Download as pdf or txt
Download as pdf or txt
You are on page 1of 28

1386

PROCEEDINGS OF THE IEEE, VOL.

61, NO. 10, OCTOBER 1973

Injection Locking of Microwave Solid-State Oscillators


RANEYURI KURORAWA
Invited P a w

Abstr4ct-Injection locking of microwave solid-state oscillators is discrtseed, based on the familiar theorem that the total impedance times the current is equal to the applied voltage. Both qtrasi-static anddynamic analyses of the lucking range, large4gn.d injection, locking stability, and AM and F Y noise are given, and recent expeztmentalwork is reviewedbriefly. No applications of injection lockingarediscassedindetail.

I. INTRODUCTION H E ADVENT of microwave negative resistance devices such as IYPATT'S and Gunndiodes made i t feasible to build economical medium-powermicrowave oscillators with solid-state reliability. Because of this feasibility, theinjection locking of nonlinearoscillators becameonce again an important subject of study. Injection locking stabilizes the oscillation frequency. In addition, it enablesmcillators to perform various functions [l] other than just generating a fixed-frequency signal; namely, to generate [2], [SI, amplify [4]-[12], amplitude limit [13]-[16], and detect [17]-[22] an FM or a PM signal. Injection locking is a well-known phenomenon which is discussed in any standard textbook [23]-[27] onnonlinear oscillators, under one of various headings such as synchronization, phase locking, frequency entrainment, or forced oscillation. The pioneering work was done by Van der Pol in the 1920's when he used appropriate approximations and solved a second-orderdifferential equationpertinentto nonlinear oscillations [28], [Dl. His contribution to nonlinear oscillations isindeed monumental. Many ideas which took clear shape in subsequent developments can traced back to Van be der Pol. However, his theory itself is not necessarily in the most convenient form t o discuss theinjection locking of microwaveoscillators: 1) with the rare exception of tunnel diodes, the current i flowing through the active part of the device in microwave oscillators cannot be represented as an instantaneous function of the applied voltage u whether the negative resistance effect is due to the transferred-electron (Gunn) effect, the IMPATT effect, or an electron beam. However, Van der Pol's theory is heavily dependent on the particular device characteristic expressed by i = -m+@ where a and y are positive constants; 2) the network in which the active device is imbedded is often far more complicated than theone described by a second-orderdifferential equation; 3) for certain purposes, Van der Pol's equation is too complicated toprovide a clear insight intohow the oscillator behaves under various approximate conditions. Van der Pol's theory has been gradually refined, extended, or simplified by many authors including Krylov and Bogoliubov [SO], Adler [31], Huntoon and Weiss [l], Loeb 1321, and

Fig. 1. Schematic diagram of a microwave osci a l tor.

Blaquiere [33], [34] just to mention a few. The same general theory was described by Slater in his celebrated book MicrowaueElectronics [35]. Slater's discussion is particularly suitable for the study of microwave oscillators. The purpose of the present paper is to review recent developments in the injection locking of microwave solid-state oscillators using the language familiar to electrical engineers with a working knowledge of the linear network theory. We will first discuss the quasi-static analysis, extending Slater's discussion. The quasi-static analysiswill reveal almost all the important aspects of both free-running and injection-locked microwave oscillators. Then we will study the dynamic analysis which is convenient for the quantitative discussions of injection-lockedoscillators, includingthe locking transient and noise behavior. Finally, we will review some of the recent experimental work on microwave solid-state oscillators. 11. FREE-RUNNING OSCILLATORS Since some understanding of the free-running behavior of microwave oscillators is a prerequisite for the discussion of their injection locking, we will first consider microwave oscillators without injection signals.

This inoikd QaQn is one of a snicr plunned on bpics of g m t r a l interest-

The Edibr.
Manusaipt received November 11, 1972; reviaed January 31, 1973. The author i w t Bell Laboratories, fiurray Hill, N. J. 07974. s ih

A . Equivalent Circuits A microwave oscillator isillustrated in Fig. 1. I n this schematic diagram, the oscillator is represented by a waveguide cavity containing a negative resistance device. The cavity is equipped with an adjustable short plunger and a slide screw tuner for optimizing the output power. In practice, the cavity may take one of various forms,cylindrical,coaxial,or any other form, and the coupling to theload may be accomplished through a window. The slide-screw tuner may be replaced by an E-H tuner. I n case the output port is in the coaxial form, the slide-screw tuner is most likely replaced by slug tuners. Regardless of the actual construction of the oscillator circuit, however, the entire circuit including the load can be represented by the equivalent circuit shown i n Fig. 2. The reference plane is located at the device terminals. Z(o) is the circuit impedance seen fromthe device and - Z ( A ) is the device impedance. The device impedance is a function of the R F current amplitude A . The frequency dependence will be

KUROKAWA: INJECTION LOCKING OF MICROWAVE OSCILLATORS

1387

- a 4 1 ~+, =u

Fig. 2. Equivalent circuit of a free-running microwave owillator.

neglected for simplicity. This is justifiable since the device impedance is usually a slowly varying function of w compared to thecircuit impedance. The equationof free-running oscillation is given by
[Z(w)

mA

- Z ( A ) ] 1 = 0.

(2.1)

This indicates that the total impedance times the current is equal to zero; namely, there is no applied voltage. Because of its simplicity,our oscillator model representedby (2.1) reveals a number of important aspects of microwave solid-state oscillators, as we shall see in the subsequent sections. I t goes withoutsaying, however, thatour oscillator model is an idealized one. There is an endless list of neglected phenomena and some of the more important ones are bias circuitoscillations, harmonic tuning effects, and subharmonic oscillations. These will be briefly discussed in Section IV-0. The reasons why the reference plane is not chosen at the oscillator output port a-a' in Fig. 1 are as follows. 1) The impedance seen looking into the cavity from the output port is a strong function of w as well as of the R F current amplitude. Besides, it is active. This prohibits the utilization of the vast amountof knowledge accumulated with passive linear networks. On the other hand, active and passive the parts of the circuit are nicely separated in the above equivalent circuit and passivelinearcircuit theorycanbe fully utilized when the Z(o) part is discussed. 2) Once the current at the device terminals is known, the current at the oscillator output port canbe obtained, if necessary, by a routine calculation of an appropriate passive linear transfer function.

F g 3. i.

(a) Impedance locus and device line. (b) Generated p w r oe as a function of device resietance.

I e

Fig. 4. Equivalent circuit o a noisy free-running oscillator. f

e. Power
The generated power is given by Re Z ( A ) = K ( A ) times ( A / f l ) * , orequivalently, by X(A)Af/2. If theoperating point moves from the left to the right along the device line, the generated power will change as indicated in Fig. 3(b). The power initiallyincreasesbecause X ( A ) increases.However, the power decreases beyond a certain point since decreases A' faster than the increase of X ( A ) . Somewhere along the device line, the maximum power is generated. The output power is given by the circuit efficiency times the generated power. In ordinary cases, the circuit efficiency deteriorates as the operating point approachesthe imaginary axis (zero-resistance line). As a result, the maximum output power isexpected to be observed slightly onthe right-hand sideof the maximum generated power point. The difference must, however, be small in well designed oscillator circuits.

B . Impedance Locus, Device Line, and Operaling Point T o appreciate the real meaning of (2.1), the locus of the circuit impedance and that of the negative of the device impedance are drawn on the complex plane by varying o and A , respectively, as shown in Fig. 3. The locus of Z ( w ) will be called the impedance locus. The arrowhead attached to the locusindicates the direction of increasing w and the scale along the locus indicates equally spaced frequencies. locus The of Z ( A ) will be called the device line. When the R F current amplitude flowing throughthe deviceis A , thepoint corresponding to Z ( A ) willbe called the operating point. The arrowhead attached to the device line indicates the direction of increasing A . The scale alongthe device line indicatesequal increments of the RF current amplitude. The intersection of these two loci corresponds to
Z(w) = Z ( A ) .

D . Noise The active device produces noise. Whatever the origin of noise may be, its effect can be represented by a series voltage source as illustrated in Fig. 4. The corresponding equation is given by
[~(w)

(2.2)

- Z ( A ) ] I = e.

(2.3)

If I is finite and the free-running equation (2.1) is satisfied, impedance locus and the device line. The steady-state oscillation frequency and amplitude canbe determined by the scales on the loci at the intersection.

The total impedance times the current is equal to the noise

I Z(w) must be equal toZ ( A ) . Consequently, in the steady state, voltage. Dividing both sides by and transferring Z ( A ) to the the operating point is located the intersection between the right side gives at
Z(0) = Z ( A )

e +i

1388

PROCEEDINGS OF TEE IEEE, OCTOBER

1973

Fig. 5. Relation between the n& impedance locus, and device

line.

vector,

no longer follows closely the value indicated by the intersection, since the stored energy in the cavity and hence t h e R F amplitude cannot change instantaneously. The high-freel quency componentof AM noise is, therefore, expected to drop Z(W)= Z ( A ) -~. j ' (2.5) with the half-power point, being inversely proportional to the Ao circuit Q. F M noise is also affected, in general, by the quick This impedance relation is illustrated in Fig. 5. The small vibration of the device line. However, as far as the low-frearrow with the length IeI/Ao indicates the noise vector quency components are concerned, the simplified model gives eI/Ao)e-* and the tail and head of the arrow determine the a clear and correct picture of oscillator noise. instantaneous amplitude and frequency of oscillation. Since c represents noise, the noise vector isa random vector. Its direc- F. Efects of Distant Reflection tion as well as magnitude changes randomly with time. This With this much understanding of the device line and imw FM and pedance locus,we will discuss a typical behavior of microwave means that both and A change randomly producing AM noise components. oscillators like the one illustrated inFig. 1. First, the probable shape of the impedance locus must be deduced. This can be E . Vibrating Device Line done conveniently on the Smith chart. The Smith chart is T o understand the oscillator noise performance more nothing but the reflection coefficient plane on which constant clearly, a new device line which representsthe right-hand side resistance and constant reactance loci are drawn. The vector of (2.5) will be considered. The new device line is vibrating drawn from the center to a given impedance point on the Smithchartindicatesthecorresponding reflection at the inboththelongitudinalandtransversedirectionsandthe the impedanceismeasured. Inthe intersection with the impedance locus determines the instan- referenceplanewhere taneous frequency w andamplitude A of the oscillation. oscillator circuit, physical structures near the device such as Strictly speaking, the amplitude of the vibration isa function the short-circuitplunger and thewaveguide post to thedevice of position along the device line since as well as el depends for the bias supply will create a large reflection which rotates A0 on the operating point. The larger the noise voltage at the slowly withfrequency.On theotherhand,the slide-screw operating point and the smaller the oscillation amplitude Ao, tuner which is located a t a distance from the device gives a the stronger the vibration the device lineand thebigger the reflectionwhich rotates quickly with frequency. The quick of changes in instantaneous frequency and amplitude, indicatingrotation is a result of the large electrical distance from the the noisier oscillation. This simplifiedpicture of oscillator device to the reflectionpoint. The longer the distance, the noise provides a considerable amount of information. For ex- quicker the phase rotation of the reflection. Because of this ample, if the circuit Q becomes high, Z(w) changes quickly quick rotation, the tip of the resultant vector most probably as with frequency. So the same displacement of the device line draws a loop or loops on the Smith chart illustrated in Fig. now corresponds to the smaller frequency deviation. In other 6. Consequently, i t is most likely that the imqdance locus words, F M noise decreases with increasing circuit Q. On the has loops. I t may be worth mentioning that the distant reflecother hand, AM noise does not change appreciably with intion has less tendency to form loops when the main reflection creasing circuit Q. This is because the amplitude deviation is vector rotates quickly. One way to avoid loops in the imprimarily determined by the scale on the device line which pedance locus is to make the Q of the close-in circuit as high is independentof the circuit Q. During the circuit adjustment, as other conditions permit and the distantreflection as small if the intersectinganglebetween the impedancelocus and as possible. device line becomes small, both FM and AM noises are exSuppose that an adjustable component, for example, the pected to become large. This isbecause, for a given amplitude short-circuit plunger, changes the location of the impedance of the vibration of the device line, the motion of the intersec- locus with a loop downward as shown in Fig.7. The operating tion and hence the corresponding frequency and amplitude as as point PIfirst moves to the right indicated by Pt but soon deviations become larger as the intersecting angle becomes as the upper edge of the loop separates from the device line, small. This simplified noise picture has its limitation. If the the operating point jumps to the left. Fig. 8 will help underdevice line vibrates very quickly, the amplitudeof oscillation stand the detailed behavior near the upper edge of the loop.

For the oscillator to be useful, noise must be small. So e will be assumed to be small. Then, to the first order, I is approximated by the noise-free oscillation amplitude A0 times some phase factor. This means that the second term on the righthand side of (2.4) can be replaced by el /Ao)e-* where Q is the phase between I and e:

Fig. 6. Formation o a loop i the impedance locns. f n

(1

+I

(I

KUROKAWA: INJECTION LOCKING OF MICROWAVE OSCILLATORS

1389

to expect and one does experience with microwave solid-state oscillators having a distant reflection or reflections. In practice, the impedance locus may contain many loops and the loops move in various directions while changing their sizes. Then the result is a more complicated and frustrating oscillator behavior. A similar complicationwill take place whenthe bias supply is adjusted while keeping the circuit unchanged. In this case, the device line moves with respectto the stationary impedance locus resulting in an almost identical effect. One word of caution may be i n order. The loops in the impedance locus cause the frequency and amplitude jumps. However, frequency and amplitude jumps do not necessarily indicate the loop. They are often a manifestation of multiplefrequency oscillation. See Section IV-0.
F g 7. Variation of the operating point as the i. impedance locue changes ita poeition.

