Introduction. Spectral Estimation Overview Periodogram Bias, variance, and distribution Blackman-Tukey Method Welch-Bartlett Method Others

Size: px
Start display at page:

Download "Introduction. Spectral Estimation Overview Periodogram Bias, variance, and distribution Blackman-Tukey Method Welch-Bartlett Method Others"

Transcription

1 Spectral Estimation Overview Periodogram Bias, variance, and distribution Blackman-Tukey Method Welch-Bartlett Method Others Introduction R x (e jω ) r x (l)e jωl l= Most stationary random processes have continuous spectra If we have time, will discuss line spectra at end of term Recall the definition of above The estimation problem is to find a ˆR x (e jω ) given only a finite data record {x(n)} N If the autocorrelation can be estimated with great accuracy at long lags, can treat as a deterministic problem With short records (or equivalently, local stationarity), is more difficult J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. Example : Data Windowed Autocorrelation Solve for the expected value of the biased estimate of autocorrelation applied to a windowed data segment. How should w(n) be scaled such that the estimate is unbiased at l =. ˆr x (l) = x(n + l )w(n + l )x (n)w (n) N n=n l Periodogram ˆR x (e jω ) N v(n)e jωn = N N n= V (e jω ) where v(n) x(n)w(n). If w(n) =c (a constant), is called the Periodogram If w(n) is not constant, is called the Modified Periodogram and w(n) is called the data window Can be estimated quickly using the FFT Is related to the biased autocorrelation estimate we discussed last time N ˆR x (e jω )= ˆr v (l)e jωl l= (N ) J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver..

2 Example : Periodogram and Biased Autocorrelation Let x(n) be a finite duration sequence, to N. Show that the biased estimate of autocorrelation is given by Example : Workspace ˆr x (l) = x(l) x( l) N and that the DTFT of ˆr x (l) is equal to the Periodogram of x(n) when w(n) =. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6 Periodogram ˆR x (e jω ) N v(n)e jωn = V (e jω ) = F{ˆr v (l)} N N n= Remember that we must use the biased estimate of autocorrelation Otherwise the estimated may be negative at some frequencies The equality N V (e jω ) = F{ˆr v (l)} no longer holds This relationship to ˆr(l) means we can use the FFT to calculate ˆr(l) very efficiently Take FFT of signal, magnitude square, and inverse FFT Requires O(N log N) operations instead of O(N )! Must take care to zero pad sufficiently Example : Periodogram as a Filter Bank ˆR x (e jω ) N v(n)e jωn N n= How is the periodogram a filter bank? What is the transfer function of the filter? How good are the filters? How close are they to ideal filters? J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8

3 Example : Work Space Example : Peridogram Generate the periodogram for a signal from a known LTI system with two sets of complex conjugate poles close to one another and near the unit circle. Repeat for various signal lengths. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. Example : MATLAB Code M = 5; % Padding to eliminate transient N = [5,5,,5,5,]; % Length of signal segment NA = 5; % No. averages cb = 95; % NZ = ; b = poly([-.8,.97*exp(j *pi/),.97*exp(-j *pi/),....97*exp(j *pi/6),.97*exp(-j *pi/6)]); % Numerator a = poly([.8,.95*exp(j**pi/),.95*exp(-j**pi/),....95*exp(j*.5*pi/),.95*exp(-j*.5*pi/)]); % Denominator b = b*sum(a)/sum(b); % Scale DC gain to Example : Pole Zero Map of ARMA Process J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver..

