 513 S degrees. It will be seen that
the beam width of this array is half that of a continuous array of aperture S. By
judicious choice of the element patterns and 5, typically 5 = 3 to 5 times the aper-
ture of the individual antennas, it will often be possible to obtain adequate angular
accuracy by seeking the peak value of the largest lobe without the need for any
phasing. This implies that the whole assembly can be rotated or that the move-
ment of the target is such that the interferometer can be set so that the target
passes through successive beams.
reference
antenna
By altering the phase between the two elements of a pair, the patterns can be
swept in the plane of the array as described by Ryle.76 Kraus54 describes a simple
form as a phase-switched interferometer. For the arrangement of Fig. 17.120, when
the antenna elements are in phase, the far field pattern is
836 VHF and UHF antennas
where E0((j)) is the normalised pattern of an individual element and \// = 2nS sin 0.
With the phase to one element reversed the pattern becomes
element pattern
interferometer
pattern
SA
Since a null position can be determined with more accuracy than a peak, the
phased-switched system would be used in the following way. The source direction
would be determined coarsely from the position of the greatest peak of the in-phase
system. With the system switched to anti-phase, the source position now lies within
the central null and can be located more accurately.
Fig. 17.121 shows an interferometer in which the two elements are dissimilar.
A special case of this arrangement (Fig. 17.122) in which S = 0 is known as the
Mills Cross (Mills and Little67). It forms the basis of a number of radio-telescope
antennas, some of which are described by Kraus.54
By using other pairs of elements at different spacing, the side-lobe structure of
an interferometer can be suppressed so that the position of a distant source can be
determined with an accuracy only limited by the size of the baseline and the fre-
quency stability of the system.
An alternative use of the system, employed by a number of satellite ground
stations, is to use a close-spaced pair of antennas to give a coarse angular position
and a much wider-spaced pair to define position within the wider beam. Clearly
any convenient form of antenna element can be used. Among those described by
VHF and UHF antennas 837
Kraus are helices and dipoles, with and without parabolic reflectors. For satellite
tracking systems where the required accuracy permits rather simpler systems, slotted
cylinders have been used. These, based on the work of Alford, provide an aperture
of 2X from a single feed. The decision to use a few complex elements or many
simple ones must depend, in part, on the reliability and consistency of phase of the
feed system under all weather conditions.
E/rf)
-H
18 rod 3-6
polarisation
cage
[cut away to show feeds]
2-7
1-8
feed elements in
radomes
-I
0-9
5 0 dia
25
/ / / / / / / / / / / / / / / / / / / / / / / / / / /
Two different aspects of dielectric loading are considered here. The first deals with
attempts to reduce the physical size of antennas by loading and the second with the
use of dielectric materials as protection, where the aim is to minimise the influence
of the dielectric.
dielectric
monopole
The results obtained above refer to coated monopoles on a ground plane. Yagi
and log-periodic antennas could be similarly treated to give a reduction in element
length even if reducing the axial length of the antenna was impractical for most
842 VHF and UHF antennas
frequencies. Work has also been done on reducing the size of cavity slot antennas
by loading with ferrite or dielectric. Table 17.3 due to Lyon and Hiatt60 gives
typical results.
0.03
0.02
0.01
W
1 5 9
N
Fig. 17.125 Height reduction factor N for various dielectric coatings (computed)
A: M r = 8 1 e r = 1
B: M r = 9 er = 9
C: ju r = 1 €r= 81
b/a= 20, see Fig. 17.124
tan 6e = tan SM = 0
This Section is only concerned with covers whose electrical thickness is small,
and not with radome design. Although, the covers are 'thin', they can nevertheless
influence the gain and impedance of the antennas they surround. The effect is
greatest when the dielectric is close to the high-impedance points of the antenna.
Thus a dielectric cover in the plane of a slot antenna will have a significant effect
not only in lowering the working frequency but also in decreasing efficiency if the
material is chosen for its structural rather than its electrical properties.
lOr
dB
height reduction N
tan = 0
To minimise these effects the cover has to be moved to a point of lower voltage
gradient. Fig. 17.127 shows some arrangements which will make the thickness and
position of the cover less critical. It must be remembered that rain, ice or snow will
effectively increase the thickness of the cover, so the spacing has to be chosen to
allow not only for manufacturing and positional tolerances but for the anticipated
build-up of moisture in one form or another, including condensation. If the antenna
design is very critical to mismatch, it may be necessary to provide hot air to keep
the weather loading within acceptable limits.
It should be obvious from the previous Section that the effect of dielectric
covers on electrically short antennas is more significant than on self-resonant ones.
844 VHF and UHF antennas
It is true that the cover may reduce the VSWR; this is probably because of its high
loss resistance. The notch antenna in a flat sheet or sharply curving surface is
another type significantly affected by the proximity of a cover.
structural dielectric
17.7.1 Definition
The plane of polarisation of a radio wave is defined by the direction of the electric
vector during at least one complete cycle. In general, both the magnitude and
direction will vary during the cycle. In a plane perpendicular to the direction of
propagation, the electric vector will describe an ellipse and the wave is said to be
elliptically polarised. Two special cases of elliptical polarisation are of particular
importance:
(a) Linear polarisation in which the electric vector always lies along a fixed line;
ratio of major to minor axis of ellipse is infinite
(b) Circular polarisation in which the electric vector describes a circle; ratio of
major to minor axis is unity
The ratio of the axes (axial ratio) is known as the ellipticity, usually expressed in
decibels, and is the parameter used to define the quality of a nominally circularly-
polarised antenna.
The electric vector can rotate in either direction. The convention adopted by the
IEEE is that, viewed in the direction of propagation, the polarisation is right-hand
for clockwise rotation, left-hand for counter-clockwise (see Chapter 1). Fig. 17.128
shows the orientation of the electric vector during a complete cycle for right-hand
circular (RHC) polarisation and gives a clue to methods of propagating such a wave.
Clearly two orthogonally polarised antennas fed with equal amplitude and having a
phase shift of n/2 will generate the appropriate field. The antennas may lie in the
846 VHF and UHF antennas
same plane orthogonal to the direction of propagation and be fed with a TT/2 shift
or they may be displaced A/4 along the direction of propagation and fed in phase.
The following Section includes many of the current methods of producing circular
polarisation.
Fig. 17.128 Orien tation of electric vector for righ t-hand circular polarisation
I /0°
TT/2
TT
3TT/2
Section 17.2.1). It should be noted that in one axial direction the turnstile is
polarised RHC but in the opposite direction it is LHC. Like most of this class of
antenna the polarisation degrades to linear in the orthogonal plane. Some improve-
ment in angular coverage for a reasonable axial ratio can be obtained by using vee
dipoles instead of straight ones. A whole family of antennas based on the dipole can
be used including crossed Yagis and log periodics. For good axial ratios it is essential
that the relative phase and amplitude of the orthogonal elements should remain
constant over the desired bandwidth, which implies careful control of element
design and tolerances. It is also necessary to maintain the quadrature phase shift
between the orthogonal components.
cavity
wall p p
I///////.// ///7/\
'o-
180°
I
1 ft 1
[Z / / / / /
n
*/ / / / / / / A
0-53A
both elements being fed (Cox and Rupp20). A variation on this (Clavin etal.1*) has
a slot with parasitically coupled inverted L monopoles on either side (Fig. 17.134).
Other arrangements include a dipole skewed slightly from the slot axis to give
appropriate coupling. Physically the neatest arrangement is that due to Sidford80 in
850 VHF and UHF antennas
which the dipole lies in the plane of the slot. As Fig. 17.135 indicates, the dipole is
a folded type coaxially fed and coupled to the slot by one or two capacitive tabs.
The whole slot/dipole assembly can be printed on one double-sided board. With a
backing cavity no more than A/12 deep an impedance bandwidth of 10% (to 2:1
VSWR) is possible. This antenna, which has been named the Sidpole after its
inventor, has a good axial ratio over a very wide E-plane, but the coupling between
the slot and the dipole has to be adjusted experimentally and is critical.
80 A/2
<n - 4 0 —
£ 0 0-2 0 4 06 0-8 10 1-2 1 4 -16 18 20
5/A
Fig. 17.136 Mutual impedance between two thin parallel \/2 dipoles
2 10
-10
0-2 0 - 4 0 6 0-8 1 0 1-2 1-A 1 6 1-8 2 0
Fig. 17.137 Mutual impedance between two thin coilinear \/2 dipoles
Walkinshaw88 gives values for the impedance of a thin dipole in the presence of
parasitic elements; a study of his figures will show that the parasitic element has
to be within A/2 to produce a significant effect on the impedance of the driven
element. It should be noted that the parasitic dipoles have a zero centre-point
852 VHF and UHF antennas
impedance. In a situation such as Fig. 17.138, where 1 and 2 are dipoles resonant at
different frequencies/! and/ 2 , the impedance of the radio equipment ZS2 a t / t
referred to the centre of dipole 2 may make this dipole a very effective parasite to
dipole 1, and vice versa. It cannot be assumed that because ZS2 is matched to the
transmission line at f2 the same will be true at f\ and in general this is most unlikely.
The effect of dipole 2 on dipole 1 cannot therefore be determined without a
knowledge of the impedance presented to its terminals at the relevant frequency.
Li 1 11 12
Fig. 17.138 Genera/ arrangement for coupling between two antennas operating at different
frequencies
Walkinshaw's paper also gives some indication of the effect of one dipole on the
radiation pattern of another. It is clear that the maximum effect occurs when the
elements are parallel and will be very much reduced if they are collinear or ortho-
gonal. When one antenna is skew to another, although the mutual coupling will be
reduced, there will be some re-radiation from the skew element in cross-polarisation,
the significance of which will depend on the system operation. In homing and
direction finding systems the presence of another antenna may assume much
greater importance. If the antenna is on the axis of symmetry of a two-element
homing system then the effect on each element will be the same. The angular
change of phase between signals received by the two homing antennas may be
disturbed, but the equiphase line will remain on the axis of symmetry. If the third
antenna is offset, however, the responses of the two homing antennas will not be
mirror-imaged about the axis and a 'squint' will develop. In the previous paragraph
the effect of terminal impedance on re-radiation was noted. This indicates that it
may be possible to choose the base impedance of the third antenna to minimise its
effect within the operating frequency range of the homing system.
The coupling between two antennas is defined as the ratio between the received
power (in a matched load) and the radiated power. The coupling between two
parallel doublet antennas, spaced d apart is (— 18.46 — 20 log10(d/X))dB in the far
field. The general formula for the coupling between two antennas then becomes
(— 22 — 20 Iog10(<2/A) + Gx + G2)dB where Gx and G2 are the gains, referred to
isotropic, of the two antennas in the direction of one to the other. This must also
VHF and UHF antennas 853
take into account any skew or displacement between them. This formula relates to
free-space coupling and will be modified by the presence of reflecting surfaces pro-
viding transmission paths other than the direct one. A striking example of this is
given by Klimkiewicz50 for collinear antennas on a common metal mast. He shows
that the coupling between the antennas does not increase uniformly with distance,
but oscillates because each antenna illuminates the metal mast. As it is common
practice to mount many antennas for mobile services on a common mast it is
important to minimise their coupling in the manner indicated in this Chapter.
The problem of coupling between adjacent antennas reaches its most severe pro-
portions on aerospace vehicles, and considerable work on coupling analysis and
prediction has been reported in the IEEE EMC Symposium Records. Of the many
papers published on this subject, those of Siegel,81'82 Palas69 and Siakiewicz and
Adams79 are worthy of study.
± resonant at
T
The formula quoted was for the far field for antennas intended for the same
frequency. The extent to which it can be used at close proximity depends on the
type of antennas and their arrangement, but it should give a reasonable approxi-
mation between parallel dipoles as close as one wavelength. When the antennas
are intended for different frequencies the effect of mismatch must be included;
it should be noted here that broad-band antennas may possess significant pass bands
at higher frequencies. It should also be noted that, although impedance characteris-
tics of antennas of the same type tend to be very similar within their design fre-
quency bands, this is not true outside these bands and it is unwise to rely on
measurements on single examples.