G. Onset of Oscillation
Three intersections appeared in the preceding discussion, but only one intersectionat a time represents the actualoscillation. The right-most intersection was not realized a stable as operating point. T o understand these points, we will study how the steady state reached in microwave oscillators. is The RF current amplitude is expected to be small just after the bias supply is turned on. Under this condition, the equation of the free-running oscillation is approximated by where - Z(0) is the small-signal device impedance. Since Z is small but nonzero because of the switch-on disturbances or Z(0) noise and since Z(w) is not equal to for any real w , w must becomplex. The real part will be expressed by w and the imaginary part by -a. Since the time factor corresponding to d - j a is & W , positive a indicates the amplitudeis increasa indicates a decreasing amplitude, ing with time and negative A concrete example in which Z(w) represents a series resonant circuit will help make the point clear. Z(w) is given by

Fig. 8. Explaining a jump of the operating point.

.
1

where & represents the loss in the cavity and RL the load. the In vicinity of the resonance, Z(w) is approximated by Z(W) 1 2j(w - wa)L ( = 2j(w - wa)L

Fig. 9. Typical frequency and power t nn hysteresis u i g of a multiple-tuned oscillator.

+ Ra + RL + 2 a L + & + RL

(2.8) (2.9)

the operating point Pa approaches the upper edge of the loop, the intersecting angle becomes small. This indicates that the oscillationbecomes noisy. Thentheintersectiondisapthe pears as the locusmoves to the positionindicatedby dotted line. No steady-state oscillationis possible near the upperedge of the loop.Consequently, the operating point jumps to the other intersection Pb. If the maximum power point is located at the position indicated by X , the power initiallyincreases as the locusmoves downward, since the operating point approaches the maximum power point. But when the operating point jumps to Pb, the power suddenly decreases and the frequency increases as illustrated in Fig. 9. If the adjustment is reversed, the locus moves upward, the operating point moves to the right, and the frequency well as as the power follows a different path; andwhen the lower edge of the loop separates from the device line, the frequency and power suddenly jump again. In this particular case, both frequency and power decrease. This is a typical behavior one has

As

where
wa

=-. 1

d Z

(2.10)

The impedance locus, the locus of Z(w) with realw, is a straight line parallel to the imaginary axis and a-constant loci are all parallel totheimpedance locuswhile w-constant loci are perpendicular to it. The right-hand side of the impedance locuscorresponds to positive as and the left-handside to negative a s as indicated in Fig. 10. In this example shown in Fig. 10, Z(0) corresponds to a positive a.Consequently, a becomes positive in order to satisfy (2.6) with nonzero small I . This means that the circuit-device interaction forces the curincreases, rent amplitude to increase. As the current amplitude the operating point moves along the device line to the left toward the intersection with the impedance locus. The figure indicates that not only the amplitude but also the frequency will, in general, change as the operating pointmoves since the

1390

PROCEEDINGS OF THE IEEE, OCTOBER

1973

Fig. 10. Explaining the onset of oecihtion

i a single-tuned o d l t r n elan.

the amplitude increaseswithtime and the operating point moves to the left along the device line toward PI. intersecAt tions PIand PI, is still positive, the amplitudeis increasing, a and the operating point does not settle down a t these points. PI, the other hand, represents the stable operating point. on T o realize the operating point Pz,the impedance locus has to be moved upward to initiate the frequency jump and then downward to the originalposition. The operating point PZ thus created is normally stable. If the amplitude increases from the steady-state value, the operating point appears on the a-negativeside of the impedancelocus and the amplitude decreases. If the amplitude decreases from the steady-state value, the operating point appears on the a-positive side of the impedance locus and the amplitude increases. This brings z the operating point back to P . The operating point PCis unstable. If the amplitude decreases, the operating point appears on the a-negative sideof the impedance locus and the amplitude furtherdecreases. If the amplitude increases, the operating point appears on the a-positive of the impedance side locus and the amplitude further increases. Consequently, any small perturbation will result in a runaway phenomenon. PI will never be realized as the steady-state operating point. Later t we shall come back to the stability of P (Section 111-J).
ANALYSIS OF INJECTION LOCKING 111. QUASI-STATIC We arenow in a position to discuss the injectionlocking of microwave oscillators in detail.

Q-eOrrTUIT

LOCI

Fig. 11. Explaining the onset of oecillation and the &abilityof W b l e operating pointa in a multiple-tuned axillator.

value of o changes along the device line. Since a is positive and the amplitude increases as long as the operating point is located on the right-hand side of the impedance locus, the amplitude growth continues until the operating pointreaches the intersection satisfying Z(o0)- Z(A0) = 0, or equivalently

A . SmalCSignal I n j c c t h If a signal with its frequency close to the free-running frequency is injected, the ocillator is synchronized or locked to the injection signal. One wayof injecting a signal is illustrated in Fig. 12. The output port to the load is separated from the injection port by means of a circulator. However, the circuit impedance seen from the device remains the same as before. The locked oscillator equation is given by

~ L ( w ~ - ~ + X ( A O ) = O &+RL-X(AO)=O (2.11)
where 00 and A0 are the frequency and amplitude correspondingtotheintersectionand X ( A ) is the imaginary part of Z ( A ) . Note that the sign of X ( A ) is chosen so that it becomes positive when the device reactance is inductive.

[Z(U,)- Z ( A ) ] I = E

(3.1)

where I$ represents the injection signal voltage seen by the device, and oi is the injection frequency. The discussion will be initially restricted to small-signal injection. In this case, the RF current amplitudewill stay approximately the same as the free-running amplitude Ao. Consequently, (3.1) gives

H . Sbbility of Free-Running Oscillation If the amplitude increasesbeyond the intersection, the circuit-device interaction forces the current to decrease since the left-hand side of the impedancelocus corresponds to negativeas.Consequently,theoperatingpointreturns to the intersection. In other words, the intersection in Fig. 10 represents a stable steady-state operating point. In case the impedancelocus has a loop, the a-constantloci look like those shown in Fig.11. One locus corresponding to a negative a is shownby the dotted line and four loci corresponding to positive as are shown by the solid lines. The small-signal impedance Z(0) corresponds to two values of a, one associated with the dotted line and the other associated with the right-most solid line. Suppose that a small current exists having the a of the dottedline and the proper frequency which is determined by the position of Z(0) along the dotted lines, then the amplitude decreases exponentially with time and eventually disappears. On the other hand, if a small current exists having the a of the solid line and the proper frequency determined by the position of Z(0) on the solid line,

Z(WJ= Z(A)

P I +Ao

e -+

(3.2)

where 4 is the phase difference between I and E . Equation (3.2) indicates that the distance from Z ( A ) to Z(ui) is given by El / A 0 under thelocking condition. There are, in general, two points on thedevice line which satisfy this condition. The correspondingvectorrelationsareindicatedbythesolidand dotted arrows in Fig. 13. The dotted arrow corresponds to an unstable condition as we shall see in Section 111-G. So only the solid arrowwill be considered. This arrowwill be called the injection vector. If one varies wi keeping El constant, the vector moves as illustrated in Fig. 14; Note that constant El approximates the constant injection power when the impedance locus is nearly parallel to the imaginary axis and the circuit efficiency does not change drastically. Since the distance from Z(u) to the device line becomes longer than IEl / A0 below and above 01, the vector relation (3.2) is not satisfied indicating that no locking takes place there. Between ol and a, locking takes place. From the to ~ , changesby 4

KUROKAWA: INJECTION LOCKING OF MICROWAVE OSCILLATORS

1391

1 c

1
w

Fig. 12. A achematic diagram of an injection-locked osciuatur and its eqnivalent circuit.

Fig. 15. Power variation as the injection frequency is swept from ol to am.

Fig. 13.

Relation between the injection vector, impedance locus, and device line.

Fig. 16. A typical power variation when the device line is slanted.

the free-running value, since the operating point comescloser to or further away from the maximum point. I t is interpower esting to note that the first-order variation becomeszero q % when the free-running operating point is locatedat the maxiFig. 14. Variation of the injection vector as the injection frequency is A good amplitudelimitercan be built mumpowerpoint. swept from ol to or. Locking i not possible below WI or above or. s utilizing this property of locked oscillators. If the intersecting angle is not perpendicular, the power variation becomes more approximately 180". The phase difference between the injec- complicated as shown in Fig. 16.As soon as the locking takes tion signal and the circulator output will change more than place at w1, the power decreases in this case, the maximum 180" as we shall discuss in Section IV-F. power point being located on the right-hand side of the interWhen the intersecting angle between device line the and the section. Thereafter, the power changes smoothly and as soon impedance locus is perpendicular, theoutput power will as the oscillator is unlocked at w, the power returns to the change as illustrated in Fig. 15. Depending on whether the free-running value. In practice, the unlocked oscillator signal maximum power point is located on the left- or right-hand and the injection signal interact with each other and createa side of the intersection, the power increases or decreases from beat frequency. This beat signal is usually observed outside

1392

PROCEEDINGS OF THE IEEE, OCTOBER

1973

Fig. 17. A typical power variation o an injection-locked oedllatar. This f picture is taken with a single-tuned I~SPATT owillator. fo: 9450 MHz. PO: mW. Pi:32.3 dB below PO. 60 Horizontal d e : 0.12 MHz/div. Vertical d e : 0.05 dl/div.

I+-""
Fig. 18. Calculation o the l o c k i n g range. f

b l
Fig. 19. Explaining the noiee behavior d a locked o d l a t o r .

the locking range if the outputis detected anddisplayed on an oscilloscope as shown in Fig. 17. Inside the locking range, the C. Noise of Locked Oscillator detected outputclosely follows the curve previously discussed. The noise of locked oscillators can be discussed by means of the vibrating device line in the same way as the free-runB . Locking Range ning case. The head of the injection vector is kept a t w i since The locking range of a single-tuned oscillator canbe easily the injection frequency is kept a t w i and the length of the calculated. Since the distance between the points on the im- vector is kept constant as the device line vibrates. Near the and wo is given by center of the locking range, the longitudinal vibration of the pedancelocuscorresponding to wo+& 2 h L from (2.8), the following relation is satisfied a t the ex- device line gives AM noise. The AM noise is approximately treme of the locking range: the same as in the free-running oscillator since the amplitude scale on the device line is approximately uniform in the range ( 3 . 3 ) of interest prith the small-signal injection. On the other hand, FM noise improves substantially, since the wi point is fixed. where B expresses the inclination of the device line from the FM noise is not completely eliminated because the arrow direction and hence the phase of the current change with the direction perpendicular to the impedance locus as shown in Fig. 18. This relation can be rewritten in terms of the free- transverse vibration of the device line as illustrated in Fig. the running power PO, injection power Pi, and the external Q 19(a). Note that the rate of phase change is frequency. The phasechange is inversely proportional to the length of the of the resonant circuit by substituting injectionvector.Consequently,theresidual FM noise im~ R L A ~ ' Pa = 3 E 1 ' / 4 R ~= Pi proveswithincreasinginjectionpower.When the locking frequency is moved away from the centerof the locking range; u ~ L / R L . L / R L = QQt. w in addition to the longitudinal vibration, the transverse vibration gives AM noise since the operating pointmoves along the The result is device line with the transverse vibration as illustrated in Fig. 19(b) tokeep the length of the injection vector constant. Note ( 3 . 4 ) that the transverse vibration gives primarily F M noise in the free-running oscillator. I t is obvious from Fig. 19(b) that the Notice that R. does not appear in (3.3) and the Q in the lock- AM component due to the transverse vibration increases as ing formula is not the loaded Q but the external Q of the theinjectionfrequencyapproachesthe lockingboundary. resonant circuit. The total locking rangeis given by %. It is Similarly, FM noise increasessince therate of thephase proportional to the square root the injection power and in- changeincreases as the injection frequency approaches the of versely proportional to theexternal Q. When the device line is locking boundary.

perpendicular to the impedance locus, cos B = 1 and the formula reduces to the familiar one without cos 0.

KUROKAWA: INJECTION LOCKING OF MICROWAVE OSCILLATORS

1393

D. FM Amplijer
Since the locking power can be substantially lower than the oscillator power and the oscillating frequency is primarily determined the by frequency of the injection signal, the locked oscillator can be used to amplify an FM signal. In this case, the injection frequencychanges with time and the oscillation frequency closely follows it. For FM amplification, the locking range must be made large compared to the frequency deviation of the injection signal to keep the noise within a tolerable range. This requires a low-Q circuit. The phase distortion introduced during the amplification also improves with decreasing Q (Section IV-G).

E . Large-Signal Injection In thepreceding discussion of locking, I El was assumed to be small and A was approximated by A0 for the injectionvector. If one wanted to be more precise, the length of the injectionvector could be varied as the operating point moved. However, as long as the percentage variation of A from A0 is small, this modificationdoes not bring about any essential change in the discussion. If the assumption of small \ E l is removed, however, A may be substantially different from the free-running valueA o .Then A must be determined from (3.1). Taking the magnitudesof both sides gives

Fig. 21. Vector relation between Z and AZ.

The possible values of A are obtained by plotting A times the distance fromZ ( A ) to Z(wJ as a function of A and equating i t t o El as illustrated in Fig. 20. The minimum of A Z(wi) - Z ( A ) occurs before Z ( A ) passes through the point on the device line closest to Z ( w i ) . If [ E l is sufficiently large, (3.5) hasonlyonesolution, which generally represents a stable locking. For smaller El , there are generally three solutions. Only the solution with the largestA gives the stable locking. The other two areusually unstable as we shall see in Section III-G. Below the solution is usually unstable.

