Tài liệu Phân tích tín hiệu P9 doc

A time shift of the input signal leads to a modulation of the ambiguity function with respect to the frequency shift v: This relation can easily be derived from 9.11 by exploiting the f

Signal Analysis: Wavelets, Filter Banks, Time-Frequency Transforms and

Applications Alfred Mertins

Copyright 0 1999 John Wiley & Sons Ltd Print ISBN 0-471-98626-7 Electronic ISBN 0-470-84183-4

Frequency Distributions

In Chapters 7 and 8 two time-frequency distributions were discussed: the

spectrogram and the scalogram Both distributions are the result of linear

filtering and subsequent forming of the squared magnitude In this chapter

time-frequency distributions derived in a different manner will be considered

Contrary t o spectrograms and scalograms, their resolution is not restricted

by the uncertainty principle Although these methods do not yield positive

distributions in all cases, they allow extremely good insight into signal

properties within certain applications

9.1 The Ambiguity Function

The goal of the following considerations is t o describe the relationship between

signals and their time as well as frequency-shifted versions We start by looking

at time and frequency shifts separately

Time-Shifted Signals The distance d ( z , 2,) between an energy signal z ( t )

and its time-shifted version z,(t) = z ( t + T ) is related to the autocorrelation

function T ~ ~ ( T ) Here the following holds (cf (1.38)):

d(&, .lZ = 2 1 1 4 1 2 - 2 w % T ) } , (9-1)

265

In applications in which the signal z ( t ) is transmitted and the time shift T

is to be estimated from the received signal z(t + T ) , it is important that z ( t )

and z(t + T) are as dissimilar as possible for T # 0 That is, the transmitted signal z ( t ) should have an autocorrelation function that is as Dirac-shaped as possible In the frequency domain this means that the energy density spectrum should be as constant as possible

are often produced due to the Doppler effect If one wants to estimate such frequency shifts in order to determine the velocity of a moving object, the dis- tance between a signal z ( t ) and its frequency-shifted version z v ( t ) = z ( t ) e j u t

is of crucial importance The distance is given by

where sEz((t) can be viewed as the temporal energy density.' Comparing (9.5)

with (9.3) shows a certain resemblance of the formulae for .Fz(.) and pEz(v),

'In (9.5) we have an inverse Fourier transform in which the usual prefactor 1/27r does

not occur because we integrate over t , not over W This peculiarity could be avoided if Y was replaced by -v and (9.5) was interpreted as a forward Fourier transform However, this would lead to other inconveniences in the remainder of this chapter

9.1 The Ambiguity Function 267

however, with the time frequency domains being exchanged This becomes even more obvious if pF,(u) is stated in the frequency domain:

We see that pF,(u) can be seen as the autocorrelation function of X ( w )

Time and Frequency-Shifted Signals Let us consider the signals

which are time and frequency shifted versions of one another, centered around

z ( t ) With the abbreviation

for the so-called time-frequency autocorrelation function or ambiguity func- tion’ we get

Thus, the real part of A z z ( v , r) is related to the distance between both signals

Example We consider the Gaussian signal

(9.12)

which satisfies 1 1 2 11 = 1 Using the correspondence

we obtain

Thus, the ambiguity function is a two-dimensional Gaussian function whose center is located at the origin of the r-v plane

Properties of the Ambiguity Function

1 A time shift of the input signal leads to a modulation of the ambiguity

function with respect to the frequency shift v:

This relation can easily be derived from (9.11) by exploiting the fact

that x ( w ) = e-jwtoX(w)

2 A modulation of the input signal leads to a modulation of the ambiguity function with respect to I-:

z(t) = eJwotz(t) + AEZ(V,T) = d W o T A,,(~,T) (9.16)

This is directly derived from (9.10)

where E, is the signal energy A modulation and/or time shift of the

signal z ( t ) leads to a modulation of the ambiguity function, but the principal position in the r-v plane is not affected

Radar U n c e r t a i n t y Principle The classical problem in radar is to find signals z ( t ) that allow estimation of time and frequency shifts with high precision Therefore, when designing an appropriate signal z ( t ) the expression

9.2 The Wigner Distribution 269

is considered, which contains information on the possible resolution of a given

z ( t ) in the r-v plane The ideal of having an impulse located at the origin of

the r-v plane cannot be realized since we have [l501

m

IAzz(v,r)I2 d r dv = IA,,(O,0)I2 = E: (9.18)

That is, if we achieve that IA,,(v, .)I2 takes on the form of an impulse at the origin, it necessarily has t o grow in other regions of the r-v plane because of

the limited maximal value IA,,(O, 0)12 = E: For this reason, (9.18) is also referred t o as the radar uncertainty principle