854 VHF and UHF antennas
element C
insulator
metal sleeve A
A practical example of this type is shown in Fig. 17.140 with the dielectric sup-
port structure removed for clarity. It was designed by the author for use on high-
performance aircraft and may be mounted on the fuselage or inside a fin cap. It is
formally described as a top-loaded dual band sleeve antenna. Sleeves A and B
constitute a conventional broad-band sleeve monopole operating over the UHF
band 225-400MHz. The inner conductor of the coaxial feeder is connected to the
closed base of sleeve B, and the dimensions of the two sleeves are chosen to
optimise performance in this band, the total height being approximately 30 cm.
printed -circuit
board - ^ - _ _ _ _ / / // glass fibre
~ cover
glidepath
parasitic
notches
glidepath notch
* localiser
notch
NX \\
coaxial feed
The rod within sleeve B forms with the inside of this sleeve a short-circuited
coaxial line having a high impedance at its open upper end over the UHF band
so that the horizontal top rod can be removed without significant effect. In the
VHF band 118-136 MHz this shorted line provides an inductance in series with the
top loading element whose dimensions are optimised for this band. The whole
antenna thus operates as a top-loaded sleeve for the VHF band. An important
point in the design of this class of antenna is that the upper band can be optimised
first and the components for the lower band added without significant disturbance
of the upper band performance.
As an example of the second class, Fig. 17.141 may be considered. This antenna,
for glidepath and localiser reception on aircraft, consists of two separate notch
856 VHF and UHF antennas
17.9 References
77 SCHELKUNOFF, S. A., and FRIIS, H. T.: 'Antennas, theory and practice', John Wiley,
New York, 1952
78 SHNITKIN, H., and LEVY, S.: 'Getting maximum bandwidth with dipole antennas',
Electronics, 31 Aug. 1962, pp. 40-42
79 SIAKIEWICZ, K. R., and ADAMS, A. T.: 'Analysis and prediction of coupling between
collocated antennas', IEEE 14th EMC Symposium Rec, 1972, pp. 315-320
80 SIDFORD, M. J.: 'A radiating element giving circularly-polarised radiation over a large
solid angle', IEE Conf. Pub. 95, 1973, pp. 18-25
81 SIEGEL, M. D.: 'Aircraft antenna-coupled inteference analysis',. IEEE 11th EMC
Symposium Rec, 1969, pp. 85-90
82 SIEGEL, M. D.: 'Near-field antenna coupling on aerospace vehicles', IEEE 12th EMC
Symposium Rec, 1970, pp. 211-216
83 SINCLAIR, G.: 'The patterns of antennas located near cylinders of elliptical cross-section',
Proc. IRE, June 1951, pp. 660-668
84 SMITH, C. E.: 'Log-periodic antenna design handbook'. Smith Electronics Inc., Cleveland,
Ohio, 1966
85 SMITH, M. S.: 'Properties of dielectrically loaded antennas', Proc. IEE, 124, 1977, pp.
837-839
86 STIDHAM, J. R.: 'Experimental study of UHF mobile radio transmission using a directive
antenna', IEEE Trans., VC-15, 1966, pp. 16-24
87 THOMAS, R. L., and JOHNSON, R. B.: 'A flush cavity VOR/Localiser antenna', Micro-
wave J., Dec, 1966, pp. 66-70
88 WALKINSHAW, W.: 'Theoretical treatment of short Yagi aerials', /. IEE, 93, Pt. IIIA,
1946,pp.598-610
89 WANSELOW, R. D., and MILLIGAN, D. W.: 'A compact, low profile, transmission line
antenna - tunable over greater than octave bandwidth', IEEE Trans., AP-14, 1966, pp.
701-707
90 WEBSTER, R. E.: '20-70 Me Monopole antennas on ground-based vehicles', IRE Trans.,
AP-5,1957, pp. 363-368
18.1 Introduction
In most radio systems the output of the transmitter and the input of the receiver
are at considerable distances from the terminals of the antennas with which they
work. Whenever this distance is an appreciable fraction of a wavelength, or greater,
then the satisfactory operation of the system demands that attention be paid to the
design of the connecting link.
In this Chapter an account will be given of the main types of transmission line
so that a sound choice can be made for any desired system. In many cases the sys-
tem will require the use of such devices as balancing units, power splitters, hybrids,
etc. These are grouped under the heading 'line components'. They are discussed in
this Chapter because many of them are built up from lengths of transmission line
and because without them neither transmission line nor antenna could function in
the desired manner.
The chief types of transmission line used by antenna engineers are the coaxial,
the open wire and the hollow waveguide. Waveguides have been discussed fully in
many specialised volumes and are not treated again here. Because of their extensive
use and wide range of application, coaxial lines are considered first and in some
detail; the open wire lines are then discussed briefly. The Section concludes with
data on other forms of transmission line which may be useful to the antenna engin-
eer. For convenience screened twin lines have been included at the end of the
section dealing with coaxial lines.
The typical coaxial line is that shown in Fig. 18.1 where the conductors are a pair
of coaxial circular cylinders, and this type will be discussed first. For a circuit at
power frequencies where the current density could be taken as evenly distributed
across the conductor cross-section with 'go' and 'return' paths it is simple and con-
venient to look upon the line as two current-carrying conductors. It is clear that as
Coaxial Transmission Lines and Components 861
these currents are equal and opposite in magnitude and coaxial in path, their net
magnetic field outside the line is zero. It may be noted en passent that, if the con-
ductors are not coaxial through some defect in manufacture, there will be a result-
ant magnetic field and this constitutes one of the three mechanisms of unwanted
coupling to and from a coaxial line.
The penetration of the wave into the conductor was discussed by Howe1 using this
concept of a rapidly attenuated wave flowing through a transmission line in a
copper medium. A derivation of ZT in terms of the constants of the conductor has
been given in detail by Schelkunoff.2
The two great advantages of the concept of transfer impedance are that it is not
difficult to measure for cables under laboratory test or as installed, and also it leads
in many cases to a simple calculation of the coupling between pairs of circuits.3 In
very many cases this coupling through ZT is the determining mechanism of leakage
out of (and into) coaxial lines, and for this reason it has been taken by the IEC as
the specified parameter in the standardisation of the efficacy of screening of flex-
ible coaxial cables.4
Because the screening of cables is of such importance in the complex and tightly
packed radio systems of the present day, the transfer impedance will be discussed in
more detail with particular attention to its variation with frequency. Referring to
Fig. 18.2, curve (a) shows the variation of ZT for a tubular conductor. It will be
seen that at low frequencies it approximates to the DC resistance per unit length,
which is shown by the broken line. This is clearly what one would expect because in
the case of a direct current the voltage drop along both the inside and outside of
the outer conductor would be the same and this would be related to the current
carried by the resistance per unit length.
DC resistance
/unit length
1.0 10 100
frequency MHz
As the frequency increases ZT decreases, slowly at first and then very rapidly so
that the leakage becomes quite negligible at the upper radio frequencies (the upper
part of the HF band, the VHF band and higher bands).
Coaxial Transmission Lines and Components 863
Because cables with tubular conductors are mechanically rigid, they are of
limited usefulness and in most installations the cables used have conductors formed
by bands of copper wire. Curve (b) shows the variation of ZT with frequency for a
typical braided conductor of the traditional design. It will be noted that with rise in
frequency there is at first a slight fall in ZT followed by a rise which becomes more
rapid until it increases linearly with frequency. From this it is clear that at the
upper frequencies braided cables have markedly greater leakages than those with
tubular conductors.
Recent studies by Homann5 and Tyni6 have shown that by using braids with
rather larger gaps between the braiding wires than had previously been considered
good practice, the braid can be 'optimised' with the result that ZT follows a curve
such as curve (c) of Fig. 18.2. The word 'optimised' which is generally used in order
to describe such braids is not well chosen. The 'optimisation' procedure aims to
balance the couplings through the braid by the mechanisms which give rise to two
voltages in antiphase. A truly optimised braid would have these exactly balanced
and a Z T of zero. However, because some deviations from the ideal construction are
inevitable the truly optimised cable withZy = 0 cannot be manufactured, and what
are known as 'optimised' cables are at best still far from the theoretically possible
perfection. It may be noted that the substitution of a cable with an optimised braid
for one with a traditional well-filled braid may typically reduce ZT by a factor of 3.
If this is done in both the interfered and interfering systems, then an improvement
of the order of 20 dB can be expected. (This figure will of course depend upon the
system frequency and the size of cable used.) Because of this improvement, 'opti-
mised' cables made to conform to IEC requirements forZ T should be used in instal-
lations where leakage must be minimised.
It is possible to reduce the leakage from a cable with a braided conductor by the
application of a second braid. With careful design the presence of this second braid
can reduce ZT by a factor of 30 or more, giving a reduction of radiated power of
30 dB. If the two braids are spaced by an intersheath ZT can be reduced by a fur-
ther factor of 10, resulting in a power decrease of 20 dB. For installations where
particularly low leakage is important a series of 'superscreened cables' has been
developed.7'8 The relative performances of these types of cables are set out in Fig.
18.3.
In order to complete the picture it should be noted that with cables having a
braided outer conductor a third mechanism of coupling is possible. This is by direct
capacitance coupling between the inner conductors by lines of electric field which
slip through the interstices of the braid. This can usually be ignored in single braid
cables, and is never of any consequence when a double braid is present.
The two characteristics of a coaxial line which are usually of prime importance
are the characteristic impedance Zo and the attenuation per unit length a. The
values of Z o have been standardised by IEC as 50 and 75 12 with some supplemen-
tary higher values (e.g. 100 and 12512) for low-capacitance cables. The 7512
standard was chosen because, for an airspaced cable with both conductors of the
same resistivity, it gives the minimum attenuation per unit length within a given
864 Coaxial Transmission Lines and Components
Superscreened cables
MM11/50 50 2.95 0.1 4
MM 12/50 50 7.25 0.03
Notes
1. Average values, very much larger values can occur
2. These would be the acceptance maxima which would apply if these cables were ordered to
meet IEC screening requirements
3. Estimated by author
4. Maximum specified in TRG 71181 (UKAEA)
5. At frequencies below about 4 to 5 GHz there is very little transfer of current from one
braiding wire to another. As a result, equal currents flow in the braid wires in contact with
the dielectric core and in those of the outer layer. For this reason it is usual to take the
effective diameter over the dielectric (DOD) as equal to the core diameter plus that of one
braiding wire. This applies to calculations of characteristic impedance, attenuation, cut-off
frequency etc.
Expressions for Z o and attenuation per unit length are given in Fig. 18.4. TheZ 0
depends only upon the ratio of the conductor diameters and \JJife. It is rare for the
relative magnetic permeability /i to be other than unity, and the only variable of
Coaxial Transmission Lines and Components 865
practical importance is the dielectric constant e. The expression for OL is more com-
plex. To begin with there are two terms: the first representing the resistive loss and
the second the loss in the dielectric between the conductors.
Conductor resistivity: p,
Dielectric constant; e
Relative magnetic permeability; ju
Power factor; p = tan 5
Frequency: f
Zo = 138 /-loglo(D/a0ohm
V e
a = 2.16
The resistive loss is proportional t o / , being controlled by skin effect. The second
expression for a shows the full dependance of the resistive term upon d and D. It
can be shown that, for a given value of D, a is a minimum in the case where px — p2,
if D/d = 3.59. For an airspaced line this gives a value of Z o = 76.7 £1. The standard
of 75 12 is chosen for low-loss air-spaced line in order to accomodate the inevitable
solid dielectric spacers which produce a slight lowering of the characteristic imped-
ance.