The terms in the square brackets are the first two terms of the Taylor expansion of Z(A +AA). When Z(w) i applied to I , s it is interpreted to be Z(wi). When Z(w) is applied to A I , w takes a value appropriate to I . This may be a complex quanA tity with itsreal part being w,. The imaginary part determines whether AI decays or grows. Neglecting the second-order term in (3.9) gives

aZ( -- A ) I AI I I COS 9 + [ Z ( w ) - Z ( A ) ] A I = 0
aA

(3.10)

I &I,

F. Locking Stabddy [36]


Suppose that the lockinghas been established and the steady-state current I is Bowing at wi. The corresponding equation is given by

where [ Z ( w ) - - @ ) ] I has been cancelled with E from (3.6). The first term on the left-hand side contains AI1 Z COS C . P From (3.7)and (3.8), it is calculated to be

I AI I I cos

I AI I Aeiuir$[ei(Wpt+f+il)+ e-i(rrti-f+it) 1
A
ei(opt+f)

[1 + e-*i(upt+f+it)1
(3.11)

[Z(O~) Z ( A ) ] I = E. -

(3.6)

Assume that I is perturbed by a small amount AI and ask whether or not decays withtime. If i t does, such an operatAI ing point is stable. T o determine the behavior A I , let I and Combining (3.10) and (3.11)gives of AI be expressed by

I = AeiVi'

I AI I ei(++f)

(3.7) where

[Z(w)

- K]AI = 0

(3.12)

where up as well as the magnitude of AI is a slowly varying between function of time compared to one R F cycle. The angle I and AI is given by

9 = opt

+ { - wit.

(3.8)

Since the magnitude of AI times cos @ is the increase AA of the R F current amplitudeA due to I as illustrated in Fig. 21, A the pert-urbed equation becomes

(Zb)- [ k ( A )

aZ( A ) + -I AI I cos @]> aA

If K were a constant, whether or not AI decays with time could be determined by investigating the locus of Z(w). According to the Nyquist stability criterion applied(3.12), AI to decays if Z(w) does not enclose K on the Smith chartwhen w varies from - 00 to 00.Unfortunately, K contains wI as well as t and the Nyquist criterion is not directly applicable. However, since i t is a slowly varying function of time compared to one F cycle, K can be considered to be a constant to R determine the instantaneous behavior AZ. If a nonzero AI of exists, Z(o) must be equal tb K.This equality determines the

1394

PROCEEDINGS OF THE IEEE, OCTOBER

1973

Fig. 22. Stability circle for explaining Type-1 and Type-2 instabilities.

imaginary part of w and hence a for a given w,. As is obvious from (3.13), K is located on a circle centered at Z ( A ) +(aZ(A)/dA)(A/2) and of radius (aZ(A)/dA)(A/2). If the circle is located completely outside the region enclosed by Z(w), AI is expected to decay with time and the corresponding locking is stable. The circle will be called the stability circle. If a part of the circle is enclosed by the impedance locus, the stability is not guaranteed. But this does not necessarily mean unlocking. Referring to Fig. 22, suppose that AI exists and its frequency w, is between 01 and w. Then K is located at Z(w,-ja) with a being real. Otherwise Z(w) cannot be equal to K with the real part of w being w,. If K is located at the point indicated by X, a is positive. This means that the magnitude of AI isincreasing with time. At the same time, K rotates counterclockwise since wi is larger than o in this case. In other words, w, isalsoincreasing , with time. When up reaches wq, the growth of AI stops. Beyond 02, AI starts decaying. If the locus of Z(wi-ja) intersects the stability circle when a varies from - QO to po k e e p ing w i constant, K does notrotate beyond the intersection with Z(wi-ja) since up becomes larger than ui and K reverses its'rotational directionbeyond theintersection. If the intersection corresponds to a negative a as illustrated in Fig. 22, AI dies away as K approaches the intersection. This means thatstable locking is possible. However, since AI initiated bynoise is amplified between u and &, the outputof 1 the oscillator will contain excess noise near a.If the locus of Z(wi-ja) intersects the stability circle with a positive a,AI increases indefinitely as K approaches this interesction and the locking is not stable. This be called Type-1 instability. will If Z(ui-ja) does not intersect the stability circle, K keeps rotating and the growth and decay cycle of AI repeats. If the net gain in one cycle (one cycle may correspond to several rotations, .since the Riemann surface may be multiple-fold) is negative, AI will eventually die away so that the locking is stable. If on the other hand, the net is positive, the lockgain ing is not stable. Thiswill be called Type-2 instability.

Fig. 23. Stability circka for possible operating pointa graphically determined i Fig. 20. n

F g 24. Graphic interpretation o the condition of the i. f


negative elope in the curve shown in Fig. 20.

the stable locking. There is one more possibility with a very small A , which represents an unstable case according to the preceding stability criterion.This possibility disappeared from our discussion because of approximating A by Ao. In the examples of Figs. 20 and 23, Type-1 instability was found at A Swhere A Z(wi) - Z ( A ) is decreasing with A . This seems to be a rather general rule. The condition for the negative slope d -A Z ( O ~- Z ( A ) < 0 )

dA

is equivalent to

G . Stability of the Graphual Solutions I t can be determined by the stability circle which of the three solutions in Fig. 20 is stable. The stability circles corresponding to the three solutions are drawn inFig. 23. I t is clear that the solution with the largest A is stable, while the other two are not stable. One intersects Z(wi--ja) with a positive a causing Type-1 instability and the other is contained in the region where a is positive causing Type-2 instability. Similarly, in the case of small-signal iifjection illustrated in Fig. 13, the dotted arrow is unstable while the solid arrow represents

Referring to Fig. 24, if the triangleformed by Z ( A ) , Z(wi), and Z ( A ) + A ( a Z ( A ) / a A ) is rotated counterclockwise by 4 to take into account the multiplication ej4 in the second term, by one can immediately see that the inequality gives the condition of enclosing Z(ui) by the stability circle. However, if the stability circle encloses Z ( w i ) , it is most likely that the locus of Z(wi-ja) intersects the stability circle with a positive a causing Type-1instability. In other words, if the slope of A I Z(wi) - Z ( A ) I is negative, Type-1 instability generally will take place. A rare exception is when the locus of Z(wi-jru) for positive a is confined in a small area on the complex plane. An extreme example is Z ( w ) = R independent of w . But in this case, the validity of the entire discussion becomes questionable.

KUROKAWA: I N P C T I O N LOCKING OF MICROWAVE OSCILLATORS

1395

H . Locking of Single-Tuned Oscdlators In order to compare the large-signal injectionwiththe small-signal case, a single-tuned oscillator will be considered. A possible device line and the impedance locus are shown in Fig. 25. A stability circle can be drawn for each operating point on the device line and the envelope of the stability circles is shown by the dotted line together with the stability circle with its center on the impedance locus. The stability circles located to the right of this particular one have net positive gains for AI and the corresponding operating points suffer from Type-2 instability. Thisis because K stays longer in the a-positiveregion than in the a-negative region and the magnitude of a is symmetrical on bothsides of the impedance locus. For stable locking of single-tuned oscillators, the center side of of the stability circle must be located on the left-hand the impedancelocus. The current amplitude must be larger A than A , associated with the stability circle centered on the impedance locus. AI Z(w,) - Z ( A ) 1 as a function of A is shown in Fig. 26 for sixdifferent points on the impedance locus. The negative-slope part suffers from Type-1 instability as we discussed in Section111-G, and the area below A , suffers from Type-2 instability. The remaining stable portions are indicated by thick lines. If Z(wi) is inside the stability circle centered on the impedance locus, unlocking takes place only by Type-1 instability. If Z ( a ) is outside the envelope of the stability circles, unlocking takes place only by Type-2 instability. The output spectra just prior to locking (or just after unlocking) are different for two types of instabilities.For Type-1 instability, the spectrum lines are crowded near the injection frequency, and for Type-2 instability, the spectrum lines are spaced apart as shown in Fig. 27. This may be reA lated to the fact that I grows near w i for Type-1 instability while no components nearwi exist in AI for Type-2 instability. Between the envelope and the stability circle centered on theimpedance locus, thereare,in general,two frequency ranges in which both types of instabilities play their role. In Fig. 25, however, the upper frequency range almost disappeared. Consider the point immediatelyinside the envelope in the lower frequency range. Curve 2 in Fig. 26 corresponds to this injection frequency. As the injection signal is gradually increased, the locking is expected to take place at a1 and the corresponding amplitude is A,. Thereafter,theoperating point follows the curve until the maximum point is reached. A further increase of the injection voltage brings the operating point to the other branch (thick portion) of the curve. There Reversing the direction of is a jump in the current amplitude. the injection voltage variation, the curve is continuously followed until the minimum point asis reached. Then the operating point jumps to the initial branch unlocking takes place and at al. Next, consider curve 3. As the injection signal is increased,locking is expected to take place a t the minimum point aa. In practice, however, the locking usually starts at a4 with a slightly higher voltage. A sufficient increase of the injection voltage brings the operating point to the other branch a as of the curve with jump in the amplitude well as the phase. The reduction of the injection voltage after the jump brings the operating pointdown to aa and the unlocking takes @ace. The fact that the locking does not take place at the smallest possible injection voltage for stable locking is understandable since the stability criterion is derived by asking whether or not the locking is stable once i t has been established and not by asking whether the locking will take place. The condition

Fig. 25. Impedance locus and device line for the discussion of large-&nal injection. The envelope o stability circles and the stability circle f centered on the impedance locus are shown by dotted linee.

Am

Fig. 26.

A I Z(*) - Z ( A ) I aa f d o m of A for six different injection frequencies indicated in Fig. 25.

for the onset of locking is not yet well understood for largesignal injection. Inthe preceding discussion, theinjectionvoltage was varied while the frequency was kept constant. If the injection frequency is varied while the voltageis kept constant, thefollowing conclusions are expected from Figs. 25 and 26. For small-injection signals, unlocking takes place only by Type-1 instability and the phase t$ changes by approximately 180" when ai is swept over the locking range. For large enough injection signals, unlocking takes place only by Type-2 instability and the phase t$ changes considerably more than 1800, the upper limit being 270" for single-tuned oscillators. In between, there is a voltage range where both types of instabilities are involved and the amplitude A as well as the phase t$ undergoes discontinuous changes depending on the history of the locking. Cartwright [37] has discussed thesame discontinuous changes in Van der Pol's oscillator. However, since its device line is perpendicular to the impedance locus, the phenomena are not as pronounced as previously discussed. The discontinuous changes are clearly observable with single-tunedIMPATT oscillators since their device line is considerably slanted.

I . Lacking Around a h o p I t is interesting to apply the above stability criterion to the case in which the impedance locus has a loop. Suppose that the injection frequency w: is located where the free-running

13%

PROCEEDINGS OF TEE IEEE, OCTOBER

1973

PJPo=

(a) a. -17.5dB, b. -17.6dB, c -17.8dB,

d.

- 18.0 dB, e.

- 20.5 dB.

Pd/P@=

- 12.8 dB, b. - 12.9 dB, c - 13.0 dB, d. - 13.1 dB, e. - 15.5 dB.
a.

(b)

Fig. 27. Spectnun changes for (a) Type-1 and (b) Type-2 instabilities The pictures a e taken w t a r ih single-tuned IMPATI oedllator. fo: 9450 MHz. PO=70 mW. Horizontal scale: 3 MHz/div.

Fig. 28. Stability circle for a small-eignal injection.

oscillation is normally inhibited as shown in Fig. 28. If the injection signal is small, the locus of Z(oi-ja) shown by the dotted line will intersect the stability circle with a positive a causingType-1instability.Lockingcannot be achievedno matter how closely Z(wi) is located to the device line. If the injection signal is large, the stability circle is situated completely outside the region enclosed by the impedance locus.

Fig. 29. Stability circk for a medium-signal injection.

This means thatstable locking is possible at frequencies where the free-running oscillation is inhibited. Between these two extremes, the stabilitycircle may be located as illustrated in Fig. 29. In this case, the locking will be stable but the out-

KUROKAWA: INJECTION LOCKING OF MICROWAVE OSCILLATORS

1397

Fig.

30. Lacking around a loop. (a) The unlocking takea place in the middle of two locking -ea. (b) A l d B increaae of the injection power eliminatee the unlocking region. The pictures are taken with a multipletuned MPATT oscillator. fo: 9470 MHz.PO: mW. Horhntal d e : 3 MHz/ 52 div. P ia 18 dB below POfor (a) and 17 dB for (b). i

put of the oscillator will show excess noise near wt and or in addition to the spectrum at w i . I t follows from the preceding discussion that if the injection frequency is swept from 01 to w~ keeping the injection power sufficiently large, stable locking can be achieved around the loop. I t is also possible that, if the injection power is properly chosen, the locking takes place in two frequency ranges separatedby a region where the oscillation is unlocked. Fig. 30 shows an example. With a certain injection power, the unlocking takes place in the middle of two stable locking ranges. An increase of the injection power Fig. 31. Exs~laining stability of the operating the eliminates the middle unlocking region. point Pa of a freerunning cadlation. Shemanaev Ivanov and [38] have discussed doubletuned oscillators extending Van der Pols theory. They predicted the same unlocking in themiddle of two locking ranges. modification is, however, necessary. If the free-running oscillaa Note thata double-tuned circuit may have loop in the impe- tion is considered as the limit of reducing the injection power dance locus depending on the coupling coefficient. while keeping the injection frequency the same as the freebe runningfrequency, theabovestabilitycriterionshould J. More about Free-Running Stability applicable. If the stability circle is small and mostly or comBriefly returning to the free-running stability discussion, pletely enclosed by the impedance locus as shown in Fig. 31, the operating point in Fig. 11was saidto be stable. A slight the small current with close to the frequency corresponding Pt o,

1398

PROCEEDINGS OF THE IEEE, OCTOBER

1973

to give a clear insight into the locking transient, response to FM signals and high-frequency components of noise. In this section, we will derive differential equations for the R F current amplitude A and phase $. The study of these equations will provide the information needed to understand the dynamic behavior of injection-locked oscillators.

