Professional Documents
Culture Documents
On The Sliding-Window Representation in Digital Signal Processing
On The Sliding-Window Representation in Digital Signal Processing
On The Sliding-Window Representation in Digital Signal Processing
SIGNAL
AND PROCESSING, VOL. ASSP-33, Nb. 4, AUGUST 1985
Abstract-The short-time Fourier transform of a discrete-time sig- can be expressed directly in terms of the values of this
nal, which is the Fourier transform of a “windowed” version of the
sampled sliding-window spectrum. This will lead us in a
signal, is interpreted as a sliding-window spectrum. This sliding-win-
dow spectrum is a function of two variables: a discretetime index, which natural way to Gabor’s representation [3] of a signal as a
represents the position of the window, and a continuous frequency var- superposition of properly shifted and modulated versions
iable. It is shown that the signal can be reconstructed fromthe sampled of a function that is related tothe window. We show a way
sliding-window spectrum, i.e., from the values at the points of a certain to determine thisfunction from the knowledge of the win-
time-frequency lattice. This sampling lattice is rectangular, and the
dow, and we elucidate this with some simple examples of
rectangular cells occupy an area of 27r in the time-frequency domain.
It is shown that an elegant way to represent the signal directly in terms
window functions.
of the sample values of the sliding-window spectrum, is in the form of
Gabor’s signal representation. Therefore, a reciprocal window is in- SLIDING-WINDOW REPRESENTATION OF
troduced, and it is shown how the window and the reciprocal window DISCRETE-TIME SIGNALS
are related. Gabor’s signal representation then expands the signal in
terms ofproperly shifted andmodulated versions of the reciprocal win- Let x ( n ) (n = . - , - 1, 0, 1, - *
+ denote a one-di-
a)
dow, and the expansion coefficients are just the values of the sampled mensional discrete-time signal and let w ( n ) represent a
sliding-window spectrum. window sequence; the signal and the window may take
complex values and they need not have a finite extent. We
multiply the signal by a shifted and complex conjugated
INTRODUCTION
version of the window and take the Fourier transform of
the product, thus constructing the function [cf. [l], (6.1)]
S HORT-TIME Fourier analysis [ l ] of discrete-time sig-
nals is of considerable interest in a number of signal-
processing applications. In order to study spectral prop- f ( ~ ! n)
, = C
03
yields the inversion formula values at the points of a certain lattice inthe fl-n domain.
This will be shown in the next section.
SIGNAL RECONSTRUCTION FROM ITS SAMPLED
SLLDING-WINDOWSPECTRUM
in which j2= dQ * represents integration over one period
2s;of course, the rather mild requirement that w ( 0 ) be Let N be apositive integer, let the sliding-window spec-
nonzero should be satisfied. There exists another wayof trumf(Q, n) be known at the points { Q = k(27r/N),n =
reconstructingthe signal from its sliding-window spec- m N ) ( k , m = -
* , - 1 , 0, 1 , * * and let the values at e),
trum, viz. by means of the,inversion formula [cf. [ 5 ] ,(2) these points be denoted by&,; hence,
and [ 6 ] , (27.12.1.5)l
03
x(m) = m c L2~s
2n=-m 2n
dQ
m
n = --m
Iw(n>I
n=-m
x(m) = m
1
Iw(n)l
2n=-co
m
'i
2~ 2a
dQ
spectrum (cf. [4, Section 4.21).
We first define the functionf(n, w ) by a Fourier series
with coefficients fkm,
n=-m
(4) m
- p(fl, n) w ( m - n) exp u f l m ] . f(n,w) = m=-m k=(N) f k m exp [-j(wmN - k -N n
The representation ( 3 ) , i.e., choosing the kernel p ( Q , n)
in (4) equal to the sliding-window spectrumf(Q, n), is the (8)
best possible one in the sense that for this choice the L2- where
represents
summation
over
one
period N.