There are other criteria by which the optimum design may be judged other than
the diameter of the outer conductor. The cases of minimum dielectric and mini-
mum conductor cross-sections for a given attenuation have been studied.9'10
It is to be noted that both terms of the second expression for a are dependent
upon \Je. This means that by reducing the solid dielectric between the conductors
both the resistive loss and the dielectric loss are reduced. In fact, in the majority of
cases the reduction of the resistive loss is the more important benefit resulting from
866 Coaxial Transmission Lines and Components
the reduction of the effective dielectric constant of the line, by the replacement of a
continuous dielectric by one of the airspaced or semi-airspaced types.
It will be seen that the first term in the expression for ot is proportional to \Jf,
while the second varies with/. For this reason, while the copper losses dominate the
attenuation at the lower frequencies, if the frequency is increased sufficiently the
dielectric loss will become the more important.
100i
10-
o
o
00
10-
01-
1-0 10 100 1000 10,000
MHz
Fig. 18.5 Variation of attenuation with frequency
(a) For small solid polythene cable RG-58B/U
(b) For medium solid polythene cable RG-213/U
(c) For semi-airspaced cable of same DOD as ib)
In Fig. 18.5 are shown plots of attenuation per unit length against frequency for
some typical cables. Curve (a) relates to a small braided cable (UR43, RG-58/U). It
will be seen that at the low frequencies the slope approximates to that of a square-
root law and that it becomes steeper as the frequency increases and the linearly de-
pendent dielectric loss becomes more significant. Curve (b) is for a larger braided
cable (UR67, RG-213/U); its attenuation is lower because of its greater conductor
area, and the steepness is greater because the reduction of the resistive loss makes
that due to the dielectric relatively more important. Curve (c) shows the variation
for a semi-airspaced line of the same diameter as the solid dielectric line of curve
(b). The attenuation is markedly lower because:
(i) The continuous tube outer conductor has a lower resistance than the braid of
the second cable (UR67, RG-213/U).
(ii) The change from solid polythene to air-spaced dielectric reduces the effective e,
reducing both resistive and dielectric losses.
Coaxial Transmission Lines and Components 867
(iii) The above change reduces the power factor, and consequently the dielectric
loss.
It will be noted that the dielectric-loss term is not dependent upon the conductor
sizes. For a given value of tan 5 it is the same for large cables as well as small ones;
in fact, it is also independent of the cross-sectional geometry and applies to any
TEM system.
The cables listed in this Figure all have solid polythene cores; they range in size
from the smallest to largest. It is useful to compare UR 61 and UR 57. These are
both the same diameter over the dielectric but the second cable being of Z o = 75 £2,
has a smaller inner conductor than the 5012 one; this results in a higher cut-off fre-
quency. The values of fc quoted have been calculated allowing for the thickness of
868 Coaxial Transmission Lines and Components
the braid wires, and therefore are not quite inversely proportional to the diameter
over the dielectric.
F = \Ed\oge- (18.2)
For air-spaced lines, and also for semi-air-spaced lines with bare conductors, the
maximum safe voltage gradient ZT, is usually taken as 10kV(peak)/cm.
For cables with solid dielectric the voltage limitation is usually set by the break-
down of the air gap between the solid dielectric and the inner conductor, although
in some cables of low Z o the breakdown between the dielectric and a wire braided
outer conductor may set the limit. In general, the safe limit for E defined by the
airgaps is very much less than the intrinsic dielectric strength of the solid dielec-
tric.11
The thermal breakdown can take place when the temperature of the inner con-
ductor exceeds the safe limit. For polythene this is usually taken as 85°C and for
PTFE (Teflon or Fluon) it is taken as 250°C. The higher working temperature of
PTFE permits the manufacture of cables of considerably higher power-handling
capacities than their polythene equivalents. This is evident from Fig. 18.7.
The temperature rise of the inner conductor at a given, power level depends upon
the thermal resistance of the layers of the cable external to it. This has been dis-
cussed by Mildner12 whose methods are generally accepted for the calculation of
power ratings. It should be noted that the power ratings quoted in specifications
and cablemakers' lists may be based upon different ambient temperatures, and this
must be allowed for when comparing power-rating data from various sources. It is
also important to allow for any difference between the operating ambient tempera-
ture in a system under design and that chosen in the specification or list. The table
in Fig. 18.8 shows the correction factors which convert the power ratings calculated
at 55°C, 40°C and 30°C to various other ambients for cables with polythene insula-
tion.
3000-
1000-
300-
100-
Fig. 18.7 Comparison of the power ratings of cables having cores of the same size but differ-
ing in material
PTFE core
Polythene core
UR102 and UR67 have DOD = 7.25 mm
UR108 and UR43 have DOD = 2.95 mm
Fig. 18.8 Correcting factors for power ratings allowing for change in ambient temperature
(polythene insulation)
This is markedly lower than the characteristic impedance in the balanced mode.
Thus for Duradio no. 68, while the balanced Z o is 100 £2, that in the unbalanced
mode is only 19.6. The use of the twin cable in the unbalanced mode can give a
convenient low Z o line for matching purposes; however, because the characteristic
impedance is very sensitive to manufacturing tolerances and because the accept-
ance tests on the cables do not include a measurement of the unbalanced Zo it is
Coaxial Transmission Lines and Components 871
wise where a long production run is planned to form the low Z o line by paralleling a
number of coaxial cables, connecting their inner conductors at each end. Fig. 18.10
shows the characteristic impedances which can be obtained by connecting two or
three cables of 50 or 75 O in parallel.
Paralleled cables, O 75 + 75 75 + 50 50 + 50 75 + 75 + 50 75 + 50 + 50 50 + 50 + 50
ResultantZ 0 ,n 37.5 30.0 25.0 21.4 18.8 16.7
It is not necessary to give here the details of the construction and performance
of individual cables because these are readily available. There is an international
specification for radio-frequency cables — IEC96 issued in several parts.4 There are
also a number of UK specifications which are aligned with the IEC specification,
but having additional cables of types for which international specifications have not
yet been agreed. Thus the UK cables of the Uniradio and Duradio range are covered
by BS2316 and the US cables of the RG series are defined in MIL-C-17-D. In
addition most cablemakers issue their own lists which usually relate their products
where possible to the relevant items in the IEC, British Standard or US MIL docu-
ments.
The most familiar of the open-wire lines are those used for telephones, but they are
also used for the feeding of HF antennas. Other systems of similar open construc-
tion are used in many antennas which are designed for installation on metal structures.
Open-wire lines have the great advantage that they are cheap compared with co-
axial cables large enough to carry the same power, and also they are easy for
experienced linesmen to install. However, their balance can be spoilt by the
presence of nearby metalwork and when this happens, they are liable to radiate or
to pick up unwanted signals. For these reasons they are usually confined to trans-
mitting systems and not used to carry very small signals which need protection
from interference.
Open-wire lines are normally air-spaced and so fx/e= 1; this term is therefore
omitted from the formulae given in this Section. From the expression for the
characteristic impedance given in Fig. 18.11 and the curve for Z o plotted in Fig.
18.12, it is seen that, when the wires are well spaced, that is 2y/d is greater than
about 50, Z o varies only slowly with the spacing. For this reason a characteristic im-
pedance of 600 H is much used for transmitter feeders because at this spacing the
inevitable small movement of the wires in the wind will not have a. significant effect
upon Z o and upon the impedance matching of the system. The rapid fall in Z o with
decrease in spacing for close wires (e.g. for Z o = 200 £2) is evident from Fig. 18.12.
872 Coaxial Transmission Lines and Components
The exact expression for Z o given in Fig. 18.11 involves the function
cosh"1(2j/(i).- This can be conveniently evaluated using the identity cosh"1* =
log^Oc + \lx2 — 1). For large values of x the approximation cosh"1* = loge2x can
be used; if x > 3.7 the approximation is within 1% and within \% for x > 5. The
values of cosh"1 for values of x up to 5.0 are tabulated in Fig. 18.13.
or 120 locu ^ + / -f -
)
or approximately 276 log 10 is accurate to 1 % if — > 3.7
which
Yd')' ' d
but the error increases rapidly as the spacing decreases further.
d — conductor diameter
2y — axial spacing;
It was found that Peek's law applies and that the voltage gradient at the conduc-
tor for the inception of corona is greater for fine than for thick wires, being pro-
portional to [1 + (0.301/r)], where r is the wire radius in centimetres.
In many cases it is necessary to fit corona rings to the spreader insulators, but
this results in an increased capacitive loading which increases the electrical length of
•8 4 0 0 -
40 50 60
spacing/diameter ratio }2y/d
the feeder. The same source14 discusses this and quotes measured values for insula-
tor capacitance.
It has been observed that some outbreaks of corona are intermittent while
others are continuous. This is because, if the onset of corona makes the impedance
mismatch at the transmitter worse, then less power is delivered and the corona is
extinguished; in other circumstances the mismatch may be reduced and the corona
become firmly established.
The lines of electric and magnetic field between two conductors of circular
cross-section are shown in Fig. 18.11. The importance of the lines of force drawn
874 Coaxial Transmission Lines and Components
there is that they can lead to further information about associated forms of trans-
mission line. For example the line AA', which is an axis of symmetry and an equi-
potential surface, could be occupied by a perfectly conducting sheet without
disturbing the fields. Because the fields either side of the sheet are mirror images
they represent the same voltage difference; hence the impedance between either
wire and the earth plane must be
Z o = 60 cosh"1 — (18.4)
d
This, then, is the characteristic impedance of a wire of diameter d parallel to a
plane conducting sheet at a distance y from the plane.
We have thus derived an expression for another line from a knowledge of the
field pattern of a well known line. This process of introducing an imaginary per-
fectly conducting surface to follow the shape of an equipotential surface is of
general application. Thus a conducting cylindrical sheet could be placed over one of
the equipotential circles surrounding the upper conductor of Fig. 18.11 so as to
Coaxial Transmission Lines and Components 875
form a pair of lines of dissimilar diameters, as shown in Fig. 18.14. It is clear that
the Z o of this pair would be the sum of the characteristic impedances of each of
the lines formed by the conductors and an imaginary conducting plane at AA'. The
ZQ of each line is calculable using the above equation. This is simple if the position
of AA' is known; however, given only the diameters of the two conductors, D and
d, and their axial separation, 2y9 it is necessary to locate the position of the equi-
potential plane. In order to do this use is made of the properties of the orthogonal
circles which define the lines of electric field and equipotential surfaces.
For the equipotential circles it is known that the product of the two intercepts
with the normal to A A' is equal to the square of the unit distance of the grid. As
this unit distance is the same for both conductors, referring to Fig. 18.14, one can
equate these products:
(yi+^OCVi-rO = (y2 + r2)(y2-r2) (18.5)
The other known dimension s is introduced through the equation
s = yx+y2 (18.6)
Solving for yx and y2
2s
(18.7)
2s
With the position of AA' known, the characteristic impedances Z0l and Z^ of
876 Coaxial Transmission Lines and Components
the cylinders with AA' can be calculated. Their sum Z 01 4- Z&, is the characteristic
impedance of the line formed by the two conductors:
The ZQ of the region between the circular conductors is equal to that between
the circle d and AA' less that between the circle D and AA'. Hence the Z o between
the circles is given by
(18.10)
The expression is exact even for the closest spacings of the circles; however, if
the eccentricity is small, the following expression quoted by Schelkunoff2 will be
sufficiently accurate for most purposes;
Q = ZQ11 — (18.11)
Coaxial Transmission Lines and Components 877
where Z o = characteristic impedance of the line if concentric, e = eccentricity
(interaxial-separation/radius-outer-conductor), and JC = D/d.
Schelkunoff also gives an expression for the attenuation of the eccentric line;
/ 2e2 e2x2 \
1 8 1 2
+ + ( >
where a = attenuation of the corresponding concentric line.
No indication is given of the limits of eccentricity for which the approximation
is reliable except that it is recommended for 'nearly coaxial' lines.