A . Difftwential Equations In the ac circuit theory, the time derivativeeverywhere is replaced by multiplication byjw. This is justifiable since

dn - Re f A e i ( u t + + ) ) = Re { (jw)mA&ot++) 1
dt"
when A and $ are constant. In the study of oscillators, however, both A and $ must be allowed to be slowly varying function of time. Then the time derivative becomes

Similarly, to the first-order approximation, the nth derivative is given by

Fig. 32. (a) Relation between the impedance locus and the device line
o a negative resistance amplifier. (b) Determination o the operating f f point.

to P will grow. Consequently, the operating point PScannot I be stable suffering fromType-2instability.Theoperating This is most likely to occur when the point will jump to PI. small-signal impedance Z(0) is located close to P .When the t impedance locus is moved downwards, the operating point will jump to PIbefore the intersection PSactually disappears.

Re { [ j ( w

+): + f $ ] " A e j ( ~ t + + ) )

(4.2)

K. Negutiue Resistance Amplifiers

This means thatRe ( Z ( w ) A d " + + ) ) will give thevoltage across the impedanceZ(w) when w is replaced by (w+ (@/tit)) - j ( l / A ) ( d A / d t ) everywhere in Z(o). Consequently, the equation of a locked oscillator becomes

If the device line doesnot intersect the impedance locus as shown in Fig. 32(a), the result is a negative resistance amplifier. When the injectionsignal is at wi, A Z(wi) - Z ( A ) may = E COS (wit e) e(t) (4.3) look like the one shown in Fig. 32(b). From this figure, the possible current amplitude can be determined for each injec- where E ( cos (wit+*) represents the locking signal and e(t) tion voltage. For small El , A is small. This is the reason why the possible noise. 9 is the phaseof the locking signal which is the injectionvector can bridge between Z ( A ) and Z(wi) which allowed to be a slowly varying function of time. Since the are far apart. As El increases, A increases but the relation is left-hand side of (4.3)can be rewritten in theform not necessarily linear. The saturation effect of the negative resistance amplifier can be studied using this relation. Care $) R e must be exercised to convert E and A to the corresponding A cos (wit quantities at the circulator ports as we shall discuss in Section IV-H and Section IV-M. Also the stability circle must be studied at each operating point to check the stabilityof the amplifier.Depending on the relative position of the device line and the impedance locus, Type-1 or Type-2 instability multiplying (4.3) by cos (wit+$) and sin (wit+$) in turn and may takeplace. The corresponding operating pointis not then integrating the results over the period T of one R F cycle give realizable in the steady state. For large injection signals, the operations of negative resistance amplifiers and of injectionlocked oscillators are similar in many respects, including the saturation effect, noise, and, sometimes, discontinuities in the gain and phase characteristics [39], [MI. = E cos ($ - e) e, (4.4)

I I

+ +

I 1

I v . DYNAMIC ANALYSIS OF INJECTION LOCKING Inthe previous section, we discussed thequasi-static analysis. I t revealed many interesting and important aspects of injection locking. However, the quasi-static analysis failed

KUROKAWA : INJECTION LOCKING OF MICROWAVE OSCILLATORS

1399

where

7 J 1--r

atanh
7 J 1--r

[Aw(t -

to)

d($r A/
1-

1 1 (4.14)

when

lAu/Au,,,l < 1 and by

In general, (4.4) and (4.5) contain higher ordertermsin &/& and ( l / A ) ( d A / d t ) ,which make the solution extremely difficult. However, if Z(w) is approximated by a singletuned circuit, they reduce to tan L
=

2-

I E I cos (4 - 9)+ e,
dt

when

I&/Au,,,l

I'>-:(

(4.15)

>l. Equation (4.14) shows how thephase

(4.8) changes after the locking signal is applied at t = O . The value

-2L(wt - w,)

- X(A) =

1 E 1 sin (4 -

+ e,

(4.9)

where use is made of (2.8).

B . Adler's Equation [31] When the injection signal is small, A is expected to stay near the free-running value A 0. Since

of t o is determined by the initial phase at t = 0. Regardless of 4 eventually settles down to 40 when the initial phase, <1 where sin # o = -&/&. This is what one expects from(4.12) with @ / d t = O . When the injection frequency is outside the locking range, (4.15) shows that undergoes a periodic variation and does not converge to a constant value. Aside from Adler himself, several authors have discussed the locking transient [41]-[47].Among them, Lisitsian [42] and Takayama [47]presented the results of numerical integration of equationssimilar to (4.8) and (4.9). Takayama showed the effect of X ( A ) while Lisitsian neglected it.

I&/&/

D . Spectra of Unlocked Oscillator Prior to locking or after unlocking, the output spectrum of from (2.11), (4.9) reduces to an oscillator shows a triangular shape as shown in Fig. 27(a). I t is fascinating and slightly surprising to see this spectrum (4.11) change on the spectrum analyzer. Several authors have successfully explained this change [48]-[51]using Adler's equa[SO] i especially s tion. The methodpresentedbyArmand appealing to the present author. Armand calculated the Fouriercomponents of Aei(oil*) directlyusing (4.15).The result is listed below: when 9 = 0 and e, = 0. This is the famous equation derived by Adler in 1946 for triode oscillators and subsequently discussed Frequency Complex Component by many authors. Under the locked condition,@/dt = 0. Since sin 4 I < 1, the locking range Aw,,, is given by a j tan - . A Wi
00)

2L(wi - 0,)

+ X ( A 0 ) = ~ L ( w,

= 2LAw

(4.10)

(4.13)
wi 2n Comparing with (3.4), notice that cos 8 is missing in the we denominator of the right-hand side. This is because the variation of X ( A ) with A is neglected in the derivation of (4.12). wi nn Under this condition, the device line is perpendicular to the impedance locus and cos 8 = 1. In low-frequency approximations,devicesrespond to external perturbations instantanewhere ously. There is no phase lag in their response, which means Awm that 7 ( A ) is zero or a t most a constant independent of A , sina = - - 7 -900 5 a 5 9 0 ' Aw which can be absorbed in Z(w). Consequently, the effect of -( A ) is inherently absentunless it is deliberately introduced. X a = - Aw COS a. The effect of X ( A ) is, however, significant in most microwave oscillators and will be discussed in Section IV-E. As the locking signal increases, Aw,,, increases, a approaches f90, and the spacingbetween adjacent spectrum lines 0 deC . Locking Transient creases. At the same time, t a n a/2 increases and approaches The great advantage of Adler's equation is its simplic- 1 Consequently, more and more spectrum lines become visi. ity. I t can be integrated directly and theclosed-form solution ble. The added lines are on the wo side of w;. The amplitudes is given by decrease in geometrical progression and hence linearly on the

+ +

1400

PROCEEDINGS OF THE IEEE, OCTOBER

1973

logarithmic scale of a spectrum analyzer. Thus the above reEl COS (40 e) > 0. (4.26) SRL -COS 4 0 > 0 SRL sult explains very well what is observed in Fig. 27(a), c-e. The Ao COS e small lines on the oppositesides of wo are due to the amplitude variation neglected in Adler's equatio,? The stability the free-runningoscillation requires positives. of With positive s, the first condition is generally satisfied, since E. Effect of Nonlinear Reactance [52] El is small. Then, the second condition becomes Adler's equation neglects the variation of the R F current COS (40 e) > 0. (4.2 7) amplitude and hence the effect of nonlinear device reactance. As a result, a factor of l/cos 0 is missing in (4.13). T o underEquations (4.21) and (4.27) uniquely determine for a given stand thefirst-order effects of the nonlinear reactance,' we will Aw. The result agrees with the conclusion obtained in Section consider (4.8) and (4.Y) under the conditions P = 0, e, = 0, and 111-G in relation to Fig. 13, as it should. e = 0. In the steady state, they reduce to . { % + R L - X ( A ) J A = E ]COS+ (4.16) F. The Circulator Output So far we have investigated the current through the de. ) {-2L(wi - 0 - X ( A ) ] A = El sin& (4.17) vice, which is equal to the current through the load RL in the will be assumed to be small. Then, A is expressed by equivalent circuit, when Z ( w ) is represented by (2.7). Most Ao+AAo where AAo expresses the small deviation from the injectionexperimentswith microwaveoscillators are, howfreehnning value Ao. Retaining only the first two terms of ever, done through a circulator as illustrated in Fig. 12. The the Taylor expansions of X ( A ) and X ( A ) around A0 in (4.16) load current I' at the circulator output is different from the current I at the oscillator port a-a'. For quantitative compariand (4.17) and substitutirig (2.11) and (4.10) gives sons between the theory and experiments, I' must be calcuSRLAAO E COS 40 = (4.18) lated in terms of known quantities. This calculation could be rRLAAo = E sin 40 (4.19) eliminated if the initial differential equations were written in -2LAoAw terms of wave parameters as they were done by Quine [54] where 40is the steady-state valueof 4 and and also by So [55]. However, this approach was not used here because of the reasons stated in Section 11-A. ai?( A ) aX(A) The current I' is given by SRL = - A rRL = A (4.20)

+I

I I

1231

I I I I

aA

8A

s and r are called the saturation factors of the device resistance and reactance,respectively. s was introduced by U s o n [53] for the noise discussion of free-running oscillators. Solving (4.18) and (4.19) for Aw gives

(4.28)
This is because, referring to Fig. 33, I is the sum of the current I' at the circulator output and the current at the circulator input which is given by E / ~ R LA slightly more convinc. ing argumentmay be as follows. The wavereflected back from the oscillator to the circulator comes out from the circulator output port andgoes into the load. Equating two waves, one from the oscillator to the circulator and the other from the circulator to the load, gives

AW = -

2
Q=t

Pi sin (40
COS

+ 0)
e

(4.21)

where 0 is defined through the relation tan 0 = S

-wse~w.

(4.22)

Since sin (#o+@) is less than 1, the locking range Au,, identical to (3.4) is obtained. To make sure that (4.21) represents the stable solution, the behavior of small perturbations AA and & from the steady-state solution must be investigated. Substituting A =Ao+AAo+AA and $=@o+@ into (4.8) and (4.9) gives the differential equations for AA and A$ t:

V - RLI - V'

- -

- RLI'

2 d R ~

24%

Since V = E - RLI and VI= -RLI', the above relation gives (4.28). From (4.18) and (4.28), the current I' is calculated in terms of known quantities:

dA A ~L-+sRLAA

I' = (A0
=

dt

I El A 4 s i n g o

(4.23) (4.24)

E + AAo)ei% - -L ~ R

dA4 -2LAo -- rRLAA =


at

1 E I A 4 cos 40.

Eliminating AA from these equations gives

(4.29)
When Aw varies from-&to 4ochanges by 180' from 900-8 to -900-8, but the phase 01I' changes by 180" times [ 1 + ( 2 / ~ ) 4 ( P i / P ~cos e]. When cos e= 1 and the injection ) power is 20 d B below thefree-runningoutput power, the phase change is approximately 192'. This is what one should expect after calibrating out all the additional phase shifts in the connectingwaveguides and the circulator. ' 1 The output power is given by +RLI1 * and the output

+&,

I $ For the steady-state solution to be stable, A has to decay with time. This requires that the coefficients be positive.

KUROKAWA: INJECTION LOCKING OF MICROWAVE OSCILLATORS

1401

i
Fig. 34.

The variation o a factor in (4.30) as a function o Aw. f f


I -0.

Fig. 33. Equivalent circuit of a single-tuned oedllator with a

circulator.

where use is made of (4.19) with through

Let A$ bedefined

$=t$o+At$+*
power variation by [52]:

(4.32)

Po

-=2

AP Po

Pi (-(=2

then Aqj is expected to be small. Substituting into(4.31) gives


cost$o

2
s

- 1)

2LA -

(Po (-- 1) [(1 1Pi 2


s

(E)
cos 8

at

+ 1 E I At$

COS $0

- 2LA at

d e

(4.33)

In thefrequency domain, & is given by

-2joLAqCf)

At$Cf> =
(4.30)
The first term in the brackets gives an elliptical variation and the second term a linear variation with Lso. Their difference may vary as shown in Fig.34. Consequently, the power variation is convex when s < 2 and concave when s>2. The observation of the output Variation gives a fairly good idea about the magnitudes of s and 8 and hence I from (4.22). When s = 2, the first-order power variation vanishes. Note that s = 2 corresponds to the maximum power point, as the following discussion indicates:

2jdA

+ I E I cos
W

t$o

+Qat

rv

j a% Q t

(z

@a COS t$o

*Cf).

(4.34)

Once Aqj is obtained, t$ can be calculated from (4.32). The phase at the circulator output is given by

-PO
aA

= -(RL$)

aA
=

a -[X(A) - &IaA
- &]A

A2

2
where

[X(A)

A2 aX(A) + y --0 aA

gives the power maximum point. But the above relation is equivalent to aX( A ) A s

(4.36)
Apart from the constant phase follows if

RL = -

aA

- = - RL.
2 2

t$~', output phase the

closely

The power variation of lockedoscillators was firstdiscussed by Huntoon and Weiss [l] using a very general approach. Several other authors have also discussed the same power variation using various approaches [56]-[SI.

(4.37)
Note that (4.37) guarantees Aqj being small. At the center of the locking range, $0-0 and (4.37) reduces to w<&. For high-fidelity PM (or FM) reproduction, the modulation frequency must be smallcompared to the lockingrange bo, 1591, [ml.