Cross Ambiguity Function Finally we want t o remark that, analogous

to the cross correlation, so-called cross ambiguity functions are defined:

W

A?/Z(V, 7) = 1, z ( t + f ) y * ( t - f ) ejyt dt

(9.19)

X ( W - ;) Y * ( w + ;) eJw7 dw

9.2 The Wigner Distribution

9.2.1 Definition and Properties

The Wigner distribution is a tool for time-frequency analysis, which has gained more and more importance owing t o many extraordinary characteristics In order t o highlight the motivation for the definition of the Wigner distribution,

we first look at the ambiguity function From A,, (v, r ) we obtain for v = 0 the temporal autocorrelation function

from which we derive the energy density spectrum by means of the Fourier transform:

(9.21)

W

- l m A , , ( O , r ) e-iwT d r

On the other hand, we get the autocorrelation function pFz (v) of the spectrum

3Wigner used W z z ( t , w ) for describing phenomena of quantum mechanics [163], Ville

introduced it for signal analysis later [156], so that one also speaks of the Wigner-Ville

distribution

41f z ( t ) was assumed to be a random process, E { C J ~ ~ ~ ( ~ , T ) } would be the autocorrelation function of the process

9.2 The Wigner Distribution 271

I Wignerdistribution I

Temporal autocorrelation Temporal autocorrelation

Figure 9.1 Relationship between ambiguity function and Wigner distribution

The function @,,(U, W) is so t o say the temporal autocorrelation function of X(W) Altogether we obtain

(9.27)

with & , ( t , ~ ) according t o (9.25) and @,,(Y,w) according t o (9.26), in full:

Figure 9.1 pictures the relationships mentioned above

We speak of W,, ( t , W) as a distribution because it is supposed to reflect the distribution of the signal energy in the time-frequency plane However, the Wigner distribution cannot be interpreted pointwise as a distribution of energy because it can also take on negative values Apart from this restriction

it has all the properties one would wish of a time-frequency distribution The most important of these properties will be briefly listed Since the proofs can

be directly inferred from equation (9.28) by exploiting the characteristics of

the Fourier transform, they are omitted

Some Properties of the Wigner Distribution:

1 The Wigner distribution of an arbitrary signal z ( t ) is always real,

This property immediately follows from (9.28)

6 If X ( w ) is non-zero only in a certain frequency region, then the Wigner distribution is also restricted t o this frequency region:

X ( w ) = 0 for W < w1 and/or W > w2

W,,(t,w) = 0 for W < w1 and/or W > w2

9.2 The Wigner Distribution 273

7 A time shift of the signal leads t o a time shift of the Wigner distribution

10 Time scaling leads t o

Signal Reconstruction By an inverse Fourier transform of W z z ( t , W ) with respect t o W we obtain the function

7- +zz(t,7-) = X * ( t - -1 4 t + 5)'

Moyal's Formula for Auto-Wigner Distributions The squared magni-

tude of the inner product of two signals z ( t ) and y(t) is given by the inner

product of their Wigner distributions [107], [H]:

9.2.2 Examples

Signals with Linear Time-Frequency Dependency The prime example

for demonstrating the excellent properties of the Wigner distribution in time- frequency analysis is the so-called chirp signal, a frequency modulated (FM) signal whose instantaneous frequency linearly changes with time:

wZE((t, W ) = 2 e-"(t - to)' e-a 1 [W - W O ] ' (9.48)

Hence the Wigner distribution of a modulated Gaussian signal is a two- dimensional Gaussian whose center is located at [to, W O ] whereas the ambiguity function is a modulated two-dimensional Gaussian signal whose center is located at the origin of the 7 - v plane (cf (9.14), (9.15) and (9.16))

Signals with Positive Wigner Distribution Only signals of the form

It can be regarded as a two-dimensional Fourier transform of the cross ambiguity function AYz(v, 7) As can easily be verified, for arbitrary signals

z ( t ) and y ( t ) we have

We now consider a signal

and the corresponding Wigner distribution

product of two cross Wigner distributions we have [l81

9.2 The Wigner Distribution 277

Figure 9.3 Wigner distribution of the sum of two sine waves

Figures 9.4 and 9.5 show examples of the Wigner distribution We see that the

interference term lies between the two signal terms, and the modulation of the interference term takes place orthogonal to the line connecting the two signal terms This is different for the ambiguity function, also shown in Figure 9.5 The center of the signal term is located at the origin, which results from the fact that the ambiguity function is a time-frequency autocorrelation function The interference terms concentrate around