The 2-wire open lines have two sources of weakness: first because of their high
characteristic impedance they are more susceptable to stray couplings and also their
working voltages are higher than is the case for lines of lower impedance; secondly,
because the energy of the guided wave is spread at significant densities over a con-
siderable area about the line axis the chances of interference are relatively high.
Various forms of open-wire lines using three or more wires have been developed
in order to give a reduced Z o without the need for critical wire spacing and to give
some increase in power-handling capacity. Multiwire lines may be either sym-
metrical and therefore balanced, or they may be unbalanced, with one set of con-
ductors partially screening the other, thereby reducing interference by reducing the
field at points not in the immediate vicinity of the line.
Expressions for the Z o of various types of open-wire line are given in Figs. 18.16
and 18.17. These all assume that the wire spacings are much larger than the wire
diameters. The validity of these expressions at close spacings is discussed in a later
Section.
In Fig. 18.16a is shown a group of three wires with the two outer ones connec-
ted together. These two can be considered as a very degenerate form of coaxial
screen.
In Fig. 18.16Z? is shown the field around a set of four wires positioned at the
corners of a square and driven so that each of a diagonal pair are at the same volt-
age, the signal voltage being applied between the opposed pairs, it is therefore a
balanced line. It will be seen that there are two planes of symmetry AA' and BB'.
These are equipotential surfaces because they are normal to the lines of electric
field. Hence the system could be halved by a conducting plane put through AA'.
This is shown in Fig. 18.16c and it represents a balanced 2-wire line above a con-
ducting plane such as the earth. It is clear that Z o is halved.
In the same way the system of Fig. 18.16& could be quartered by conducting
planes lying in AA' and BB'. One of these quarters is shown in Fig. 18.16J; it is a
single conductor placed in a conducting right-angled corner. Because the four quar-
ters of Fig. 18.16Z? are connected in series-parallel, the Z o of each quarter is the
same as that of all four taken together; and for this reason the expression for Z o
given in Fig. 18.16<i is the same as that in Fig. 18.16&.
The formulae quoted for Z o in Fig. 18.16& and c are among those given by
McLean and Bolt,15 who also present data on other open-wire lines and the effects
of nearby conducting planes.
878 Coaxial Transmission Lines and Components
o oo
S -I
A--H H--A
T o o
h
2.
Z o = 138log K
\fSi = s2 =s Zo =
Zo =
(d) Wire in corner
The wire of diameter d is spaced at distances sl/2 and s2l2 from the walls of a
conducting right-angled corner
Zo = 138log l 0
Coaxial Transmission Lines and Components 879
+ +
+ +
c
s ; 0
4- +
-0-
-fr:
4- +1
= 276
880 Coaxial Transmission Lines and Components
The characteristic impedances of some other open-wire lines are given in Fig.
18.17. The solution of the case of two wires in parallel above ground shown in Fig.
18.17& is derived from Fig. 18.17a by placing a conducting plane in the equipoten-
tial surface midway between the pairs of wires. The Zo of each pair of paralleled
wires taken against the plane is half that of the system of Fig. 18.17a. The line with
two wires in parallel is not often used by design although it may arise inadvertently
when the feed to a balanced line is out of balance.
Figs. 18.17c and 18.17J show examples of unbalanced lines in which one set of
conductors is partially screened by the other; they may be regarded as rudimentary
coaxial transmission lines.
Zo = 60 cosh" (18.13)
* - » » * •f
It will be noted that the first term is that of the corresponding line with circular
conductors and the second term represents the effect of distorting the line from a
circular cross-section.
Omar and Miller19 extend this analysis on a semi-empirical basis to line cross-
sections formed by rectangles whose major and minor axes define confocal ellipses.
This is extended to other rectangular cross-sections giving an accuracy in general of
better than 2%. A circular inner conductor within a square-section outer conductor
is shown in Fig. 18.19. The expression given by Schelkunoff20 for Z o is that for a
Coaxial Transmission Lines and Components 881
normal coaxial line with the addition of the factor 1.078. This factor in effect
replaces the square by a circle of diameter 1.078 times that of the inscribed circle.
The effect of the factor is to add 4.51 £1 to the value of Z o calculated for a circle of
diameter D. The reason why there is a limitation on the value of D/d is that, for a
very large inner conductor, the uniform distribution of charge on this conductor,
which would occur if the outer were circular, would be seriously distorted by the
proximity of the mid points of the sides of the square in which regions the charges
would concentrate. An expression for the Z o of a circle within a square applicable
to all diameters of inner conductor has been derived by Parzen.21
by AZQ compared with that of a line with a single wire inner, where
Z o = 1381og10*i (18.14)
Values of AZ 0 calculated for an air-filled system are given in the bottom row.
Applying this to a balanced twin line it is seen that the use of 7-strand wire
would add 3.7612 for each conductor, i.e. 7.52 O for the whole line. It is clear that
this would also be the increase for the 4-wire lines of Figs. 18.16 and 18.17.
It is sometimes convenient to have the inner conductor of other than circular
cross-section, in which case it is necessary to determine the radius of the equivalent
circle. Three shapes and their equivalents are shown in Fig. 18.21.
T
a
h—b—H
, = 0.5a De= 1.18a 0 e —0.40 ( a + 6 )
Fig. 18.21 Diameters of circles equivalent to simple shapes
The values quoted for the equivalent circles are derived on the assumption that
the spacing of the conductors is large compared with the dimensions a and b. In
such circumstances the current distribution will not change as the conductor is rota-
ted about its axis and will be that for the conductor in free space.
The derivation of the equivalent circle for the thin ribbon or strip is published
by Anderson23 as a demonstration of the conformal method of computing charac-
teristic impedance.
It will be seen that the equivalent diameter for the field outside the square
quoted above, 1.18c, differs from the value when the field is inside the square,
1.078, shown in Fig. 18.19a. The value for the angle is a good approximation when
a and b are nearly equal but if b became very small, the equivalent diameter would
increase and approach the value of 0.5Ckz applicable to the ribbon. These consider-
ations do at least define the limits between which the equivalent diameters for
intermediate bja ratios must lie.
Coaxial Transmission Lines and Components 883
18.4.2 Conductors in enclosures
In some cases the field around a conductor will be limited in extent by parallel con-
ducting planes. The simple case of a conductor parallel to a plane is considered in
the discussion of Fig. 18.11 and those of series and paralleled conductors in Figs.
18.16c and lS.llb. A conductor in a square is shown in Fig. 18.19a and one in an
angle in Fig. \S.\6d.
Other arrangements are shown together with expressions for their characteristic
impedances in Fig. 18.22. The expression for the conductor sandwiched between
two parallel sheets is shown in Fig. 18.22a. The expression for Z o is quoted by
Jasik.16 It will be noted that the second term within the squared bracket which cor-
rects the first term (which by itself is often quoted as an approximate formula) is
surprisingly small, being no more than a correction of 0.58% for d/D = \ .
Fig. 18.22& shows an arrangement of balanced twin conductors between two
planes, in which Z o is twice that for the previous arrangement. The reason for this
can be seen from Fig. 18.22Z?, where the broken line is an equipotential surface. If a
conducting sheet were placed in this position the field pattern would not be
disturbed but the transmission line would behave as two lines of the type shown in
Fig. 18.22a, and would have twice the impedance of the line of Fig. 18.22a.
The unbalanced line of Fig. 18.22c and the balanced line of Fig. \%22d are like-
wise linked in Z o . The dotted line of Fig. \%22d marks an equipotential plane mid-
way between the conductors. If it were occupied by a conducting sheet, one would
have a line of the type of Fig. 18.22c on each side of A A'. Hence the Z o of Fig.
18.22^ must be twice that of Fig. 18.22c.
The expression for Z o given in Fig. 18.22c is applicable if the diameter of the
central wire is small. A detailed analysis by Chisholm24 has led to expressions which
are accurate for large-diameter wires and which can allow for displacements of the
wire from the central line. Corresponding comments apply to the expression given
in Fig. l&.22d which is derived from the preceding case.
2D D -
1
2h
I
I
I
-4-
Fig. 18.22 Characteristic impedances of some wires within enclosures
4
F "I
4D
0-0367 (tf/P)4 1
-
-
97fi
276 J
(c)Z0 = ^tanh^ cT < D, ft
ITU
ITU U
U
for an expression accurate for large values of d see Chisholm24
= 2276log
{d)Z00 = 7 6 l o g 1 10 0^§tanh^
§tanh^ d<Dfh
for an expression accurate forr large values double the expression given by
Chisholm24 for the case of (c) above
Considering Fig. 18.23; case a, when d/b < 1, can be regarded as a slot in an infi-
nite sheet; this is sometimes used as a 'slot line' feeder, but more frequently as the
basis of resonant slot antennas. Fig. 18.23& shows the two extreme cases of d/b < 1
and d/bP- 1. The intermediate cases have been considered by Morton.26 In Fig.
18.24 are shown three curves relating Zo with d/b. The left-hand one applies when
Coaxial Transmission Lines and Components 885
257
276 log 10 4(2 +dlb)
log lo 4(1+26/c0
d/b<\ d/b>\
d
L 377 d/b 276log10 4 -
\ b
157T 2 (Bd
138logl0|--
0.693 + ^
d/b- 1
64.3 (Sd
138logl0 - ~
log io \n b
d/b> 1
64.3
138 Iog10( 2.16-
3.06 \ b
\-b/d
-d—H
d/b> 1
64.3
138logl0 2~
b
\-b/d
_ b— d/b>\
d
/2 d
4d
188- 138logl0
_ I b
2 300-
<u
"8
200-
100-
3 4
the spacing to width ratio d/b
Returning to Fig. 18.23 it will be noted that, when d/b > 1 for cases (c) and (d),
Zo is the same; this illustrates the point made previously that, when a non-circular
conductor is well away from adjacent conductors, it can be rotated about its axis
without affecting Z o .
In case (e) it will be noticed that, for d/b > 1, the multiplying factor 2.16 is the
factor 1.078 given for a circular conductor in a square in Fig. 18.19a divided by
that relating the width of a strip to the equivalent diameter from Fig. 18.21.
Example (g) of Fig. 18.23 has been added to the six cases given by Magus and
Coaxial Transmission Lines and Components 887
25
Oberhettinger by imagining a plane conductor, to be placed between the strips of
case (b) in order to derive an expression for Z o for the unbalanced line formed by a
strip parallel to a large conducting plane.
u-
13-
E 1-2-
1-1-
1-0-
1 2 3 4 5 6
d/b
All the transmission lines considered so far have had a uniform cross-section.
Conical lines have a varying cross-section but maintain a constant characteristic
impedance and therefore waves can travel along them without impedance mis-
match. The lines are again considered to be air-spaced and the term y/jjje omitted
from all the formulae for characteristic impedance.
Two typical examples of conical lines are illustrated in Fig. 18.26; the left-hand
line is the conical counterpart of the coaxial line and the right-hand one is the
counterpart of the balanced twin open-wire line. It is evident that the wavefront in
these lines will not be plane but a section of a sphere, and, further, will expand
keeping its shape constant.
Lengths of coaxial conical line have been used for many years as a means of
joining one coaxial line to another of lesser diameter without the impedance
888 Coaxial Transmission Lines and Components
discontinuity which would accompany any sudden steps in the diameters of the
conductors. With the aid of stereographic projection it is possible to derive the
characteristic impedance of a given conical line from that of a corresponding uni-
form line and vice versa. This technique was described by my onetime colleague Dr
P. G. Rohde in an unpublished internal note; because it is not well known in the
antenna literature the use of stereographic projection in the theory of conical lines
is given here in some detail, together with examples of its use in the determination
of the characteristic impedance of related uniform lines.
9j angle
between axes
8 half angle
of cones
In Fig. 18.27 is shown a pair of cones with apices at 0, the centre of a sphere.
These form a conical transmission line whose intercepts on the sphere are A and B.
Consider a tangent plane EFGH touching the sphere at Q. Then the projection of A
and B from P, the other end of the diameter through Q, giving A' and B' is a stereo-
graphic projection.