G. Phase Modulated Injection Signals


Suppose thatthe injection signal phase is modulated; ! ! namely, P is a function of time. P itself may be large. but a/& be small compared to o for the approximation must used in the derivation (4.4) and (4.5) to be valid. Therefore, of a twill be assumed to be small. T o illustrate the methodof / d analysis using the simplest yet meaningful case, wewill assume that r = 0. This avoids lengthy equations in following the discussion. Under the assumption that the oscillator is locked and e, = 0,(4.9) becomes

H . Noise Oscillator noise is an old subject. In 1938, Berstein [61] discussed the noise of free-running oscillators using the Einstein-Fokker equation. Since then, theSame subject has been discussed by many authors [ S I , [62]-1741 including Rytov and Malakhov. Both Rytov and Malakhov discussed not only free-running oscillators but also locked oscillators. AM and FM noise had been, however, of little practical interest until the importance of microwave solid-state oscillators for F M

1402

PROfXEDINGS OF THE IEEE, OCTOBER

1973

amplification was clearly recognized and a separate measurement of their AM and FM noise was reported by Josenhans [75]. The most straightforward method to discuss these noise components may be t o utilize (4.8) and (4.9). T o illustrate the method without giving lengthy equations, wewill again assume small-signal injection and Y = 0 [76]. With finite e, and e, and Y = 0,(4.23) and (4.24) become

Forthefree-running oscillator, a similar calculationfrom (4.8) and (4.9) with E = O gives

KT
(4.47)

-2LAo-

dA4

at

I El A + c 0 s 4 ~ e,. +

. -

(4.39)

In the frequency domain, and AA are given by &

(e-

;2 ($ j+

(4.48)

A4 =

- e.Q
2jwLAo

+ I E I cos

40

If e(t) represents random noise

e,e,* = 0.

(4.43)

The factor 2 in the right-hand side of (4.42) appears because the factor 2 in front of the integrals of (4.6) and (4.7) gives a factor of 4 when it is squared and the cosine and sine in the integrands give a factor of 3. Under the assumption that the device noise is predominant, ] e l f is related to the effective noise temperature T of the device as follows:
-

k T = - - els ela RL =- e l 2 4 R A )~ ( R L &> ~ R L RL R,

The formulas (4.47) and (4.48) were derived by Edson [53] through a different approach. They are identical to those obtained from (4.45) and (4.46) by letting Pi-tO. However, this method of taking the limit is not justifiable as the derivation of (4.47) and (4 48). Somewhere in the process, as the unlocking condition approaches, the assumption of small & will break down. A comparison of (4.45) and (4.47) shows that the low-frequency components of the phase fluctuation improve when injection locked. On the other hand, the amplitude fluctuatim does not improve. As a matter of fact, * increases when w i #COO, since cos* 40is less than one. If the effect of upconversion noise from the bias circuitis included in e, and e,, they are most likely correlated. Then a term involving e&,* will appear in addition to the terms with w a n d m w h e AA * is calculated. The calculation itn self is straightforward. However, without definite knowledge about the correlation, the result is not worth presenting here. For the discussion of upconversion noise, the reader is referred to [52], [76]-[81].

I A A1

I I

I. Noisy Injection Signal


If the injection signal contains FM noise, 9 is randomly fluctuating.Then 4 also-fluctuatesrandomly.The mean9 square fluctuation of $ due to the fluctuation in can be calculated from (4.32) and (4.34):

where 7 is the circuitefficiency defined by R L / ( R L + & ) . (The value of in [52], [76] is one-half the value used here. The other half is attributed to the negative frequency band in the references. However, in the present paper, only the positive frequency will be considered.) The mean-square fluctuations of & and AA can now be obtained from (4.4Q) and (4.41) :

kT -

A more complete account of the amplitude and phase fluctuations of a noisy oscillator with a noisy injection signal is presented in [76].

$PO

J. Effects of Nonlinear Reacbnte on Noise


2kT 9RL

In theprevious noise calculations, the saturation factor is t assumed to be zero.. When r is finite, the calculation remains straightforward but becomes lengthy. In many oscillator experiments, the operating point is near the maximum power point so that s is of the orderof 2 and the frequency interest' of is low enough sothat thefollowing relation is satisfied:
(Qat

:' -

+ pi Po

(4.50)

If this is the case, the time derivative on the left-hand side of (4.8) can be neglected compared to theremaining terms. This

KUROKAWA: INJECTION LOCKING OF MCROWAVE OSCILLATORS

1403

is the approximation used by Thaler et al. With this approximation and finite r , the phase and amplitude fluctuations corresponding to (4.45) and (4.46) are given by

[SI.

1 &I2
-

r l ~ o O S ~e C
=

KT 1 --

(Q-

pi cos2 (t$o e) f> -" p0 cosz e

(4.51)

1Olog- = 10 log C

1 A ,' I ~ B A
L

[dB in B H ] z.

K. Noise Power SuppIcssion Factors

IA A I ~ = -

2kT

(Q-

:j2 +

pi COS^ (do e) p0 COS~B

(4.52)

Heins et al. [82] calculated ] A + / and * at the center of the locking range assuming s = . Their results are, of course, identical to (4.45) and (4.49) with cos t$o= 1, although small-signal injection was not assumed. They defined I the noise power suppression factors S and St by the ratio of the to ratio of the to free-running - and 2, respectively:

It$-t$~/

It$-t$~l It$/

The phase fluctuation corresponding to(4.49) becomes

I t $ -t$op
s 2 '
=
+

(3

(4.55)

As we studied in Section IV-F, the amplitude and phase at the circulator output are expected to beslightlydifferent from A and 6. Consequently, their fluctuations are also expected to be differentfromthose of A and t$. Under the /At$'/ turns to out be same approximation used above, (l+d(Pi/Po) cost$^)^ times that given by (4.51). Onthe otherhand, isgiven by replacing Pi/Po inthenumerator of (4.52) by

1AA'I

Note that * becomes larger or smaller than the freerunning fluctuation when the quantity inside the above bracket is larger or smaller than cosz (&,+e)/ cosz 8, respectively. Since this is equivalent to the condition that

IA A ' I

Note that SI+&= 1 and that the total FM noise is given by SI times the FM noise in the injectionsignal plus S times the I FM noise of the free-running oscillator. I t might, however, be more appropriate to use the corresponding quantities at the circulator output. SI'given by l+\Qd is times SIand S by (1+ d/pilp~)~ SZ. differences are, however, , ' times The negligible when the injection signal is small and the frequency of interest is low. Independent of the calculation by Hines et al., SIand SZ were also calculated by Isobe and Tokida [83], [84] using an analog computer and by Sugiura [85] using an extension of Adler's equation. Sugiura's calculation is not restricted to the center of the locking range.

L. Large-Signal Injection In the case where the injection signal is large,as discussed by Hansson and Lundstrom [86], %+RL- X ( A ) is no longer close to &+R&-IE(Ao) which is equal to zero from (2.11). For small perturbation AA from the steady-state value A , therefore, the second term on the left-hand side of (4.8) becomes

is larger or smaller than zero, the amplitude fluctuation reaX( El cost$ A A mains the sameas the free-running fluctuation whenthe d l -- A ) A A A = [ El cost$ lator is adjusted for the maximum power or for t$o=O. From aA A (4.30), t$a=O corresponds to the stationarypower point in the sRLAA. (4.56) locking range. The expression for lt$'-t$o'l is given by replacing cos* (t$o+B)/cos* 60 inthenumerator of (4.53) by Similarly, the corresponding term in (4.9) becomes [COS'(t$o+e)/cm' @]+(QB~ w/%)' COS' $0. - T(A A A ) ] ( A A A ) I) The results of FM and AM noise measurements are often [-22L(~i - & . presented in terms of the root-mean-square frequency deviation and thecarrier to noise ratio in decibels in a certain bandwidth B . They are given by

+I

1404

PROCEEDINGS OF THE IEEE, OCTOBER 1973

sion to multiple-tuned oscillators. One method to do this is to utilize the following approximation:

Fig. 3 Graphic interpretation of the stability COnditiOnS . for a large-eignal injection.

The saturation factorss and r are calculated at the operating point Z ( A ) . The equations corresponding to (4.23) and (4.24) are given by

dAA 2L-

dt

+ I E l Ac o s 4 A A + s R L A A
=

-2LA

-+
dt

dA4

I El sin4 A A - T R L A A
A
=

- I E I A4 sin4

(4.58)

I E I At$ COS$.

(4.59)

With the first two terms in the Taylorexpansion, the impedance locus is, in effect, approximated by a straight line with (4.63) into equallyspacedfrequencymarkers.Substituting (4.4) and (4.5) gives two first-order differential equations for A and 4. The subsequent analysis is almost identical to that for single-tuned oscillators. The stability criteriaas well as the noise expressions have been obtained in [88], [89]. They provide useful information about the behavior of multiple-tuned oscillators. As a matter of fact, the discussions presented in Sections II-A-III-C an attempt to give an intuitive exare planation to the results of this analysis. Several authors have simplified or extended this analysis and compared their results with experiments. However, i t is often overlooked that A and 4 are both referred to the device port and hence they have to be converted to quantities directly measurable at the output port before quantitative.comparisons are made between the theory and experiments. For example, if the transfer function from the device port to the oscillator port a-a' in Fig. 12 is given by T(w), the current I' at the circulator output becomes

Eliminating AA from (4.58) and (4.59) gives where Ed is the injection voltageto the circulator and RL is the circulatorimpedance.Note that AM-FMnoise conversion may take place in this process of conversion. Similarly, the injection voltage applied to the circulator must be converted to the injection voltage seen by the device for the calculation of A and 4. The conversion of Ei to E has been done when the locking range is calculated in terms of P and POin both lossi less [88]and lossy [go] cases. Finally, it is important torealize that thewhole analysisis based on the approximation(4.63). The first two terms of the Taylor expansion become often inadequate when even @/dt<<u and ( l / A ) ( d A / d t ) <<u are well satisfied. This i s especially true when the injection signal is large. The stability criterion for large-signal injection obtained by a straightforward extension of the analysis will give erroneous results unless the circuitis single-tuned orclose to it. Also, the case discussed in Section III-Jcannot behandledbythisapproximation
(4.63).

For the stable operation, & has to decay with time. This requires that the coefficients be positive:

I El c o s 4
A

+;sRL>O

(4.61)

Referring to Fig. 35, the first condition indicatesthat thecenter of the stabilitycircle be locatedon the left-hand side the of impedance locus. Similarly, the secondconditionindicates that Z(wi) be located outside the stability circle. They are identical to the stability conditions discussed in SectionIII-H, as is expected. The noise under the large-signal injection can be discussed by addinge, and e, on the right-handsides of (4.58) and (4.59), respectively. The calculation of the amplitude and phase fluctuations is straightforward. However, i t is worth noting that e, and e, at the operating point Z ( A ) may be quite different [87] from those at the free-running operating point Z ( A o ) .

M . Multiple-Tuned Oscillator When several tuning elements are provided and they are adjustedby trial anderror, microwaveoscillators tend' to showmultiple-tunedcharacteristics as discussed inSection II-F. Consequently, it is desirable to extend the above discus-

N . Multipk-Device Oscillators In parallel with efforts to increase the output power of individual devices,it is always interesting to explore the p i bility of combining the output powers from several devices. The combined oscillator can provide a higher power than a single-device oscillator, or it can have longer life by biasing individual devices more conservatively, or both.In early days of microwavesolid-statedevices, the above possibilitywas studiedbyfirstbuilding single-deviceoscillators andthen combiningtheir outputsbyhybridcircuits [91]-[93] or a transmission line [94]. When the number of oscillators to be combined is two [95]-[98], the osallator performance can be maderelativelystable.Once a stablecombiningcircuitis

KUBOKAWA : INJECTION LOCKING OF MICROWAVE OSCILLATORS

1405

Fig. 36. A poeaible p w r combiner. Nonreciprocal circuits help suppress oe undesired interactions between devices.
CAVITY

Fig. 38.