4 Example : MATLAB Code Example : Signal figure h = circle; z = roots(b); p = roots(a); hold on; h = plot(real(z),imag(z), bo,real(p),imag(p), rx ); hold off; axis square; xlim([-..]); ylim([-..]); axislines; box off; Signal Length: Sample Index J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. Example : MATLAB Code n = ; w = randn(m+n,); x = filter(b,a,w); % System with known nx = length(x); x = x(nx-n+:nx); % Eliminate start-up transient (make stationary) figure; k = :n-; h = stem(k,x); set(h, Marker, none ); set(h, LineWidth,.5); box off; ylabel( Signal ); xlabel( Sample Index ); title(sprintf( Length:%d,n)); xlim([ n]); ylim([min(x) max(x)]); x 5 Example : True True J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6

5 [R,w] = freqz(b,a,nz); R = abs(r).^; f = w/(*pi); subplot(,,); h = plot(f,r, r ); set(h, LineWidth,.); box off; ylabel( ); title( True ); xlim([.5]); ylim([ max(r)*.]); subplot(,,); h = semilogy(f,r, r ); set(h, LineWidth,.); box off; ylabel( ); xlim([.5]); ylim([.*min(r) max(r)*5]); xlabel( ); Example : MATLAB Code x 5 Example : Periodogram Estimate N:5 NA: :95% True J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8 x 5 Example : Periodogram Estimate N:5 NA: :95% True x 5 Example : Periodogram Estimate N: NA: :95% True J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver..

6 x 5 Example : Periodogram Estimate N:5 NA: :95% True x 5 Example : Periodogram Estimate N:5 NA: :95% True J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. x 5 Example : Periodogram Estimate N: NA: :95% True Example : MATLAB Code for c = :length(n), nx = M+N(c); Rh = zeros(na,nz/+); kh = :NZ/+; fh = (kh-)/nz; for c = :NA, w = randn(m+n(c),); x = filter(b,a,w); % System with known x = x(nx-n(c)+:nx); % Eliminate start-up transient (make stationary) rh = (/N(c))*abs(fft(x,NZ)).^; Rh(c,:) = rh(kh). ; Rha = mean(rh); % Average Rhu = prctile(rh,-(-cb)/); % Upper confidence band Rhl = prctile(rh, (-cb)/); % Lower confidence band J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver..

7 Example : MATLAB Code figure; subplot(,,); h = plot(fh,rh(,:), g,fh,rhl, b,fh,rhu, b,f,r, r,fh,rha, k ); set(h(), LineWidth,.); set(h(:), LineWidth,.5); set(h(), LineWidth,.5); set(h(5), LineWidth,.5); ylabel( ); title(sprintf( N:%d NA:%d :%d%%,n(c),na,cb)); xlim([.5]); ylim([ max(r)*.5]); legend(h([ 5]),,, Average, True,); subplot(,,); h = semilogy(fh,rh(,:), g,fh,rhl, b,fh,rhu, b,f,r, r,fh,rha, k ); set(h(), LineWidth,.); set(h(:), LineWidth,.5); set(h(), LineWidth,.5); set(h(5), LineWidth,.5); ylabel( ); xlabel( ); xlim([.5]); ylim([.*min(r) max(r)*5]); Periodogram Properties The main disadvantage of the Periodogram appears to be excessive variance As with any estimator, we would like to know the bias, variance, covariance, confidence intervals, and distribution We will consider these each in turn J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6 [ ] E ˆRx (e jω ) = Periodogram Mean N l= (N ) E[ˆr v (l)] e jωl [ ] E[ˆr v (l)] = E x(n)w(n)x (n l)w (n l) N n= = r x (l)w(n)w (n l) N n= = r x (l) N n= = N r x(l)r w (l) w(n)w (n l) The expected value of ˆr v (l) is the true autocorrelation r x (l) windowed with r w (l) = N w(n) w( n). [ ] E ˆRx (e jω ) Periodogram Mean Continued = = N N l= (N ) N l= (N ) π E[ˆr x (l)] e jωl r x (l)r w (l)e jωl = R x (e ju ) ( π π N R w e j(ω u)) du [ ] Since E ˆRx (e jω ) R x (e jω ), the Periodogram is a biased estimator Here R w (e jω )= W(e jω ) is the ESD of the window Ideally, we would like R w (e jω )=πnδ(ω) The smearing limits our ability to distinguish sharp features (e.g. distinct peaks) in the J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8