We now introduce the theorem that if a pair of conductors is erected on A' and
B' normal to the tangent plane, then the characteristic impedance of this pair is the
same as that of a conical pair. This follows from the discussion of spherical waves
on cones by Schelkunoff.27
In order to illustrate the method Fig. 18.28 and 18.29 have been prepared to
show the projections used in order to derive the characteristic impedances of the
two conical lines shown in Fig. 18.26. Consider first Fig. 18.28, which shows a
coaxial conical line consisting of two conductors, the inner one of half angle dx and
the outer of half angle 62. The conducting parts are shown shaded. The circle is of
unit diameter. The circles are projected from P onto the tangent plane at Q. The
lines of projection are shown dotted. The circles on the sphere project to give two
Coaxial Transmission Lines and Components 889
circles on the plane both centered on Q, one of radius QA and the other of radius
QB. We consider a coaxial line erected upon the circles; its Z o will be equal to
1381ogJ^H (18.16)
In order to express this in terms of the angles 6t and 62, which define the conical
line, it should be noted that the angle subtended at P by any arc of the circle is
half that subtended at O. Consequently
QA = tan B2 /2 and QB = t a n ^ / 2
Hence
[tanflx/2
1381og10 (18.17)
{tm62/2
By similar reasoning it is seen that the twin line erected upon the circles AA' and
BB' of Fig. 18.29 has a characteristic impedance
and because
y = tan 0i/2
and
d - tan — tan
890 Coaxial Transmission Lines and Components
the Z o of the twin conical line is
Taking the usual simplification for 9X much greater than 62, the expression
becomes
Z o = 2761og10 (18.20)
It may be noted that the reason why the dotted cones of projection from P can
Coaxial Transmission Lines and Components 891
intersect both the sphere and the tangent plane in circles, is that the cone of projec-
tion is elliptical in cross-section and the circular cap and tangent plane both make
the same angle with the axis of the projection cone from P. It is an important
property of stereographic projection that any circle on the sphere is projected as a
circle on the tangent plane.
In Fig. 18.30 are shown two conical lines which are of special use in antenna
theory: the cone normal to a plane conductor which is introduced in the study of
the monopole excited against a plane, and the two identical cones with a common
axis (which mathematicians would describe as only one cone) on which dipole
theory is based.25
The expression for the Z o of the former is derived from the equation of Fig.
18.27# by putting dl = 7T/2; the corresponding expression for the two cones is
obtained by doubling this. It is to be noticed that the Z o of a cone inclined to a
conducting plane will be half of that shown in Fig. 18.27& for the twin conical line.
If the cones of Fig. IS30b were rolled flat one would have the conical strip of
Fig. 18.31a. A half of this, driven against a conducting plane, is shown in elevation
in Fig. 18.3 \b. We now derive an expression for the Z o of the conical strip starting
with the arrangement shown in Fig. 18.31c. This shows the intersections of the con-
ducting surfaces of the half strip system with the sphere. The strip intersects the
sphere along an arc subtending 20 at the centre. This is projected from P as a
straight line BB' in the tangent plane. A narrow strip of uniform width is shown
erected upon BB'. The plane conductor intersects the sphere around the equator.
This cut projects as the circle AA' in the tangent plane. We have thus the arrange-
ment of a strip within a circular conductor as shown in Fig. 18.3 Id.
892 Coaxial Transmission Lines and Components
The ZQ of a strip within a circle has been previously noted (Fig. 18.24/) to be
equal to that of a circular conductor of diameter half the width of the strip within
the same circle, or
Id
Zo = 1381og10— (18.21)
b
A' d
Zo = 1381og10(2cot<9/2) (18.23)
For a conical strip on a plane, we double this in order to obtain
ZQ = 276 log 10(2 cot0/2) (18.24)
for the conical strip of Fig. 18.3 Iff.
Coaxial Transmission Lines and Components 893
Now the relationship between the width of the strip and the diameter of the
equivalent circular conductor will only apply if the width is small compared with
the diameter of the outer conductor; that is in the present case if 6 is small.
An expression for other values of 6 is derived in the following way; consider the
strip line of Fig. 18.32a having half angle 0, and the complementary one having half
angle f— 6 shown in Fig. IS32b. These are not only complementary geometrically;
they support complementary fields, and therefore from Babinet's principle their
characteristic impedances are related in air by the formula
ZoZo = T/2/4 where n = 1207T - 377.
We thus have accurate expressions for 6 small and near to 7r/2; an exact value can
be obtained for the case 0 = n/4 because then the two lines are identical, and there-
fore
Zo = Zo = ??/2 = 188.5 O (18.26)
It will be found that at 6 = n/4 either of the expressions gives a value of Zo with-
in \% of the exact one. We thus have a solution for the characteristic impedance of
a conical strip in two expressions which between them are accurate to better than
|% throughout the whole range 6 = 0 to n/2,
(a) Z o = 2761og10(2cota/2) for0<Ji.e. Z O >188.5H (18.27)
128.6
(b) Zo = ; /nn nxx fotd>i i.e. Z o < 188.5O (18.28)
Fig. 18.33 Characteristic impedances of a double conical sheet and a conical sheet normal to
a plane
(a) Double conical line of thin sheet
Zo = 276loglo2cot0/2 if 0 < 45° or Zo > 188.5 SI
orZ
128 6
o = (/n o
90n —
- Me\ ' f 0 > 4 5 ° or Zo < 188.5 a
Iog 10 l2cot — J
(b) Conical sheet normal to plane
Z o = 138log l o 2cot0/2 if 0 < 45° or Z o > 94.25 O
64 3
or Z = if e > 4 5 o r Z
° 7 ' or> fl\ ° « < 94-25 "
The above equations are approximate; each has its greatest error within the ranges
of 0 recommended above when 0 = 45° when the error is less than j%.
It is to be noted that with this line there is a complementary split cylinder line
of half angle (TT/2 — d) which is complementary in the sense of Babinet's principle
and which supports fields which are complementary. These bear this relationship
Coaxial Transmission Lines and Components 895
because the cross-sections of the pair can be projected to give a complementary pair
of conductors within the equitorial plane.
A number of examples are now given which illustrate the ways in which the
characteristic impedance of a new line can be derived from a known one by stereo-
graphic projection. Some of these lines are uniform and parallel and some are
conical. Thus from the expression for the Z o of the double conical sheet shown in
Fig. 18.33 we use the projection of Fig. 18.35 in order to derive theZ 0 of the uni-
form line formed by a strip between two semi-infinite planes, and we use a different
projection in Fig. 18.36 in order to obtain that of a strip parallel to a single semi-
infinite plane. Then going back to the expression for the strip between two semi-
infinite planes derived in Fig. 18.35, the projection of Fig. 18.38 gives the Z o of
the conical line formed by a conical sheet normal to a plane; in the upper part of
Fig. 18.39 this leads to the Z o of a cylinder parallel to a thin sheet, and this result is
then used with the projection of the lower part of the figure in order to give the Z o
of a cone on the edge of a semi-infinite plane sheet.
896 Coaxial Transmission Lines and Components
By this process of projecting a known line to produce another and then re-
projecting back in a different way a limitless number of conical and uniform lines
could be produced, together with expressions for their characteristic impedances.
The transformations described above will now be considered in more detail.
may be written
257
forZ f t < 188.5 ft
cone on edge of
sheet
Fig. 18.39 Cylinder parallel to a sheet edge and a cone on the edge of a sheet
(a) For cylinder parallel to thin sheet
+1
Z o = 138log l0 2
128.6
orZ o = if Zo < 94.25 n
Iog 10 4fl
{b) For cone and sheet
for0 < 45°, Z0> 94.25
64.3
orZn = for0 > 45°, Z O < 94.25
g10
Iog 90 -e
10((2cot
Coaxial Transmission Lines and Components 899
Also, by expanding cot ((90 — 0)/2) in terms of cot 0/2
Z
o= ( 1 2 8 'L_^\\ (18-29>
logio I
may be written
7
; /
logio 2'y/s/d-I
As is noted in Fig. 18.35, the first equation applies if Z o > 188.5 and the second
if Z o < 188.5.
It is clear that if the conical strip of Fig. 18.35 were rotated clockwise so that
the point M reached the point P, then the right-hand semi-infinite plane would
move to infinity and disappear leaving the strip near a single semi-infinite plane.
This arrangement is shown in Fig. 18.36. Because of the equality of the angles
marked 0, tan0 =D/s = d/D = y/d/s; hence cot0/2 = y/s/d 4 y/l 4 s/d and the
equation
Z o = 2761og lo (2cot0/2)
can be rewritten in terms of s and d:
logio 12 c o t - —
may be written
128.6
Zo = J-J= r==y for Z o '< 188.5 « (18.32)
logi02
The two halves of Fig. 18.39 show that the transmission line formed by an angu-
lar strip normal to a plane is equivalent to that of a cone of equal half angle on the
edge of a semi-infinite sheet. This is done by the two projections shown. The upper
one equates the angular strip and plane to a cylinder parallel to the edge of a strip,
while the lower part shows that the cone on an edge is likewise equivalent to a
cylinder and sheet.
It is now necessary to establish some relationship between the two cylinder-
sheet lines. In fact they are both the same as can be seen:
In the upper Figure, MOQ = [Or/2) + 0]; hence 6?M = [(TT/4) + (0/2)] and
t}PB = [(TT/4) - (0/2)]; also QCP = 0.
Therefore QA = tan [(TT/4) + (0/2)], QB = tan [(ir/4) - 0/2)], OC = cot 0 =
tan[(7r/2)-0]and
,
bf (n , x ... .
tan T 4 - - -f tan - —-
tan
li"lj (1835)
v - -nn\—7rv
tan^ + - J - t a n ^ - -
In the lower Figure, BQ = tan [(JT/4) - (0/2)] and AQ = tan [(JT/4)+ (0/2)] and
(
n \
4~2
(1836)
V = I* e\ (n e^
tan^ + - j - t a n ^ - - -
Hence a'/b' = d'/b" and the two cylinder-sheet lines have the same characteristic
impedance, and therefore for the cone on sheet edge line,
Coaxial Transmission Lines and Components 901
ZQ = 1381og lo (2cot0/2) for0 < 45° or Z O > 94.25 ohms (18.37)
or
64 3
for0 > 45° or Z O < 94.25 ohms (18.38)
/90-fl
2
log10 2 cot
I
which are the same expressions as in Fig. 18.33
It will be seen en passant that it has been shown that the Zo of the line formed
by cylinder and sheet is also equal to that given by the above expressions. In order
to express Z o in terms of the dimensions of the cylinder-sheet line it is convenient
to introduce st and s2 as in Fig. 18.39; then it can be shown that, for the cylinder
and sheet,
or
10R ft
ZQ = 1_ i f Z 0 < 94.25 ohms (18.40)
1
log io 4 -
The upper expression can be rearranged in terms of the diameter d and axial
spacing from the edge, y, so as to give
300
200
o
INI
100
2 U 6 8 10 12 tt 16 18 20
Fig. 18.40 Comparison of the characteristic impedances of a cylinder near an edge and near a
plane
For cylinder and edge
Consider the case of the two-wire balanced line shown in Fig. 18.11 for which
the characteristic impedance is plotted in Fig. 18.12. When the two conductors are
far apart the current is evenly distributed around the periphery of their circular
cross-sections. In these conditions the simple formula
Zo = 2761o g l 0 ^j (18.42)
a
applies and we have the right-hand part of the curve of Fig. 18.12.