9.2 The Wigner Distribution 279

The multiplication of $==(t, T) and $ h h ( t , r ) with respect t o r can be replaced

by a convolution in the frequency domain:

l

271

WE ( t , W ) = - W,, ( t , W ) : W h h ( t , W )

That is, a multiplication in the time domain is equivalent t o a convolution of

the Wigner distributions W,,(t,w) and W h h ( t , W ) with respect t o W

Convolution in the Time Domain Convolving z ( t ) and h ( t ) , or equiv- alently, multiplying X ( W ) and H ( w ) , leads t o a convolution of the Wigner

distributions W,,(t,w) and W h h ( t , W ) with respect t o t For

(9.67)

= I W z z ( t ’ , ~ ) W’h(t - t ’ , ~ ) dt’

calculating the Wigner distribution of an arbitrary signal x ( t ) is that (9.28)

can only be evaluated for a time-limited z ( t ) Therefore, the concept of windowing is introduced For this, one usually does not apply a single window

h ( t ) t o z ( t ) , as in (9.65), but one centers h ( t ) around the respective time of

Using the notation

9.3 General Time-Frequency Distributions

The previous section showed that the Wigner distribution is a perfect time- frequency analysis instrument as long as there is a linear relationship between instantaneous frequency and time For general signals, the Wigner distribution takes on negative values as well and cannot be interpreted as a “true” density function A remedy is the introduction of additional two-dimensional smooth-

ing kernels, which guarantee for instance that the time-frequency distribution

is positive for all signals Unfortunately, depending on the smoothing kernel, other desired properties may get lost To illustrate this, we will consider several shift-invariant and affine-invariant time-frequency distributions

9.3 General Time-frequency Distributions 281

9.3.1 Shift-Invariant Time-Frequency Distributions

Cohen introduced a general class of time-frequency distributions of the form

By choosing g ( v , T ) all possible shift-invariant time-frequency distributions can be generated Depending on the application, one can choose a kernel that yields the required properties

If we carry out the integration over u in (9.71), we get

T,,(t,w) = - ss g ( u , r ) AZZ(v,r) eCjyt eCjWT du d r (9.73)

271 This means that the time-frequency distributions of Cohen's class are com- puted as two-dimensional Fourier transforms of two-dimensionally windowed ambiguity functions From (9.73) we derive the Wigner distribution for

g ( v , r ) = I For g ( v , r ) = h ( r ) we obtain the pseudo-Wigner distribution

The product

is known as the generalized ambiguity function

Multiplying A z z ( v , r ) with g ( v , r ) in (9.73) can also be expressed as the convolution of W,, ( t , W ) with the Fourier transform of the kernel:

with

G ( t , W ) = - 1 // g(v, r ) ,-jut e-jWT dv d r (9.76)

2T

That is, all time-frequency distributions of Cohen’s class can be computed

by means of a convolution of the Wigner distribution with a two-dimensional impulse response G ( t , W )

In general the purpose of the kernel g(v, T ) is t o suppress the interference terms of the ambiguity function which are located f a r from the origin of the

T-Y plane (see Figure 9.5); this again leads t o reduced interference terms

in the time-frequency distribution T,,(t,w) Equation (9.75) shows that the

reduction of the interference terms involves “smoothing” and thus results in

a reduction of time-frequency resolution

Depending on the type of kernel, some of the desired properties of the time-frequency distribution are preserved while others get lost For example,

if one wants t o preserve the characteristic

(9.77)

the kernel must satisfy the condition

We realize this by substituting (9.73) into (9.77) and integrating over d w , d r ,

du Correspondingly, the kernel must satisfy the condition

in order t o preserve the property

(9.80)

A real distribution, that is

is obtained if the kernel satisfies the condition

Finally it shall be noted that although (9.73) gives a straightforward inter-

pretation of Cohen’s class, the implementation of (9.71) is more advantageous

For this, we first integrate over Y in (9.71) With

(9.83)

9.3 General Time-frequency Distributions 283

Convolution

Figure 9.6 Generation of a general time-frequency distribution of Cohen’s class

we obtain

T,,(t, W ) = // T ( U - t , r ) z*(u - -) 7- z(u + -) 7 ,-JUT ’ du d r (9.84)

Figure 9.6 shows the corresponding implementation

9.3.2 Examples of Shift-Invariant Time-Frequency

Distributions

Spectrogram The best known example of a shift-invariant time-frequency distribution is the spectrogram, described in detail in Chapter 7 An interest- ing relationship between the spectrogram and the Wigner distribution can be established [26] In order to explain this, the short-time Fourier transform is expressed in the form

F z ( t , w ) = z(t‘) h*(t - t’) ,-jut‘ dt’

CC

(9.85)

J-CC

Then the spectrogram is

Alternatively, with the abbreviation