Single-cavitymultipledevice

oecillator.

lines, each about a quarter wavelength long terminated by a device and arranged radially about a common bias network and a common output circuit. The resistor &u incorporated into each coaxial center conductor eliminates the mode prob, lem. The capacitanceC between the outputcoupling disk and each coaxial center conductor provides the necessary coupling to the common load RL.The bypass capacitor in the bias network is located about a quarter wavelength away from the hub of the oscillator. The second one proposed by Magalhaes and the author [103]is illustrated in Fig. 38. In this oscillator, each device is mounted at one end of a coaxial line which is coupled to the sidewall of a common waveguide cavity. Each eoe coaxial line is located where the magnetic field b c m s maximum when thecavityresonatesinthe desired mode. The other end of the coaxial line is terminated by a tapered absorber. The quarter-wavelength transformer in each coaxial line improves the circuit efficiency and, at the same time, i t Fig. 37. A schematic diagram of Ruders symmetrical helps to suppress parametric-type oscillations by keeping the mnltiple-device oedlktor. circuit impedance low over the(angular) frequency range from wo to ~ W O . The analyses of both oscillators have been carried out ouilt, the small-signal injection locking of the combined os[la], Although the eigenfunction approach was ex[105]. cillator can be performed without difficulty. However, when tensively used to perform complicated calculations systemthe number of oscillators to be combined exceeds two, i t becomes increasingly difficult to stabilize the combined oscilla- atically, the foundation of the analyses was a straight extentor. Small changes in the load condition, or in the bias supply,sion of the oscillator theory describedearlier. The results show how such oscillators can be made free from the mode or even in the environmental temperature bring about sudden changes in the mode of operation. The output power as well as problem. I t is alsoshown that the small-signallocking bethe oscillation frequency literally jumps from one value to havior of such oscillators is identical to that of single-device another. This is the well-known modeproblem of multiple- oscillators. This is experimentally confirmed with the second oscillator. device oscillators. Although several reports indicate that the power combiners of this type could be operated under laborabe tory conditions, it must noted that their circuit adjustment 0. Validity of the Theory is critical. One scheme to improve the stability is illustrated in From the beginning of this paper, we have assumed the Fig. 36. In this scheme [99]-[loll, several oscillators or nega- existence of the device impedance - Z ( A ) . - Z ( A ) expresses tive-resistanceamplifiers are connected together by circula- the device voltage when, multiplied by the current, both at torsanddrivenby a master oscillator. Duetothe non- thefundamentalfrequency of oscillation. All the possible neglected. Since the reciprocal property of the circulators, undesired interactions effects of other frequency components are between active devices are reduced. As the number of oscilla- device is nonlinear, in reality, the voltage as well as the curtors or amplifiers to be combined increases, however, the loss rent is expected to contain harmonic components. When the in the circulator will soon exceed the addedpower in the stage. harmonic components aren% large, however, the theory gives This limits the possible number of oscillators of which the surprisingly accurate results, thanks to the integration over output powers can be combined efficiently. the period of one R F cycleperformed to obtain (4.4) and Recently,two multiple-device oscillators free fromthe (4.5). When the amplitudes and phases are all slowly varying mode problem were proposed. The first one proposed by functions of time, they can be considered to be constant over Rucker [lo21 is illustrated in Fig. 37. This symmetrical os- one R F cycle. Then the orthogonality between cosine and sine cillator consists of N ( N = S in Fig. 37) coaxial transmission functions wipes out the integrations involving harmonic comn

1406 ponents. The quasi-static analysis uses the result after the integration, which becomes, for example, (2.1) or (3.1) in the steady state. When some of the harmonic components are large, their effect may no longer be negligible. The device impedance will be a function of the amplitudesof these harmonic components as well as of the phase differences between them. An example is second harmonic tuning [106], [107]. A similar situation arises when a bias-circuit oscillation or parametric-type oscillation is about to take place or is taking place. For each case, a separate analysis becomes necessary. Experimentally, complicated phenomena are observed when locking is attempted under these conditions. For example, the injectionsignal may initiate biasoscillation and itlocks the oscillator at one of the sidebands created by the bias oscillation. From the practical point of view, it is recommended to design the oscillator circuit so as to suppress such complications. I t has been wellestablished how toavoid bias-circuitoscillations inGunn [log] and IMPATT oscillators [lo91 and parametric oscillation in IMPATT oscillators [IIO].

PROCEEDINGS OF THE IEEE, OCTOBER

1973

stabilitycriterionobtainedundertheimplicitassumption that the nonsynchronous perturbation stays in the system is, therefore, of questionable value, although i t happens to agree with the stability criterion against Type-2 instability when the circuit is single-tuned as discussed in Section 111-H and Section IV-L.

v. BRIEF REVIEW EXPERIMENTAL OF WORK A great number of locking experiments havebeen reported
in the literature. t is virtually impossible for a single individ: I ual to cover them all. Undoubtedly, many importantones will be left out. However, the following brief review is hoped to give the reader some idea about the recent activities in the field of microwave solid-state oscillators.

A . Locking Range One of the first locking experiments with Gunndiodes was et uZ. [118] and that with IMPATTS by reportedbyHakki Stover and Show [119]. They both showed experimentally that the locking range was proportional to the square rootof the injection power. Since then, many authors have reported P . Describing Functions similar results [120]-[123]. The square-root relation is held by a wide variety of circuits as long as the injection power is Theconcept of device impedance or admittancedates back to the early 1930s. In the early 194Os, the dependence kept sufficiently low. This is because the length of the injecof device admittance on the signal amplitude was explicitly tion vector in Fig. 13 is proportional to I El which is in turn used to determine the stability of free-running oscillators by approximately proportional to the square root the injection of Pierce [lll]. His result was later published by Edson in his power and the frequency markers on the impedance locus can be considered to be uniformly spaced over a narrow frequency book Vacuum Tub6 Oscillators [112]. The same concept appeared in thefield of servomechanisms in the early 1950s with range. However, as the injection power increases, the locking a different name-the describing function. The simple con- range is expected to deviate from the square-root relation becept was quickly extended to multiple-input describing func- cause A changes with the operating point. With a further intions [113], [114]. They describe the response of the nonlinear crease of the injection power, Type-2 instability will be inelement to two or more sinusoidal signals which are simul- volved in the process and the locking range becomes a more taneously applied. Later, the input signals were extended to complicated function of injection power. The square-rootrelainclude random signals. Most of the applications of describing tion holds even when the oscillator contains many active defunctionare, however, restrictedto memoryless elements; vices, provided that the mode-problem is successfully supX - 0 . Still another extension was madeby Bonnen[115], pressed. One example is shown in Fig. 39. This indicates the [116]. He introduced the concept of incremental input de- locking range of the 12-IMPAT~-diode single-cavity oscillator scribing functions. An incremental input describing function illustrated in Fig. 38. The locking range starts to deviate from is the response of the nonlinear element to an additional small the square-root relation when the injection power approaches input when the main signal is already in existence. He defined about 18 dB below the free-running outputpower. two types of incrementalinput describing functions. The Frequency Simulation synchronous one is the response to an additional small input B . LowI t is difficult to make quantitative comparisons between having the same frequency as the main signal. The nonsynthe chronous one is the response to an additional small input hav- the theory and experiments in microwave frequency range. For example, the locking formula contains the external Q of ing a fixed frequency different from that of the main signal. of Qat is usuallyestablished by The conceptof describing function was recently introduced to the circuit and the value the microwave field by Gustafssonet aZ. [ 1171. They discussed means of a locking experiment and the locking formula. For the stability of injection locking using the incremental input many purposes, such a procedure is adequate. However, this describing functions. The stability criteria derived are, in our does not check the magnitude of the proportionality factor theoretically derived. Indeed, several papers have been pubterminology, as follows. The system is stable: namely, the synchronous and nonsynchronous AIS do not grow with time lished with the proportionality factor different by a factor of circle with positive two. Kenyon used a low-frequency simulation of microwave if Z(wi-ja) does not intersect the stability oscillator circuits tocheck the proportionality factor. He built a and if the centerof the stability circle is not enclosed by the ! in the 350-kHz range, meaimpedance locus, respectively. The first condition agrees with a negativeresistanceoscillator the stability criterion against Type-1 instability. However, sured the circuit Qea and the locking range independently, the second condition differs from the stability criterion against and established that the proportionality factor in (3.4) was Type-2 instability. The discussion in Section 111-F shows that correct within 2 percent, which is within the limit of experithe frequency of AI could stay at w i but not at any other fre- mental error. A similar technique was adapted to verify variwith quency. Thefrequency of AI keeps changing with timeif i t is ous theoretical predictions about oscillator behavior not at w i . Thus the nonsynchronous perturbation with a fixed multiple-tuned circuits [124], including the noisy conditions, frequency does not stay to grow or decay in the system. The frequency jumps, and parasitic oscillation.

KUROKAWA: INJECTION LOCKING OF MICROWAVE OSCILLATORS

1407

Fin. 4 . Schematic d a r m for eelf-injection locking. 0 iga

cussed in Section 111-H. Khandelwal [132] reported his observation of two different types of locking and called one the triggered oscillation. Although his descriptionis brief, i t seems that he observed Type-1 and Type-2 instabilities.

E . Pulsed Operations Injection locking has also beenperformedwith pulsed oscillators [133]-[139]. In this case, the onset of oscillation is triggered by the injection signal rather than by noise. Consequently, the jitter at the-leading edge of pulse train is considerably improved and pulse to pulse coherency can be maintained. However, if unlocking takes place during a pulse, the pulse envelope becomes noisier than thefree-running case F . Subharmonic and Harmonic Injection So far, we have discussed onlyinjection at thefundamentalfrequency.However, a nonlinearoscillatorcanbe or theharmonic synchronized either subharmonic by the signal as discussed by Khokhlov [140]. Several authors have reported the successful subharmonic or harmonic locking of microwave solid-state oscillators [141141461. The improvement of FM noise of a subharmonically locked IMPATT oscillator has also been demonstrated [ 1471.
G . Siaeband Locking When the bias supply a free-running oscillator is moduof lated, the output shows many sidebands. If a signal with its frequency close to one of thesesidebands is injected, the oscillator can be locked at thissidebandfrequency. If the modulation frequency in the bias circuit is changed, the sidebandfrequency stays the same while the frequency of the carrier as well as those of all the other sidebands changes.Selecting either the carrier or only one of the sidebands by a bandpass filter, one can obtain an F M signal. Fukatsu and Kat0 [148] reported such a locking of an IMPATT oscillator. Their result did not quite meet the requirements for a multichannel F M terminal.

Fig. 39. Locking bandwidth versus injection power for the ~Z-XI~PATTdiode osdllator illustrated in Fig. 38. The free-running output power is adjusted to 10 W, and the frequency i 9.1 GHz. s

C . Noise The FMnoise improvement of microwave solid-state oscillators by injection lockingwas first reported by Josenhans [76]. He also pointed out that the AM noise did not improve by injection locking. This was later confirmed by Ashley and Palka [125]-[127]. FortheFM noise improvement,Hines et al. [82] and Sugiura and Sugimoto [128] reported that their experimental results on IMPATT andGunn diodes, respectively, showed good agreement with the theory. On the other hand, Ashley and Palka [126], [127] reported some discrepancy in the low-frequency range. Below 10 kHz, the FM noise improvement was not as large as expected. Since the circuit parameters used in the experiments are not fully disclosed, it is difficult to assess their result. However, since the effect of reverselocking[I291 (by which the oscillatordesigned to be locked tends to control the frequencyof the master oscillator) is known to be prominent in the low-frequency range [130], this might be the primary cause of the discrepancy. The effect of AM noise in the injection signal is too small to account for the discrepancy. D . Lurge-Signal Injection Tamai [131]observed a hysteresisinthe lockingof an IMPATT oscillator. This is most likely the hysteresis, associated with the involvement of two types of instability, dis-

H . TRA P A TT Oscilhtors Injection locking of a CW TRAPATT oscillationwasreported by Evans [149] in connection with a TRAPATT amplifier. When the injection frequencywas slightly below the cenZ ter of the amplifier passband, and the inputpower exceeded a certain value, the output power increased rapidly, the noise decreased by about 70 dB, and the efficiency increased to about 35 percent. He identified this mode of operation as a locked oscillation. Prager et al. [I501 also reported the injection locking of their anomalous-avalanche-diode oscillator in pulsed operation. I . Self-Injection Locking A new type of oscillator circuit illustrated in Fig. 40 was