As the conductors come closer together so that diameter and spacing become
comparable, the currents concentrate upon the parts of the wires which are close
together. This reduces Z o below that which would apply if the distribution were
uniform. For such close spacings it is essential to use the exact formula
Consideration of the resistive film is of use when assessing the effect of the de-
formation of one line in order to form another. For instance, take the square-
section conductor shown in Fig. 18.42. As noted in Fig. 18.21, this is equivalent
for the purpose of calculating Z o to a circle of diameter 1.18 times the side of the
square. Such a circle is shown. If one considers this in terms of the hypothetical
resistive film, it is seen that in going from the square to the circle one removes the
corner pieces (which are cross-hatched) and replaces them by the segments (which
are shaded). It will be seen that the two arrangements are comparable, particularly
when it is borne in mind that, while the corners are slightly smaller than the seg-
ments, there would be a concentration of current on them, or one could say that
being nearest to the outer conductor the film resistance between them and the
outer conductor was least.
906 Coaxial Transmission Lines and Components
18.7 Layout of cable runs for minimum interference
Earlier in this Chapter the quality of cable screens was discussed and the concept of
the 'transfer impedance' ZT was introduced. Fig. 18.43 shows a cable acting as a
source of interference. If the current flowing in this is /, then the signal voltage on
the outside developed by leakage through the screen is:
VT = ZTI volt/unit length (18.44)
This voltage will induce a current / ' on the outside of the screen of a second cable,
and thereby develop a voltage V*T — Z^l' per unit length on the inside of this cable.
The reduction of ZT and Z'T by the improvement of the cable screens has been dis-
cussed; however, it is just as important to minimise the coupling between the out-
sides of the cables so that / ' shall be as small as possible for a given VT. This is the
next matter to be discussed.
o oT
Fig. 18.44 Two parallel cables spaced from a conducting plane
Mohr gives curves from which L\ L" and M can be read given the diameter of the
cable screen d, the height of the cable axis from the screen h, and D the spacing of
the cable axes. He derives an expression for Vd, the interfering voltage in CD, which
is given here, with some simplification valid at radio frequencies;
Vd = coi- (18.45)
L'L"
where / = current in AB, Z'T and Z? = transfer impedances of AB and CD, Rc and
Rd = terminating impedances of CD.
This expression shows, as one would expect, that the interfering voltage is pro-
portional to the expression
M/L'L"
which is a parameter dependent upon the local geometry, i.e. the installation layout.
908 Coaxial Transmission Lines and Components
In order to display the variation of M/L'L", Fig. 18.45 has been prepared. The
solid curve refers to a pair of Uniradio 43 cables (equivalent to RG-58/U) rest-
ing on a conducting plane; the value of h = 5 mm represents the case where the
cable screen is separated from the plane by the thickness of the PVC jacket. (For
simplicity, the fact that the dielectric constant of PVC is greater than unity is
ignored). At the left-hand end of the curve the pair of cables are close with their
10-
10-
h = 5mm
001
25 50
D.the spacing of the cable axes, mm
jackets in contact. It is seen that in this position M/LfLff* is great, but that as the
cables are separated this parameter falls rapidly, and when their spacing is 50 mm it
has been decreased by a factor of 80. This means that the interfering voltage has
decreased by the same factor and the power coupled decreased by 18 dB. This
demonstrates the importance of cable separation in determining levels of interfer-
ence.
* As V = L" in this case we use the parameter M/L2 later in this Section and in Fig. 18.45.
Coaxial Transmission Lines and Components 909
2
In the same figure the dashed curve shows the variation of M/L with spacing for
a pair of cables 25 mm from the sheet. It is evident that the decrease in M/L2 is less
marked than when the cables are close to the plane. The dotted curve shows that at
a distance of 500 mm away from the plane there is still less reduction in coupling
on moving the cables apart. Comparing the three curves, it is seen that the marked
decrease in M/L2 as the cables are separated only occurs when the cables are very
close to the sheet, a closing in from 500 mm to 25 mm bringing but a small
improvement compared with that obtained in the last part of the approach to the
plane. It is clear that the effects of the dielectric constant of the PVC outer layer
would be to give an even greater benefit from close contact of the cable and the
plane that was calculated when the data for Fig. 18.45 were prepared.
From these considerations one can deduce two rules for the installation of cable
runs near to conducting sheets:
(i) The cables should be held as close as possible to the sheets
(ii) Cables which might interfere with each other should be held as far apart as pos-
sible.
The above discussion does not take into account the effects of resonance when
the sections of the cable between earthing points are an appreciable fraction of a
wavelength long. The effects of resonance are difficult to assess because the
Q-factor of the circuits between the cable screen and the earth sheet are not easy to
define. However, general principles for guidance are clear:
(i) Because the effect of resonance is only to apply a multiplying factor, the coupl-
ing between the circuits will still be reducible by the choice of cables with low
values of ZTi and by good layout which minimises M/L2.
(ii) At the highest frequencies the losses in PVC sheaths will tend to damp the
resonances, and this damping will be increased the closer the cable is held to the
conducting screen because of the concentration of the field within the PVC.
After considering the coupling between two cables near to a conducting plane it
is useful to study that between those installed in a metal cylinder such as the fuse-
lage of an aircraft. In Fig. 18.46^ is shown a cable along the axis of the cylinder.
For any TEM wave guided by the cable within the cylinder the equipotential sur-
faces will be as represented by the broken circles. If the cable is moved to the left as
in Fig. 18.46& then the field remains TEM but the equipotential surfaces will be
squeezed together. They will remain circular and will form a family for each of
which
OAxOB = OP2
It will be noted that the displacement of the inner conductor increases the volt-
age gradient on one side but decreases it on the other. An extreme case is shown in
Fig. 18.46c, where a Uniradio 43 cable is held with its PVC touching the wall
of the cylinder; the centre of this diagram is shown in Fig. 18.46d It is seen that
within the region about A A' the conductors can be taken to be parallel planes and
therefore the voltage gradient may be taken as uniform across AA\ Corresponding
910 Coaxial Transmission Lines and Components
to the various points A n on the line AA' (see Fig. 18.46Z?) are equipotential points
B n . As the potentials of the An points are known because of the uniform gradient,
the potentials of points B n can also be found and the voltage variation across the
diameter of the cylinder be determined.
pvc sheath
Using this approach, the variation of voltage gradient across the diameter of a
fuselage 2.44m wide has been calculated for two cases: where the Uniradio 43
cable touches the wall at one end of the diameter and secondly where the axis of
the cable is 50mm away from the wall. The data are plotted in Fig. 18.47 where
the voltage gradient in volts/metre is shown for an assumed potential difference of
1.0 V between the cable and fuselage. Curve a is for the cable touching the fuselage
while curve b is for one spaced 50 mm away. For the close-spaced cable the greatest
Coaxial Transmission Lines and Components 911
voltage gradient between screen and fuselage is 245 V/rn, much greater than the
maximum of 30 V/m for the other one, but curve a falls away so rapidly that at
only 50 mm away from the wall the gradient is only 2 V/m and it falls below
100 1
0001
0-5 10 15 20 2-5
distance along fuselage diameter, m
0.001 V/m at distances beyond 2.0m along the diameter. At the far end of the
diameter the voltage gradient for cuve a is only l/25th of that for curve b.
Consider a second cable at the other end of the diameter; the potential difference
912 Coaxial Transmission Lines and Components
between it and the wall would be the voltage gradient at the wall multiplied by the
distance of the axis from the wall. Thus, if a pair of cables were put one at each end
of the diameter with sheaths touching the wall and then both were moved to have
their axes 50 mm away from the wall, the voltage gradient at the second cable
would rise by a factor of 25; and because the axial separation from the wall has
increased by 10 (i.e. from 5 mm to 50 mm) the voltage induced on the cable would
increase by a factor of 250. The power picked up in the second cable would thus
increase by (250)2 or 48 dB.*
Once again one sees the importance of keeping cables close to the conducting
screen, this time for the two reasons
(a) In order to reduce the voltage gradient in the main volume of the fuselage
(b) In order to minimise the mutual coupling between cables
In some cases an installation specification sets a maximum limit on the voltage
gradient within the fuselage due to signal leakage. The requirements of the specifi-
cation should be worded so that the gradient in the immediate vicinity of the cable
is not limited; otherwise one might exclude the possibility of mounting a cable
close to the screen as for curve a in favour of an installation such as that for curve
b. This choice would, in the case considered above, reduce the maximum gradient
existing at the cable surface by a factor of about 6 at the expense of increasing the
gradient in about 99.99% of the rest of the volume of the fuselage by a factor of
about 20.
This assumes that the presence of the cable does not distort the field which would exist in its
absence. This is not exactly so and the assumption tends to reduce the ratio of induced voltages
so that the figure of 10 deduced is an underestimate.
Coaxial Transmission Lines and Components 913
it is clear that the equipotential surface which was circular in Fig. ISASa will be
further distorted by a swing nearer towards A in the region of Q.
If now a conductor C is introduced as in Fig. 18.48e, so that B and C are sym-
metrically placed with respect to A and the plane, then both must be at the same
potential and PQ must be pinned back by C as well as by B. Again the increased
pressure between the tubes of induction on the plane side of the conductors results
in peripheral currents on B and C which allow tubes of force to move toward A and
further reduce the mutual inductance between A and B. It is clear that the mutual
inductance (or coupling) between A and B has been decreased just by the presence
of C — and likewise the coupling between A and C is less because of the presence of
B.
(i) In any waveguide mode of resonance, the field strength near to a metallic boun-
dary will be low — yet again one sees the advantage of running cables close to a
metal screen.
(ii) The lower cut-off frequency for any structure such as an aircraft fuselage will be
at its maximum when all metallic items such as the frames of seats and the metal
racks for non-radio items such as safety equipment are as well bonded electrically
to the fuselage as are the units holding the radio and electronic equipment. Each
unit well bonded reduces the volume which is free to resonate.
Because this Chapter is about transmission lines, interference has been discussed
as if it began and ended with cable runs. In fact, a very large number of interference
problems arise through leakage from the units of the radio equipment rather than
from the cables which join them. It is often thought that, if touching the outside of
a cable changes the level of interference in a given installation, then a signal has
leaked out through the cable screen. In fact, all that it means is that a signal is flow-
ing on the outside of the cable — in many cases it gets there through defects in the
screening of the equipment boxes and not through a deficient cable braid.
The screening of plugs and sockets is discussed later.
The pairs of lines can be joined by a coned section such as that shown in Fig.
18.49d. The cones are proportioned so as to join the conductors within the planes
AB and CD; as a result, their apices are slightly displaced longitudinally. For this
B
A
Fig. 18.49 Junctions between coaxial lines of the same Zo but different sizes
(d) For 1% match keep half angle of outer cone < 20°
For 4% match keep half angle of outer cone < 40°
reason the section differs from the conical line shown in Fig. 18.27fl;but if this
slight difference is neglected and if the half angle of the outer cone is kept below
20°, one can equate tan0/2 to 0/2 to an accuracy of 1% and so obtain a match to
within 1%. It will be noted that even if the Z o of the conical line were equal to that
of the adjacent lines, there would still be small discontinuities at the ends of the
cones, where the spherical waves on the conical section meet the plane waves on the
uniform sections. The slight displacement of the apices of the cones gives an excess
916 Coaxial Transmission Lines and Components
capacitance at the ends of the conical section which tends to correct for the change
from plane to spherical wave fronts.
1 I
X
xR B' tx &'
— / 0 A i
i i
;
\ ,
This effect can be reduced by having two sections AB and CD each X/4 long in
series. In this case, referring to the lower part of Fig. 18.50&, the impedance trans-
formation takes place in two stages; firstly following a semicircular arc AB in the
diagram of impedance normalised to ZOi, and then following a second arc BC in the
diagram normalised to Z02- If the frequency is increased so that the lengths AB, BC
are more than X/4 long the first length takes the impedance into the inductive part
* The Cartesian form of the transmission-line circle diagram has been discussed by a number of
authors including Willis Jackson.34
Coaxial Transmission Lines and Components 917
of the diagram to B', then the second half will start in the inductive region at B',
pass some way along its length before the axis is reached and then end up near to C.