norm of p ( Q , n) takes its minimum value. To see this, we Note that the functionf(n, w ) is periodic in n and w , with
multiply both sides of ( 3 ) and (4)by x*(m), sum up over periods Nand 27rlN, respectively. The'inverse relationship
all m, and conclude from the equivalence of the right-hand
has the form
sides of the resulting equations that f ( Q , n) and p ( Q , n) -
f ( Q , n) are orthogonal in the sense
5
n=-a
-!-
2~ 2a
1dfl { p ( Q , n) - f(Q, n ) )f*(Q, n) = 0;
wmN - k
N
hence, the relationship Furthermore, we define the function Z(n, w ) by
00
to an interval of length N , with m taking on all integer by expressing f(n, w ) in terms of f k m via (8), expressing
values. g ( n , w ) in terms of g(n) via (lo), and using the inversion
With the help of the functions f(n, w ) , X(n, w ) and a relationship (12). Notethestrongresemblancebetween
similar function @(a,w ) associated with the window w ( n ) , (18) and the inversion formula (3). Equation (18), which
(7) can be rewritten as expresses the signal as a combination of properly shifted
and modulated versions of the reciprocal window,is in the
f(n, w ) = NX(n, w) @'"(a,w). (13)
form of Gabor's signal representation [cf. [3], (1.29)].
In fact, we have now solved the problem of reconstructing Gabor's signal representation thus provides a way to ex-
the signal from its sampled sliding-window spectrum: press the signal directly in terms of the sample values of
1) from the samplevalues f k m we determine the function the sliding-window spectrum.
f ( n , w ) via (8); Equation (16) can be transformed into
2) from the window w (n) we derive the associated func- m
allowed, the function X(n, w ) can be found with the help exp [ - j k 2T
z n ] = 1for k=m=O
of (13); and (19)
4) finally, the signal follows from X(n, w ) by means of 0 elsewhere.
the inversion formula (12). From (19) we conclude that the discrete setof shifted and
A simpler reconstruction method will be derived in the modulated versions of the window, w(n - mN) exp
next section. [jk(2n/N)n], and the corresponding set of versions of the
Problems may arise in the case that @ ( n ,w ) has zeros. reciprocal window, g(n - m N ) exp [ j k ( 2 a / N )n ] , are, in
In that case homogeneous solutions h"(n,w) may occur, for a certain sense, biorthonormal.
which the relation Gabor's signal representation may be nonunique in the
Nh"(n, w ) @"(n,w ) = 0 (14) case that g ( n , w ) has zeros. In that case functions Z(n, w )
may occur, for which the relation
holds. Equation (14),whichis similar to (13) with f(n,
w ) = 0, can be transformed into the relation 2(n, w ) g(n, w ) = 0 (20)
holds. Equation (20), which is similar to (17) with T(n,
w ) = 0, can be transformed into the relation
(dl
duces to a simple correlation, Fig. 1. Sketches of (a) a three-point, symmetrical window w ( n ) , and its
m corresponding reciprocal window g(n) in the case of (b) maximum over-
lap, (c) minimum overlap, and (d) partial overlap.
hm= n = -a
x(n) w*(n - m>, (26)
and so do (15) and (19); note that, moreover, the coeffi- It will be clear that inside its finite extent, .the window
cients fom become real when the signal x(n) and the win- w(n) should take nonzero values. Note that if in the min-
dow w (n) are real. Equation (18), on the other hand, re- imum-overlap case the window is uniform inside its finite
duces to a simple convolution, extent, and if the signal x(n) vanishes outside the extent
m of the window, then (7) tells us that the array of coeffi-
x(n> = m = - m f~mg(n- m), (27) cients fko is proportional to the discrete Fourier transform
of the signal, as can be expected.
and so does (21). Furthermore, (8) and (9) then constitute
EXAMPLES
a normal Fourier transform pair, and so do (10) and (12).