1408

PROCEEDINGS OF THE

IEEE,OCTOBER 1973

[5] P. Mastalli, S. Randi, and G. Vannucchi, A new microwave repeaterforFMradiolinks,AlbFreq.,vol.37,pp.41~25,May1968. [6] T. Isobe and M. Tokida, Noise loading performance o a phasef locked IMPATT oscillator for multichannel FM signals, Proc. ZEEE (Lett.), vol. 56, pp. 87-75, May 1968. [7] , Anew microwave amplifier for multichannel FM signals using a synchronized oscillator, ZEEE J . Solid-state Circuits, vol. SC-4, pp. 4OO-408, Dec. 1969. , Power amplification for FM and PM signals with synchro[8] nized IMPATT oscillators, ZEEE Trans. Microwave Theory Tech., VOI. MTT-18, pp. 906-911, NOV.1970. [9] K. Kohiyama and K. Monma, A new type o all solid-state 11 f GHz band F M transmitter combining a Gunn diode and a n IMPATT diode, Proc. ZEEE (Lett.), vol. 57, pp. 1232-1233, June 1969. (101 D. A. Fleri and R. J. S c, Amplifying properties of Gunn d l oc i lator in injection locked mode, Proc. (Lett.), vol. 57, pp. 1205-1206, June 1969. [ l l ] E. M d and A. BeIlardo, Thin-film injection-locked oscillators and negative-resistance amplifiers for 2-GHz radio repeater, VI. POSTSCRIPT ZEEE J . Solid-state Circuits. vol. SC-7, pp. 23-32, Feb. 1972. When one sees the sudden change of an oscillation spec1121 Y. Takayama, Power amplifier with IMPATT diodes in stable and injection-locked modes, ZEEE Trans.Miclowave T h y Tech., trum as locking takes place, one is witnessing one of the most vol. MTT-20, pp. 266-272, Apr. 1972. profound effects of nonlinearity. Such a phenomenon is totally [13] C. W. Carnahn and H. P. Kalmus, Synchronized oscillators as FM receiver limiters, Elecirmiw, vol. 17, pp. 108-112, Aug. 1944. absent in the realm of linear network theory familiar to elec(141 W. M.Hubbard et al., A solid s t a t e regenerative repeaterfor tricalengineers. Thanks to the pioneeringwork of various guided millimeter-wave communication systems, Bell Sysf. Tech. authors, however, i t can now be understood as a corollary of J., VOI. 46, pp. 1997-2018, NOV.1967. [15] J. H. Johnson, R. C. Shaw, and H. L. Stover, yde-band osdllatthe basic theorem that the total impedance times the current ing limiters at X-band and L-band frequenaes, ZEEE J . Sdidis equal to the applied voltage. Note that the only necessary State Circuits, vol. SC-3, pp. 163-165, June 1968. ingredient in this paper was this basic theorem; everything 1161 J. R. Ashley and F. M. Palka, Improvement of a microwave discriminator by an injection phase-locked osallator, ZEEE Trans. else was derived from it. Micrawaoe T h y Tech. (Corresp.), vol. MTT-18, pp. 1001-1002, The present paper is, in sense, an attemptto draw a road a Nov. 1970. [17] J. R. Woodyard, Applicantionof the autosynchronized oscillator to map in the field of injection locking. Local roads are comfrequencydemodulation, Proc. IRE, vol. 25, pp. 612-619, May pletely left out. There might be more scenic routes of which 1937. the present author is unaware. But the author has tried to [la] C.L.Ruthroff, Injection-IockeddllatorFM receiver analysis, Bell Syst. Tech. J . , vol. 47, pp. 1653-1661, Oct. 1968. indicate in a coherent manner how to reach major points of T. L. H. IV, Injection-locked avainterest on the map. The map incomplete since highways in [19] lanche&borne and C. F M Elmendorf,Proc. ZEEE (Lett.), vol. 57, is diode oscillator receiver, the field of injection locking are still under construction. In pp. 214-215, Feb. 1969. particular, the study multiple-frequency oscillationsis in its (201 E. Proni, FM demodulator employing a n injection-locked oscilof lator, Alto Freq., vol. 38, pp. 95-103, Feb. 1969. infancy. We have a reasonable idea how to suppress the bias [21] K. Murakami,A newphase-locked demodulator with injection locking, Electron. Commun. Japan,voL 52, pp. 119-120, Feb. 1969. oscillations in Gunn and IMPATT oscillators and the paramet[22] G. S. Hobson and M. Thomas,Directfrequencydemodulation ric-type oscillation in IMPATT oscillators. But very little is with frequency-locked Gunn oscillators, Electron. Lett., vol. 7, known about theinjection locking of such multiple-frequency pp. 67-68, Feb. 11, 1971. 1231 T. 1. Stoker. Nonlinear Vibrations. New York: Wilev. 1950. oscillations, much less about its potential applications, any. if i24j J. Cudningham, Introduction to Nonlinear A G y s i S . New York: McGraw-Hill, 1958. ACKNOWLEDGMENT [25] N. Minorsky, Nonlinear Oscillations. Princeton,. N. .J.: Van NosThe author wishes to thank Prof. A. E. Siegman of Stan- [26] trand, 1962.-Nonlinear Osciudiorr in Phys,sicd Systems. New York: C. Ha-, ford University and the Editorial Board of the PROCEEDINGS McGraw-Hill, 1964. I271 A. BlaquiCre, NonlinearSystcm Analysis. New York: Academic OF THE I E E E who gave him an opportunity to write on this Fress, 1966. fascinating field. The author also thanks M. E. Hineswho [28] B. Van der Pol, Forced oscillations in a circuit with nonlinear resistance, Phil. Mag., vol. 3, pp. 65-80, Jan. 1927. made his paper on the parametric oscillation of IMPATT diodes [29] , The nonlinear theory of electricosdllations, Proc. ZRE, available to the present author prior to its publication. V O ~ .22, pp. 1051-1085, Sept. 1934. Many colleagues at the Bell Laboratories contributed to [30] N. Krylov and N. Bogoliubov, Zntrodutth to Nonlinear Mechanics. Princeton, N. J.: Princeton Univ. Press, 1943. the drawingof the road map presented here. Among them are: f [31] R. Adler, A study o locking phenomena in oscillators,Proc. IRE, J. G. Josenhans, D. N. Kenyon, W. 0. Schlosser, C. A. vol. 34, pp. 351-357, June 1946; also this issue, pp. 1380-1385. Brackett, and F. M. Magalhaes. P. W. Dorman supplied the (321 J. Loeb, PhCnomtnes hereditaires d a m lea servomecanismee, un critere general de stabilite,Ann. Tclccommur., vol. 6, pp. 346-352, oscilloscope pictures used in this paper. Dec. 1951. The author also thanks F. H. Blecher-and K. M. Poole for [33] A. Blaquiere, Les oecillateurs non le et le diagramme de i s d Nyquist, J . Phys. Radium, voL 13, pp. 527-542, Nov. 1952. their encouragement and support during his writing. [34] , Adaptation gCntrak de la mtthode du diagramme de Nyquist dans le domaine non fin&, J . Phys. Radium, voL 13, pp. REFERENCES 636-644, Dec 1952. [35] J. C. Slater, Microwasc Electrcnrics. New York: Van Nostrand, [l] R. D. Huntoon and A. Webs, Synchronization of oscillators, 1950, pp. 205-210. Roc. IRE, vol. 35, pp. 1415-1423, Dec 1947. [36] K Kurokawa, Stability of injection-locked oecillators, Proc. . [2] B. S. So and P. A. Goud, Injectionphasdocking properties of biasZEEE (Lett.),V O ~ .60, pp. 907-908, July 1972. modulated IMPATT diode oscillators, in 8th ZW. Conf. on 1371 M. L. Cartwright, Forced oscillations in nearly sinusoidal ayeMicromaw and Optical Generation and AmpliJEcdkm MOGA 70, tems, J . Z . Ekc. Eng., voL 95, pt. 111, pp. 88-96, jan-Mar. w Sept. 1970. 1948. [3] B. Martin and G. S. Hobson, Angle modulation of frequency[38] G. D. Shemanaev and E. N. Ivanov, Locking a two-circuit d l locked Gunn oscillators, Elcdron. Lctt., vol. 7, pp. 399-401, July 15, lator,Radio En#. Electron., vol. 5 , pp. 45-46, Sept. 1960. 1971. [4] S.V. Judd, Simple repeater for frequency modulated transferreed[39] M. E. H i m , Negative-resistance diode power amplification, electron oedllatom, Electron. LC#., vol. 4, p. 33, Jan. 1968. ZEEE Trans. Elcck.orr Dcoicu, voL ED-17, pp. 1-8, Jan 1970.

studied by Ohta et a2. [151], [152]. In this scheme, a portion of the output signal is fed back into the oscillator through a circulator and a directional coupler. This scheme was called self-injection locking. In spite of the name, this is a free-running oscillator with a complex impedance locus. There is no independentvoltagesource whichrepresents the injection signal in the equivalent circuit. Because of the two separate signal paths,forward and backward, this scheme gives the largerdegree of freedom in selecting the components for a desiredimpedancelocus.However,thisscheme will suffer from the complications discussed in Section 11-F and it remains to be seen whether or not this type circuit becomes of of practical value.

W :

Roc.

KUROKAWA: INJEmION LOCKING OF MICROWAVE OSCILLATORS

1409
by injection locking in oscillators, Trans.Inst. Elec. Com. Eng. Japan, vol. 52-B, pp. 784-785, D e c 1969. J. Jwnhans, Noise spectra of Read diode and Gunn oscillators, PTOC. ZEEE (Lett.), VOL 54, PP. 1478-1479, oct. 1966. K. Kurokawa, Noise in synchronized oscillators, ZEEE Trans. Microwave T k o r y Tech., vol. MTT-16, pp. 234-240, Apr. 1968. R. F. Scherer, Investigations of the noise spectra of avalanche oscillators, ZEEE Trans. Microwave Theory Tech., vol. MTT-16, pp. 781-788, Sept. 1968. J. J. Goedbloed, On the upmnverted noise of IMPATT diode oscillators, presented a t 8th Int. Conf. on Microwave and Optical Generation and Amplification, MOGA 70, Sept. 1970. M. T. Vlaardingerbroek and J. J. Goedbloed, On the theory 0; noise and injection phase locking of IMPATT diode oscillators, Philips Res. Rep., vol. 25, pp. 452-471, Dec. 1970. K. Mouthaan and H. P. M. Rijpert, Nonlinearity and noise in the avalanche transit-time oscillator, Philips Res. Rep., vol. 26, pp. 391-413, Oct. 1971. M. T. Vlaardingerbroek, Theory of oscillator noise, Eleciron. Lett., vol. 7, pp. 648-650, Oct. 21, 1971. M. E. Hines, J. C. R. Collinet, and J. G. Ondria, FM noise suppression of a n injection phase-locked oadllator, ZEEE Trans. Microwave Theory Tech., vol. MTT-16, pp. 738-742, Sept. 1968. T. Osobe and Tokida, Effects of phase locking on modulation characteristics, Proc.ZEEE (Lett.), vol. 5 5 , pp. 453-454, Mar. 1967, - Noise reduction of oscillators by phase locking, Electron. , Commun. Japan, vol. 50, pp. 52-59, Nov. 1967. T. Sugiura, On the reduction of oscillator noise by injection locking, Electron. Commun. Japan, vol. 51-B, pp. 115-116, Aug. 1968. G. H. B. Hansson and K. I. Lundstrom, Phaselocking of negative conductance oscillators, in 1971 European Microwave Conf. Proc., A 6/4: 1 4 , Stockholm, Sweden, 1971. Also, Stability criteria for phase-locked oscillators, ZEEETrans.Microwaw Theory Tech., V O ~MTT-20, pp. 641-645, Oct. 1972. I. Tatauguchi. N.R. Dietrich, and C. B. Swan, Power-noise characterization of phase-locked IMPATT oscillators, ZEEE J . SolidSfate Circuits, vol. SC-7, pp. 2-10, Feb. 1972. K. Kurokawa, Some basic characteristics of broadband negative resistance oscillator circuits, Bell Syst. Tech. J . , vol. 48, pp. 19371955, JUly-AUg. 1969. Y. Okabe and S. Okamura, Analysis of the stability and noise of oscillators in free, synchronized and parallel running modes, Electron. Commun. Japan, vol. 52-B, pp. 102-110, Dec. 1969. M. Ohtomo, Experimental evaluation of noise parameters in Gunn and avalanche oscillators, ZEEE Trans. Microwave Thwry Tech., VOI. MTT-PO, pp. 425437, July 1972. H. Fukui, Frequencylocking and modulation of microwave silicon avalanche diode oscillators, Proc. ZEEE (Lett.), vol. 54, pp. 14751477, OcL 1966. S. Mizushina, 2 oscillators combined with 3-dB directional couplers for output power summing, Proc. ZEEE (Lett.), vol. 55, pp. 2166-2167, Dec. 1967. J. R.Nevarez and G. J. Herskowitz, Output power and loss analyeis of 2 injection-locked oscillators combined through a n ideal and symmetric hybrid combiner, ZEEE Trans. Microwave Theory Tech., VOI. MTT-17, pp. 2-10, Jan. 1969. W. 0. Schloeser and A. L. Stillwell, A traveling-wave approach to the high-power solid-state millator, Proc. ZEEE (Lett.), vol. 56, p. 1588, Sept. 1968. R. J. Soca and R. I. Harrison, A multidiode avalanche oscillator for increased CW microwave power output, Proc. ZEEE (Lett.), vol. 54, pp. 10061007, July 1966. V. P. Rubanik, Mutual synchronization of selfsscillating systems, Radio Eng. Elec&ron.Phys., vol. 7, pp. 1584-1592, Oct. 1962. T. Suezaki and S. Mori, Mutual synchronization o two oscilf lators, Elechon. Commun. Japan, vol. 48, pp. 56-65, Sept. 1965. W. 0.Schlceser, Noise inmutually synchronized oscillators, ZEEE Trans. Microwave Theory Tech., vol. MTT-16, pp, 732-737, Sept. 1968. E. F. Scherer and M. J. Barrett, A broadband multistage avalanche amplifier at X-band, in 1969 ZSSCC Dig. Tech. Papers, p. 82, Feb. 1969. J. J. Sie and W. J. Crowe, A one watt CW X-band avalanche diode source or power amplifier, in 1969 P-G MTT Symp.Rec., pp. 266272, May 1969. I. Tatsuguchi, frequency-modulated A phase-locked IMPATT power combiner, in 1970 ZSSCC D ~ KTech. Papers, pp. 18-19, . Feb. 1970. C. T. Rucker,A multiple-diode high-average-power avalanchediode oscillator, ZEEE Trans. Microwooc T h y Tech. (Corresp.), V O ~ .MTT-17, pp. 11561158, Dec 1969. K. Kurokawa and F.,M. Magalhaes, An X-band 10-watt multipleIMPATTosdllator, Proc. ZEEE (Lett.), vol. 59, pp. 102-103, Jan. 1971. K. Kurokawa, An analysia of Ruckers multideviae symmetrical