Thus the error introduced in AB is compensated in BC. The optimum ratio of the
impedances ZQ and Zo" to be chosen for the X/4 lengths depends upon the desired
frequency bandwidth and the impedance transformation ratio ZQ/ZQ required. It
has been shown by R. A, Smith14 that, when simplifying assumptions are made, the
greatest bandwidth is given when the transformation ratio of each X/4 length is the
same, that is:
Zj = [(Zf0)3Z£]m and 2% = [Zj(Z^) 3 ] 1/4 (18.46)
In general the improvement on this solution which can be expected from an exact
solution is small, and also likely to be obscured by the inevitable junction discon-
tinuities between the lines of differing cross-section. It may be noted that the above
solution has advantages when high-power working is required because the VSWR in
the X/4 sections is at a minimum.
Where a transformation from one Z o to another is required to operate over a
wide frequency it is usual to use a tapered section. This is made as long as con-
venient - say a wavelength at the lower end of the frequency band. With this
arrangement there is a small discontinuity associated with each short length of the
taper. These are all reflected back to the input with progressively lagging phase, and
so are subject to destructive interference - hence the resultant VSWR at the input
is small. This principle of securing broadbanding by spreading a discontinuity pro-
gressively along a length of line is of wide application. In the special case where the
change in Z o per unit length is constant, one has the exponentially tapered line.
This has attracted considerable attention and been discussed first by Wheeler35 and
then by others including Christiansen,36 Niutta37 and Rohrman.38
The construction of a truly exponential line may present difficulties in produc-
tion and not offer any great advantage over more simple forms of taper such as
those discussed by Gurley,39 by Klopfenstein40 and by Gent and Wallis.41
The second special case is when either Z 2 or Z 3 is zero. In this case one end of
AB is at earth potential and the impedance is often described as 'unbalanced'. This
term is unfortunate because its use implies that all impedances are either 'balanced'
or 'unbalanced', while, in fact, these are but two special cases among an infinity of
arrangements of the equivalent circuit of Fig. 18.51.
If one considers TEM-mode transmission lines they mostly fall into two classes:
those with a balanced and those having an unbalanced characteristic impedance.
The most common forms of the balanced lines are the open-wire pairs and the
screened twin cable. The unbalanced lines are usually but not always of coaxial
form. It would be possible to have other lines such as the unbalanced line, which
would result if an open-wire pair were made of one thick wire and one thin one;
however, such a line would not be generally useful.
In order to simplify their connection to the balanced or unbalanced transmis-
sion lines it is usual to design antennas so that their input impedances are either
balanced or unbalanced. This is achieved by ensuring the symmetry of the antenna
where a balanced input is required and by choosing a feed point at, an earth
potential point where an unbalanced input is needed.
These points are illustrated in Fig. 18.52 in which (a) shows the most simple of
the antennas with a balanced input — the centre fed A/2 dipole; (b) shows a unipole
where the feed point is shown at the earthy end in order to present an unbalanced
input; (c) shows a A/2 dipole fed partway between centre and end; its unbalanced
impedance cannot be fed satisfactorily with either a balanced or an unbalanced line.
It sometimes happens that it is necessary to feed an antenna with a balanced
Coaxial Transmission Lines and Components 919
input from an unbalanced line such as a coaxial cable, also, although, less fre-
quently, an unbalanced antenna from a balanced line. Such connections can be
made without ill effect provided balancing units or baluns are used. Without such a
unit serious troubles could arise. Thus, for a coaxial cable connected directly to a
dipole, the antenna would provide a coupling between the inside and outside sur-
faces of the outer conductor. In the transmitting mode the currents induced upon
the outside of the cable would radiate in unwanted directions; also in the receiving
mode any interference signals induced upon the outside of the screen would be
coupled inside and fed to the receiver input.
The most generally used class of balun is that based upon a shorted X/4 of line
placed as a sleeve around a coaxial line. Such an arrangement is shown in (a) and
(b) of Fig. 18.53. The X/4 sleeve is open-circuited at AB and short-circuited at the
other end. For this reason there is a high impedance (which would be infinite were
it not for losses) between B and the outside of the sleeve. This being so, a balanced
impedance can be connected to BC without any coupling to the outside of the
sleeve or cable.
A simplification of the sleeve balun occurs when the sleeve is replaced by an
open stub as in Fig. 18.53c. It is clear that rotation of this about the axis through
C would generate a balun of the sleeve form. While the balanced load could be con-
nected across BC as with the balun of Fig. 18.53a, it is usual to connect A to C by a
short link and then connect the load across AB. This arrangement has the advantage
of greater bandwidth.
There are a large number of variations of the X/4 sleeve balun and information
about many of these is given by Jasik.16 In some designs of antenna, advantage is
taken of a X/4 sleeve which exists as part of the radiators in order to provide balanc-
ing. An instance of this is shown in fig. 18.53d. In this arrangement the coaxial feed
is led along the axis of one arm of the dipole; this arm itself forms a X/4 sleeve
around the coaxial line and so acts as a balun.
It will be noticed that the units so far described have not effected any imped-
ance transformation; for instance, the system of Fig. 18.53& is matched by joining
the points BC by an impedance equal to theZ 0 of the coaxial line. There are other
balancing units which introduce an impedance transformation so as to give match-
ing loads of 2Z0 or 4Z0. The 'rat race' described later in this Chapter can be connec-
ted as a balancing unit with a 2Z0 match. The arrangement shown in Fig. 18.54
gives a match with 4Z0.
The core of the 4:1 transformer is a half wavelength of line AB; because of its
length the voltages at A and B will be in antiphase. Let there be a terminating resist-
ance of 2Z0 at B; this will present an impedance of 2Z0 at A. If a second load of
2Z0 is added at A as shown in Fig. 18.54#, the resultant will be a termination of Z o
presented to the line A A', which will thus be matched.
If the line is now bent into a circle and A and B brought together as in Fig.
18.54&, then, because the voltages across AC and BC are equal and in antiphase, the
pair of loads AB form a balanced load of 4Z0. Hence a balanced transmission line of
characteristic impedance 4Z0 could be connected at AB without disturbing the
920 Coaxial Transmission Lines and Components
impedance match on AA' or exciting stray currents on the outside of the coaxial
part of the system.
Because of the 4:1 step up in impedance this balun is of use where a coaxial line
must be connected to an open-wire line; for instance, a 75 Q coaxial cable can be
connected to a 300 £2 open-wire line.
A B C
T
B C
Where it is desired to have a wideband balun the devices so far mentioned cannot
be used. In these circumstances a tapered balun is sometimes of help. This has the
form shown in Fig. 18.55. Here the outer conductor of a coaxial line is sliced
Coaxial Transmission Lines and Components 921
diagonally over a length AB which is long compared with a half wavelength. At A,
the inner conductor and the pointed end of the outer conductor are connected to
an open-wire line. One can look upon its operation as a process of guiding whereby
the wavefront within the coaxial line at B spreads out as it approaches A so as to fill
the region about the balanced line; or for a wave coming from A to B, the remnant
of the outer conductor guides the wave, wrapping it up into a bundle small enough
to enter the coaxial line. The taper can also act as an impedance transformer, and
there is no special limit on the transformation ratio which this balun can be
designed to give.
2Z 0 2Z 0
B,
characteristic
impedance of
line = Zo
For matching
Resistor AC = 2Z0
Resistor BC = 2Z0
Hence terminating resistance AB = 4Z0
Fig. 18.54 4:1 transformer and tapered baluns
reflection coefficient at the coaxial input of —13 dB for a balun designed to match
from 50 £1 coaxial to 10012 balanced over the band 0.2 to 2.0 GHz. The data given
do not include the performance as a balun; only the input reflection is quoted (i.e.
only its performance as an impedance transformer is given). It should be noted that
at some frequencies in the 10:1 band resonances in the structure could make the
quality of the balancing action very sensitive to the mechanical tolerances achiev-
able in production.
The rat race has several uses. First, it can be used as a power divider in order to
give two signals equal in amplitude and phase. In this case, consider power fed in at
C, this will split into two equal parts; one will travel clockwise towards B and the
other counter-clockwise towards D. The signal reaching B by the clockwise path
will be delayed by TT/2 with respect to the phase at C. The signal passing counter-
clockwise will reach B by a path of 5X/4 having a phase delay of 57r/2, and so be in
phase with the signal arriving by the shorter path. In a like manner both paths will
provide adding contributions at D, their sum like that at B being delayed by TT/2
Coaxial Transmission Lines and Components 923
with respect to the phase at C; hence the two outputs at B and D are in phase. It is
clear that the two paths from C to A will give phase changes differing by TT/2, and
that consequently they will have a zero resultant at this point; for this reason, no
signal will leave the loop except at B and C. Thus the power input at C will divide
equally in amplitude and phase between the outlet ports B and D.
The second use of the rat race is to form the sum and difference of a pair of sig-
nals. If two signals are fed into the loop, one at K and the other at D, it is easy to
show by arguments analogous to those of the previous paragraph, that their sum
emerges from C and their difference from A.
In the design of some variable phase shifters, the two previous uses of the rat
race are pombined. First a signal is fed into C; this splits equally between B and D.
Two short-circuited lines of variable length are connected at B and D, the lines
being ganged so that one is always A/4 longer than the other. These short-circuited
lines reflect the signals back to B and D with two different phase changes, the sig-
nal from the line which is A/4 longer being delayed in phase by n radians. Because
of this difference in phase, the two signals will appear at the outlet A. Variation in
the lengths of the ganged lines will be apparent as phase changes in the signal at A.
If a signal is fed into the loop at A it will divide equally between outputs at B
and D, the two signals being in antiphase. This being so, the voltages at B and D are
symmetrical with respect to earth, and therefore a balanced impedance could be
connected across these points. It will be noted that, for impedance matching, the
balanced impedance should be 2Z0, e a c h half matching one of the two outputs
across which the connection is made. Used in this way, the rat race gives a balancing
unit or balun with an impedance transformation by a factor of two.
Impedance matching within the rat race is achieved when the characteristic im-
pedance of the loop is equal to \/2Z0> where Z o is the characteristic impedance of
the input and output lines at A, B, C and D. If these lines have Z o = 50 £1, that of
the loop should be 71 £2 although the use of a standard 75 £2 line would not result
in a significant error.
The rat race is mostly formed from waveguide or strip line for microwave use;
however, it can be formed from coaxial cable for use in the UHF and VHF bands.
When made from flexible cable it is easy to bring the desired outlets close together
in order to simplify connections.
If the network is fed by five signals at A, B, C, D and E, each will cancel at each
of the other input points but all will add at 0. Hence provided that the phase of the
sources at B and D have been reversed by an extra X/2 of line, the polygon acts as a
five-way signal combiner.
Similar reasoning can be applied to any polygon with an uneven number of sides.
If it is desired to combine an even number of signals, or to split power into an
uneven number of parts, this can be done by using a polygon of one more side than
the number of parts and then omitting a connection from one of the junctions,
preferably the one opposite to 0 .
The matching of the polygon is achieved when the characteristic impedance of
the quarter-wave lines OA and OE is
Coaxial Transmission Lines and Components 925
Z o = —=• (18.47)
n = 3 4 5 7 9 11
Z o = 57.7 50.0 44.7 37.8 33.3 30.212
It will be seen that in each of the above cases the Z o of the polygon will not dif-
fer much from 5012 and, when made of stripline, require strip widths differing little
from that used for 5012.
u
A C
A C
j—WWWW
or socket is sold as being of a given well-designed type, it does not necessarily mean
that it has been made under the costly conditions of controlled manufacture which
ensure products of consistently good quality. Having bought good plugs and
sockets, they must be assembled correctly if the benefits of their quality are to be
enjoyed.
All these considerations become of increased importance at the highest frequen-
cies, and there is much to be said for purchasing those lengths of cable which are
critical in performance from a manufacturer who is prepared to supply as a unit a
length of cable with appropriate plugs and sockets made to a specification such as
MIL-T-81490(AS).
Having obtained good components, the installation should be designed so that
the critical radio-frequency plugs and sockets are mechanically protected and their
environment kept as undemanding as possible.