Note thatif in the maximum-overlapcase the window w (n) To elucidate the conceptsof this paper, we consider two
vanishes for n # 0, then (7) or (26) tell us that the array simple examples of window sequences, and determine the
of coefficientsfom is proportional to thesignal x@), as can corresponding reciprocal window sequences for different
be expected. values of the ‘shifting distanceN . Our first example is the
Minimum overlap occurs when the window has afinite three-point, symmetrical window [see Fig. ](a)]
extent and the shifting distance N is chosen equal to this
finite extent. In that case the formulasof the previous two
sections again simplify drastically, andthe relationship
between the window w (n) and the reciprocal window g(n)
takes the simple form
If
elsewhere;
inside the extent of w(n) (29)
note that for a = 0.16, we are dealing with a three-point
lo outside the extent of w(n). Hamming window. For the maximum-civerlap case ( N =
l ) , we find
872 IEEETRANSACTIONS ON ACOUSTICS,SPEECH, AND SIGNALPROCESSING,VOL. ASSP-33, NO. 4, AUGUST 1985
for m = 0
- exp
[-an]
1;
for - ( N - 1) In 5 0
for m # 0
\O elsewhere. (42)
We use this example to show the possible nonuniqueness
of Gabor’s signal representation. In the limiting case CY =
0, the function g(n, w ) has zeros for w = r ( 2 a / N ) ( r =
. . . , - 1, 0, 1, and anarray of coefficients Zkm
e),
2q1, 0) = Rh c
r= -CQ
6 w
( ; -)
-- PR , (37)
and yields a zero result when substitutedin Gabor’s signal
where 6( - ) represents the Dirac delta function. The ho- representation.
mogeneous solution h(n) thus takes the form
CONCLUSION
h(2m) = 0
Inthispaper wehave studied the short-time Fourier
h(2m + 1) = (-1)”h. (38) transform of a discrete-time signal,or, as we prefer to call
BASTIAANS:SLIDING-WINDOWREPRESENTATION 873
it,the sliding-window spectrum. Thissliding-window [3] D.Gabor,“Theory of communication,” J. insf. Elec. Eng., vol. 93,
part 111, pp. 429-457, 1946.
spectrum is a function of two variables: a discrete time [4] M. J. Bastiaans,“Signaldescription by means of alocalfrequency
index, which represents the position of the window, and a spectrum,” in Transformations in Optical Signal Processing, W. T.
continuous frequency variable. We have shown that the Rhodes et al. Eds. Bellingham: Society of Photo-Optical Instrumen-
tation Engineers, Proc. SPIE, vol. 373, pp. 49-62, 1981.
signalcanbereconstructedfromthesliding-window [SI C. W. Helstrom, “An expansion of asignal in Gaussianelementary
spectrum, when we know its values at the pointsof a cer- signals,” iEEE Trans. Inform. Theory, vol. IT-12, pp. 81-82, 1966.
tain time-frequency lattice. This lattice is rectangular, and [6] N. G . de Brujin, “A theory of generalized functions, with applications
to Wigner distribution andWeyl correspondence,” Nieuw Archiefvoor
the rectangular cells occOpy anarea of 27r; hence,the Wiskunde vol. 21, no. 3, pp. 205-280, 1973.
coarser the sampling in time, the finer the sampling in
frequency, and vice versa.
The most elegant form to represent the signal in terms
of the sample values of the sliding-window spectrum, is
by means of Gabor’s signal representation. We therefore Martin J. Bastiaans was born in Helmond, The
Netherlands, in 1947. HereceivedtheM.Sc.
had to introduce a reciprocalwindow, and we have shown degreeinelectricalengineeringandthe Ph.D.
how the window and the reciprocal window are related. degreeintechnicalsciencesfromEindhoven
Gabor’s signal representation then expresses the signal as University of Technology, Eindhoven, The
Netherlands, in 1969 and 1983, respectively.
a superposition of properly shifted and modulated ver- Since 1969 he has been an Assistant/Associate
sions of the reciprocal window. Professor with the Department of Electrical En-
gineering,EindhovenUniversity of Technology,
REFERENCES where he teacheselectricalnetworktheory.His
research includes a system-theoretical approachof
[ l ] L. R. Rabiner and R. W. Schafer, Digital Processing of Speech Sig- all kinds of problems that arisein Fourier optics, such as partial coherence,
nals. EnglewoodCliffs, NJ:Prentice-Hall, 1978, chap. 6. computer holography, image processing and optical computing; his current
[2] A. V. Oppenheim, “Digital processing of speech,” in Applications of interest is in describing signals by means of a local frequency spectrum.
Digital Signal Processing, A . V. Oppenheim,
.. Ed. Englewood Cliffs, Dr. Bastiaans is a member of the Institute of Electrical and Electronics
NJ: Prentke-Hall, 1978,-chap. 3. Engineers and of the Optical Society of America.