1401 -,

Special problems in IMPATT diode power amplifiers, in

1972 ZSSCC Dig. Tech. Papers, pp. 34-35.

1411 R. V. Khokhlov, On the theory of locking effect for the small amplitude of external force, Dokl. Akad. Nauk SSSR, 97, pp. 411vol. 414, Mar. 1954. [42] R. R. Lisitsian, Oscillator locking to the frequency of a n external excitation, Radio Eng. Electron., vol. 2, pp. 68-88, Apr. 1957. [43] R. C.Mackey, Injection locking of klystronoscillators, I R E Trans. Microwave Theory Tech., vol. MTT-IO, pp. 228-235, July 1962. [44] T. M. While and W. B. Jones, Frequency transients in synchronized oscillators, ZEEE Trans. Circuit T h y (Corresp.), vol. CT-11, pp. 279-281, June 1964. [45] M. J. E. Golay, Normalized equations of the regenerative oscillator -Noise, phase-locking, and pulling, Proc. ZEEE, vol. 52, pp. 13111330, Nov. 1964. [46] L.J. Paaorek, Injection locking of oscillators, Proc. ZEEE, vol. 53, pp. 1723-17y27, Nov. 1965. 1471 Y. Takayama, Dynamic behavior of nonlinear power amplifiers in stable andinjection-locked modes, ZEEE Trans.Microwave T h y Tech., V O ~ . MTT-20, pp. 591-595, Sept. 1972. [48] T. J. Buckman, The frequency spectrum of a pulled oscillator, Proc. IRE, vol. 40, pp. 958-961, Aug. 1952. [49] H. L.Stover, Theoretical explanation for the output spectra of unlocked driven oscillators, Proc. ZEEE (Lett.), vol. 54, pp. 310311, Feb. 196% [SO] M.Armand, On the output spectrum of unlocked driven oscillators, Proc. ZEEE (Lett.),zol. 57, pp. 7 9 S 7 9 9 , May 1969. [Sl] J. Dekleva and I. Zanchi, Improvedcalculation for the output spectra of unlocked driven oscillators, Proc. ZEEE (Lett.), vol. 6 0 , pp. 135-136, Jan. 1972. (521 K.Kurokawa, A n Zntroduction t tk Theory of Microwave Circuits. o New York: Academic Press, 1969, pp. 380-397. 1531 W.A. Edson, Noise in oscillators, Proc. IRE, vol. 48, pp. 14541466, Aug. 1960. 1541 J. P. Quine, Injection phase-lyking characteristics of LSA-mode transferred-electron oscillators, Proc. ZEEE (Lett.), vol. 57, pp. 715-717, Apr. 1969. [55] B. C. So, Injection phase-locking properties of microwave IMPATT diode and Gunn diode oscillators with system applications, Ph.D. dissertation, University of Alberta, Edmonton, Alta., Canada, 1971. [56] P. Jochen, Equivalent circuitfor injection-locked negative resistance oscillators, Electron. Lctt., vol. 6, pp. 61-62, Feb. 5, 1970. [57] T. L.Osborne, Amplitude behavior of injection-locked oscillators, ZEEE Trans. Microwave T h y Tech., vol. MTT-18, pp. 897-905, Nov. 1970. [58] H. J. Thaler, G. Ulrich, and G;, Weidmann, Noise in IMPATT diode amplifiers and oscillators, ZEEETrans. Microwave Theory Tech., V O ~ .MTT-19, pp. 692-705, Aug. 1971. 1591 E. S. Voronin, Synchronization of self-oscillator with modulated external forces. Radiokkrrika, vol. 14, pp. 48-56, Feb. 1959. [60] H. L.Stover andR. C. Shaw, Injection-locked oscillators as amplifiers for angle-modulated signals, in 1966 G-MTT Symp. Dig., pp. 60-65. (611 I. Berstein, On fluctuations in theneighborhood of periodic motion of a n auto-oscillating system, Dokl. Akad. Nauk SSSR, vol. 20, pp. 11-16, Jan. 1938. (621 S M. R y t w , Fluctuations in oscillating systems of the Thomson . type. I, Sw. Phys.-JETP, vol. 2. DR. 217-224, Mar. 1956. , Fluctuations in oscillatingsystems of .the Thomson type, (631 11,SW. PhyS.-JETP, V O ~ .2, PP. 225-235, MU. 1956. 1641 J. A. Mullen, Background noise in nonlinear oscillators, Proc. IRE, V O ~ .48, pp. 1467-1473, Aug. 1960. [65] A. N. Malakhov, On the amplitude and phase fluctuations of a n auto-oscillator, Rdiophysics, vol. 6, pp. 495-500, Mar. 1963. , Synchronization of a generator by the random quasi-sinus(661 oidal signal, Radiophysics, vol. 6, pp. 501-512, Mar. 1963. , The solution of nonlinear equations f o r the amplitude and [67] phase fluctuations of a n auto-oscillator, Radwphysics, vol. 7, pp. 710-721, Apr. 1964. , Effect of natural noise in auto-oscillator upon their mutual [68] synchronization, Radiophysics, vol. 8, pp. 1160-1170, June 1965. 1691 , Spectrum line form of a n oscillator with smallamplitude and frequency fluctuations, Radiophysics, voL 10, pp. 886-888, 1967. [70] E. Hafner, The &ecta of noise in oscillators, Proc. ZEEE, vol. 54, pp. 179-198, Feb. 1966. [71] B. N. Biswas, RMS frequency error of injection-synchronized millators, ZEEETrans.Circuit Thcory ( C o r n . ) , vol. CT-17, pp. 637-639, Nov. 1970. [I21 T. Saito, T. Takagi, and K. Mano, Response of a pull-in oscillator for random noise coinadent with synchronizing signal, Electron. Commun. Japan, vol. 5 1 - 4 pp. 20-28, Jan 1968. I731 , Fundamental consideration of the stabilization o an oscilf lator by injection locking, Electron. Commun. Japan, vol. 52, pp. 73-80. Am. 1969.

v.

1410
oscillator, ZEEE Trans. Microwave T h y Tech. (Corresp.), vol. MTT-18, pp. 967-969, NOV.1970. [lo51 , The single-cavity multipledevice oscillator, ZEEE Trans. Minowave Theory Tech., vol. MTT-19, pp. 793-801, Ob. 1971. [loa] C.B. Swan, IMPATT oscillator performance improvement with second harmonic tuning, Proc. ZEEE (Lett.), vol. 56, pp. 16161617, Sept. 1968. [lo71 C.A. Brackett,Characterization of second-harmonic &e& in IMPATT diodes, Bell Syst. Tech. J . , vol. 49, pp. 1777-1810, Oct. 1970. [lo81 W. C. Tmi and F. J. Roeenbaum, Bias circuit oscillations in Gunn devices, ZEEE Trans. Electron Dmicm, vol. ED-16, pp. 196-202, Feb. 1969. [lo91 C.A. Brackett, The elimination of tuning-induced burnout and bias circuit oscillations in IMPATT oscillators, to be published in Bell Syst. Tech. J . [110] M.E. Hinea, Large-signal noise, frequencyconversion, and parametric instabilities in IMPATT diode networks, Proc. ZEEE, vol. 60, pp. 1534-1548, Dec 1972. [ l l l ] J. R. Pierce, Oscillator behavior, June 19, 1943, unpublished. [112] W. A. Edeon, Vacuum Tube Oscillators. New York: Wiley, 1953, pp. 438-440. [113] J. C. West, Analytical Techniqws for Nonlineur Corrirol Syskms. London, England: English Univ. Press, 1960. [114] A. Gelb and W. E.VanderVelde, Mulliple-Znput Describing Functions and Nonlinear System Design. New York: McGraw-Hill, 1968. [115] 2 Bone? Stability of f w d oscillations in nonlinear feedback . systems, IRE Trans. Automat. COW., vol. PGAC-6, pp. 109-111, D e c 1958. , Relative stability of oscillations in nonlinear control s p [116] tern, in Proc. ZFAC (Basel, Switzerland, Aug. 1963), pp. 214114. [ i l l ] L. Gustafeaon, G. H. B. Bansson, and K. I. LundstrBm, On the use of describing functions in the study of nonlinear active microwave circuits, IEEE Trans. Microwave T h y Tech., voL MTT20, pp. 402409, June 1972. [I181 B. W. Hakki, J. P. Beccone, and S. E. Plauaki, Phaselocked GaAs CW microwave oscillators, ZEEE Trans. Electron Devices (Corresp.), vol. ED-13, pp. 197-199, Jan. 1966. [119] R. C. Shaw and H. L. Stover, Phase-locked avalanche diode oscillators, Proc. ZEEE (Lett.), vol. 54, pp. 710-711, Apr. 1966. [120] T. A. Midford and S. Wanuga, Phase locking of a silicon avalanche transit time oscillator, Proc. ZEEE (Lett.), vol. 54, pp. 993-994, July 1966. [121] V. M. Vald-Perlov, A. V. Krasilov, and A. S. Tager, The avalanching transit time diode-A new semiconductor microwave device, Radio Eng. Electron. Phys., vol. 11, pp. 1764-1777, Nov. 1966. [l22] S. V. Judd and S. J. Hewitt, Phase locking of Gunndect osallator, Electron. Lctt., vol. 3, pp. 107-108, Mar. 1967. (1231 K. Sugawara, A wide-band 4-GHz Esaki-diode injection-locked oscillator, Proc. ZEEE (Lett.), vol. 57, pp. 215-216, Feb. 1969. f [124] N. D . Kenyon, A lumped-circuit study o basic oscillator behavior, Bell Syst. Tech. J . , vol. 49, pp. 255-272, Feb. 1970. [125] J. R. Ashley and F. M. Palka, Noise properties and stanbilization of Gunn and avalanche diode oscillators and amplifiers, in 1970 G M T T SF#. Dig., pp. 161-164. [126] , Measured FM noise reduction by injection phase locking, Proc. ZEEE (Lett.), vol. 58, pp. 155-157, Jan. 1970. [127] -, Reduction of FM noise in microwave diode oscillators by cavity and injection stabilization, in 1971 I E E E G M T T Symp. Dig., pp. 94-95. [128] T. Sugiura and S. Sugimoto, FM noise reduction of G u n n d e c t oscillators by injection locking, Proc. ZEEE (Lett.), vol. 57, pp. 77-78, Jan. 1969. [129] B. S.Perlman and T. E. Walsh, Criterion for nonreciprocal injec-

PROCEEDINGS OF THE IEEE, OCTOBER

1973

tion locking of bilateral microwave oscillators, ZEEE Trans. Microwave T h y Tech. (Corresp.), vol. MTT-18, pp. 507-508, Aug. 1970. [130] I. Back, FM noise in an injection-locked oscillator when reverse locking exists, Electron. Lctt., vol. 7, pp. 346-348, June 17, 1971. [I311 T. Tamai, Phase-lccked oscillations of silicon avalanche diodes, Japan. J . Appl. Phys.. vol. 5, p. 1115, Nov. 1966. [132] D. D . Khandelwal, Injection signal characteristics of high-power avalanche diode oscillators, Proc. ZEEE (Lett.), vol. 58, pp. 928930, June 1970. [133] P. L. Fleming, Synchronization of microwave d l a t i o n a in GaAs, Proc. IEEE (Lett.), vol. 53, pp. 1665-1666, Oct. 1965. 11341 H. B. Munson, Jr., Frequency control of pulsed GaAe Gunn &t e c diodes by injection locking, Proc. ZEEE, vol. 53, pp. 1781-1782, Nov. 1965. [135] H. Gelbwachs and S. Mao, Phaselocking on pulsed Gunn d l lators, R o c . ZEEE (Lett.), vol. 54, pp. 1591-1592, Nov. 1966. [I361 H. Pollmann and B. G. Bosch, Injection priming of pulsed G u m osdllators, IEEE Trans. E[cdron Dcdccs (Corresp.), vol. ED-14, pp. 609-610, Sept. 1967. [137] M. I. Grace, Injection locking of pulsed avalanche diode d l lators, Proc. IEEE (Lett.), voL 55, pp. 713-714, May 1967. (1381 F. R. Domer and R;H. Kyle, Phase locking an IMPATT device in the puleed mode, Proc. ZEEE (Lett.), vol. 55, pp. 1753-1754, Oct. 1967. [139] F. M. Magalhaea and W. 0. Schlosser, Synchronization of pulsed oecillators, Proc. IEEE (Lett.), voL 56, pp. 766-767, Apr. 1968. [140] R. V. Khoklov, A method of analpis in the theory of sinusoidal self-oscillations,IRE Trans. Circuit Thwry, vol. CT-7, pp. 398413, Dec 1960. [141] M. Sugiyama, Experiments on a harmonic-locked oecillator, Proc. ZEEE (Lett.), vol. 56, p. 780, Apr. 1968. [142] J. A. Mashock and T. K. Oshii, An intereating pulling deet in a tunnel diode oscillator, ZEEETrans. Microwave Theory Tech., V O ~ .MTT-17, pp. 347-348, J u 1969. ~ I1431 A. S. Tager and A. M. Tsebiyev, Synchronization of microwave avalannche-transit time diode oacilbtor by a n external harmonic signal, RadioEng. Electron. Phys., vol. 14, pp. 1110-11 16, July 1969. (1441 H. G. Oltman and C. H. Nonnemaker, Subharmonically injection phaselocked Gunn oscillator experiments, ZEEE Trans. Microwave Thwry Tech. (Corresp.), vol. MTT-17, pp. 728-729, Sept. 1969. [145] G. H. Ghien and G. C. Dalman, Snbharmonically injection phaselocked IMPATT oscillator experiment, Electron. Lett., vol. 6, pp. 240-241, Apr. 16, 1970. (1461 I. Schmideg, Harmonic synchronization of nonlinear oscillators, Proc. IEEE (Lett.), vol. 59, pp. 1250-1251, Aug. 1971. [147] R. Perichon, Frequency-modulation noise of subharmonically injection phase-locked IMPATT oscillator, IEEE Trans. Microwave Theory Tech. (Corrr~p.),V O ~ .MTT-18, PP. 988-989, NOV.1970. 11481 Y. Fukatsu and H. Kato. Freauencv conversion with gain through ~. sideband locking of an IMPA-TT diode oscillation, h o c . ZEEE (Lett.), VOL 57, pp. 342-343, Mar. 1969. [149] W. J. Evans, CW TRAPATTamplification, ZEEE Trans. Minomaw Theory Tech. (Corresp.), vol. MTT-18, pp. 986-988, Nov. 1970. [150] H. J. Prager, K. K. N. Chang, and S. Wn&brod, Power amplification withanomalousavalanche diodes, ZEEE Trans. Microwave Theory Tech., vol. MTT-18, pp. 956-963, Nov. 1970. [151] T. Ohta andK. Murakami, Reducing negative resistance oscillator noise by self-injection, Electron. Ccmmun. Japan, vol. 51-B, pp. 80-81, O a . 1968. (1521 T. Ota and M. Hata, Noise reduction of oscillator by injection locking, J . Znst. Elec. Chem. Eng. Japan, vol. SJ-B, pp. 487-494, Sept. 1970.

You might also like