Considering the requirements of good plugs and sockets we will start with the
electrical contact. If the screening is to be good, it is necessary that there should be
good contact all around the outer conductor; even a perfect contact at one point
on the periphery does not suffice. Thus in Fig. 18.58a is shown a pair of coaxial
lines with outer conductors in contact at A. Continuity tests would show this as a
perfect junction, and yet it would leak signal badly because in effect the two
lengths of coaxial line are joined by a short length of open twin line which radiates
both directly and by way of currents induced on the outside of the coaxial lines.
Coaxial Transmission Lines and Components 927
It is good practice to arrange these contacts as at AA' of Fig. 18.58Z?. Here the
two parts locate at B, the clamping ring forcing both together around the ring of
contact. It is likely that the two parts will be in contact at only three points, but
the narrowness of the gap between B and C will present a low admittance to any
escaping signal.
The inner conductor can give rise to bad contacts, particularly in designs in
which a central pin protrudes. Also where the inner contact element is soldered
onto the inner conductor of the cable there is the danger that this conductor may
become embrittled and crack with subsequent vibration. This is particularly so
when cables having multistrand inner conductors are used.
For satisfactory performance it is imperative that there be good contact between
the cable braid and the housing of the component mounted on it. The principles are
shown in Fig. 18.58c. The housing is counterbored with a locating step at AA and
provided with an internal locking thread. Two metal washers are slipped over the
braid wires which are folded back over the first washer. A locking ring at C holds
the whole assembly, gripping the braid wires between the step in the housing and
the washers, thus ensuring a sound mechanical grip on the braid and a good
electrical contact around the housing.
In some cases a rubber washer is inserted between the washers B and the locking
ring C in order to ensure the sealing of the cable entry. This is bad practice because
there is a chance that, through ageing or thermal cycling, the rubber washer will
yield with the serious consequences that:
(a) The good contact at A is lost with the result that there is signal leakage
(b) The mechanical grip on the braid is relaxed and the cable works loose
(c) As the cable becomes loose mechanical stresses are taken on the inner conductor
and inner contact which may then fail.
The waterproofing of the cable entry is important but it must not be allowed to
put at risk the contact integrity of the cable.
In some systems there is the possibility of interference through the generation of
intermodulation products. These are generated in nonlinear contact resistances, and
the best way of minimising them is to make sure that all contact resistances are very
small. Hence the precautions taken in order to ensure a good screening will help to
control intermodulation.
The generation of intermodulation products in cables and connectors has been
studied by Benson et al4S>46 It has been concluded that the nonlinear effects are
worst in ferrous alloys and those based upon aluminium. The parts in contact
should be held together with sufficient force to break through any oxide film on
the surfaces so that good metal-to-metal contact is established. It has been shown
that in many cases the contacts between dissimilar metals give rise to large inter-
modulation products, particularly when a hard metal is in contact with a softer one,
because in this case the one may flow before the film on the harder contact is
broken. Extreme cleanliness of all interfaces is required.
It is important that the characteristic impedance of the plug and socket when
connected should be close to that of the cable with which it is intended to function.
928 Coaxial Transmission Lines and Components
If it were possible to maintain the diameters of the conductors used in the cables
through the connector and if the same material were used for insulation as in the
cable, all would be well. However, the inner conductor of the plug or socket must
be larger than that of the cable so that it can be bored to accept the cable wire; and
also, in order to maintain a good voltage rating, the mating face is usually made
larger than the cross-section of the cable dielectric. Furthermore, for mechanical
reasons, it is usual to have locating steps on the inside surface of the housing and
upon the inner conductor. In any practical case the necessary changes in diameter
are accomplished by a combination of the arrangements of steps and coned sections
shown in Fig. 18.49.
18.12 References
1 HOWE, G. W. O.: 'The application of transmission formulae to skin-effect problems' /.
IEE, 54,1916, p. 472
2 SCHELKUNOFF, S. A.: 'The electromagnetic theory of coaxial transmission lines and
cylindrical shields', Bell Syst. Tech. I, 13, Oct. 1934, p. 532
3 MOHR, R. J.: 'Coupling between open and shielded wire lines over a ground place,' IEEE
Trans., EMC-9, Sept. 1967, pp. 34-45. Also 'Coupling between lines at high frequen-
cies, IEEE Trans., EMC-9, Dec. 1967, pp. 127-129
4 International Electrotechnical Commission, Geneva, Publ. 96: 'Cables for radiofrequencies'.
96-0: 'Guide to the design of detailed specifications'; 96-1 'General requirements and
measuring methods'; 96-2 'Relevant cable specifications'
5 HOMANN, E.: 'Geschirmte Kabel mit optimalen Geflectschirmen', Nachrichtentech. Zeit.,
21, 1968, pp. 155-161
6 TYNI, M.: 'The transfer impedance of coaxial cables with braided conductors'. EMC Sym-
posium, Wroclaw, Poland, 22-24 Sept. 1976
7 FOWLER, E. P.: 'Some recent work on the screening efficiency of braided and multi-
screened cables'. IEE Conf. Pub. 65, Electrical Interference in Instrumentation, June
1970
8 UK Atomic Energy Standard Specification AESS(TRG)71181: 'Superscreened coaxial
cables for the nuclear power industry; Pt. 1:* 'General requirements and tests'; Pt. 2:
'Cable data sheets*
9 BLACKBAND, W. T.: 'The choice of impedance for coaxial radio frequency cables', Proc.
IEE, 102, Pt, B, Nov. 1955, pp. 804-14
10 DUMMER, G. W. A., and BLACKBAND, W. T.: 'Wires and r.f. cables'. Pitman, London,
1961
11 DAVIS, R., AUSTEN, A. E. W., and JACKSON, W.: 'The voltage characteristics of poly-
thene cables', /. IEE., 94 Pt. Ill, (29), May 1947; and 95 Pt. Ill, (34) Mar. 1948
12 MILDNER, R. C: 'The power rating (thermal) of radio frequency cables', /. IEE, 93 Pt.
Ill, 1946, pp. 296-304
13 BLACKBAND, W. T.: 'Wire-braid screens for r.f. cables', /. IEE, 102, 1955, pp. 363-364
14 SMITH, R. A.: 'Aerials for metre and decimetre wavelengths'. Cambridge Univ. Press,
1949
15 McLEAN, F. C, and BOLT, F. D.: 'The design and use of radio-frequency open wire trans-
mission lines and switchgear for broadcasting systems', /. IEE., 93 Pt. Ill, July 1946,
pp.191-210
16 JASIK, H.: 'Antenna engineering handbook', McGraw-Hill, 1961
17 DUNCAN: 'Characteristic impedance of grounded and ungrounded open wire transmission
lines', Communications, June 1938
Coaxial Transmission Lines and Components 929
18 SHEBES, M. R.: 'Coaxial cable with confocal elliptical cylindrical conductors', Radio-
tecknika, (Moscow) 4, July/Aug. 1949, pp. 36-44
19 OMAR, Y. A. and MILLER, C. F.: 'Characteristic impedance of rectangular coaxial trans-
mission lines' Proc. AIEE, 71, Section T2-75, 1952, pp 81-89
20 SCHELKUNOFF, S. A.: 'Electromagnetic waves', van Nostrand, 1943
21 PARZEN, P.: 'The capacitance/unit length and characteristic impedance of coaxial cables
with one slightly non-circular section', /. Appl. Phys., 18 Aug 1947, pp. 774-779
22 MEYERS, Letter, Elect. Rev. (120), 1942, p. 791
23 ANDERSON, J. C: 'The calculation of characteristic impedance by conformal transfor-
mation', /. Brit. IRE, Jan. 1958, pp. 49-54
24 CHISHOLM, R. M.: 'The characteristic impedance of trough and slab lines', IRE Trans.,
MTT-4, July 1956, pp. 166-172
25 MAGUS and OBERHETTINGER: 'Die Berrechnung des Wellenwiderstandes einer Band-
leitung mit kreisformingem bzw. rechtikigem Aussenleiterquerschnitt', Arch, fur
Elektrotech., 37,1943, pp. 380-390
26 MORTON: 'On the parallel plate condenser and other two dimensional fields specified by
eliptical functions', Phil. Mag., 2, 1926, pp. 827-33
27 SCHELKUNOFF, S. A.: 'Advanced antenna theory', Wiley, 1952
28 THOMPSON, J. J.: 'Recent researches in electricity and magnetism' 1893
29 WILLOUGHBY, E. O.: 'Some applications of field plotting', /. IEE, 93 Pt. Ill, 1946, pp.
275-293
30 WHINNERY, J. R., JAMIESON, H. W. and ROBBINS, T. E.: 'Coaxial line discontinuities',
Proc. IRE, 32, (11), Nov. 1944, p. 695
31 MILES, J. W.: 'Plane discontinuities in coaxial lines', Proc. IRE, 35, Dec. 1947, pp.
1498-1502
32 MEINKE, H. H.: 'Coaxial line elements which produce no disturbing fields at high fre-
quencies', Hochfrequenztech. u. Elektroakust., 61, May 1943, pp. 145-161
33 KRAUS, A.: 'Measured curves of reflection coefficients of compensated inhomogeneities
in coaxial lines and the optimum dimensions deduced from them', Rohde u. Schwarz
Mitt., 8, Dec. 1956, pp. 1-12
34 WILLIS JACKSON: 'High frequency transmission lines', Methuen, London, 1945
35 WHEELER, H. A.: 'Transmission lines with exponential taper', Proc. IRE, 27, 1939,
pp.65-71
36 CHRISTIANSEN, W. N.: 'An exponential transmission line employing straight conductors',
AWA Tech. Rev., 7, 1947, pp. 229-40
37 NIUTTA, A.: 'Impedance matching using a special type of exponential line', Poste e Tele-
comm., 17, Aug. 1949, pp. 417-23
38 ROHRMAN: 'Improvement of the transformation properties of the exponential line by
compensation arrangements', Arch. Elektr. Ubertr., 4, Jan. 1950, pp. 23-31
39 GURLEY, J. G.: 'Impedance matching by non-uniform transmission lines', IRE Trans.,
PGAP-4, Dec. 1952, pp. 107-109
40 KLOPFENSTEIN, R. W.: 'A transmission line taper of improved design', Proc. IRE, 44,
Jan. 1956, pp. 31-35
41 GENT, A. W. and WALLIS, P. J.: 'Impedance matching by tapered transmission lines', /.
IEE, 93 Pt. IIIA, 1946, p. 559-563
42 MARCHAND, N.: Transmission line conversion', Electronics, 17, 1944, pp. 142-145
43 CLOETE, J. H.: 'Exact design of the Marchand balun', Microwave J., May 1980, pp. 99
and 102
44 CLOETE, J. H.: 'Graphs of circuit elements for the Marchand balun', Microwave /., May
1981,pp. 125-128
45 BAYRAK, M. and BENSON, F. A.; 'Intermodulation products from non-linearities in
transmission lines and connectors at microwave frequencies', Proc. IEE, 122, 1975,
pp.361-367
46 AMIN, M. B. and BENSON, F. A.: 'Non-linear effects in coaxial cables at microwave
frequencies', Electron. Lett., 13, 1977, pp. 768-770
Index
Dr Alan Rudge has had more than 15 years experience in antenna research
and design at Universities and Institutions in the U.K. and the U.S.A. He is currently
Managing Director of ERA Technology Ltd in the U.K., where he has established an
international centre for antenna research and design.
Dr Ken Milne is an independent electronic systems design consultant. He was
previously Director of Research for Plessey Radar Ltd, where he specialised in Radar
Systems and Microwave Antenna Design.
Dr David Olver is at Queen Mary College, University of London, where he
lectures and researches in microwaves and antennas. He is Honorary Editor of IEE
Proceedings ‘Microwaves, Optics and Antennas’,
Dr Philip Knight is with the British Broadcasting Corporation (BBC) Research
Department where he has worked on broadcast antennas and propagation problems
associated with broadcasting.