8 [ ] E ˆRx (e jω ) Periodogram Asymptotic Bias = N N l= (N ) r x (l)r w (l)e jωl Most windows are relatively constant over a range of samples proportional to the window length: { N l N r w (l) l >N Most stationary signals have an autocorrelation that approaches zero as l Thus, the periodogram can be considered to be an asymptotically unbiased estimator lim N E [ ˆRx (e jω ) ] = R x (e jω ) Periodogram Asymptotic Bias Continued [ ] lim E ˆRx (e jω ) = R x (e jω ) N The estimator is unbiased as long as N n= w(n) = π R w (e jω )dω = N π π and the mainlobe window width decreases as N Not of much relevance (asymptotic result) Practically, the bias is introduced by the sidelobes and smearing of the spectrum by the main lobe Can reduce with better windows, but can only trade main lobe width for decreased sidebands If signal is white noise (flat spectrum), the smearing of the convolution does not cause bias J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. Periodogram Covariance Theory ˆR w (e jω )= N w(n)e jωn N n= = N w(n)e jk π N n N n= The text lists the covariance of the Periodogram, but does not derive it Details of the derivation can be found in [] An outline of the derivation is provided here Suppose for now that no zero padding is used so we calculate the DFT only at the N points that we have samples of the signal Let w(n) be a WGN process with variance σw and zero mean Orthogonal Components Recall that when zero padding is not used, the exponentials are orthogonal (e jk π n)( N e jl π n) N = n=<n> = n=<n> Now consider W (e jω ) evaluated at ω = k π N. W (e jω )= N n= e j(k l) π N n { N k =, ±N,±N,... otherwise w(n)e jωn If w(n) is a WGN process, then W (e jω ) is also a Gaussian random variable, since it is a linear combination of Gaussian RVs. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver..

9 WGN Mean and Variance E [ W (e jω ) ] = N n= E[w(n)]e jωn = cov [ W (e jω ),W(e jω ) ] =E [ W (e jω )W (e jω ) ] [( N )( N ) ] =E w(k)e jω k w(l)e jω l = = N k= N k= N l= l= E[w(k)w(l) ]e jω k e jω l k= l= N e j(ω ω )k = σ w N k= σ wδ(k l)e j(ω k ω l) WGN DFT Mean and Variance Continued If we don t use zero padding, π ω = m N where m and m are integers. Then cov [ W (e jω ),W(e jω ) ] = σ w = σ w ω = m π N N k= N e j(m m ) π N k k= e j(ω ω )k = σ wnδ(m m ) Thus, for WGN and no zero padding, the random variables are uncorrelated and therefore independent! The variance of the RVs is σ w. Alternative, the number of uncorrelated estimates in the frequency domain scales with N. More data more uncorrelated samples. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver χ Random Variables v= v= v= v= v=5 v= v=.8 v= v=.6 v= Let x i be a set of IID RVs with a Normal distribution, x i N(, ) Then the RV χ where v χ = x i i= has a χ distribution with v degrees of freedom The PDF and CDF is available in MATLAB and equivalent x = linspace(,,) ; df = :5; % Degrees of freedom nd = length(df); nx = length(x); P = zeros(nx,nd); C = zeros(nx,nd); for c=:5, P(:,c) = chipdf(x,df(c)); C(:,c) = chicdf(x,df(c)); ls{c} = sprintf( v=%d,df(c)); figure h = plot(x,p); set(h, LineWidth,.5); box off; xlim([ x(end)]); ylim([.5]); legend(ls); figure h = plot(x,c); set(h, LineWidth,.5); box off; xlim([ x(end)]); ylim([.5]); legend(ls); MATLAB Code J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6

10 WGN Mean and Variance Continued E [ W (e jω ) ] = N n= E[w(n)]e jωn = cov [ W (e jω ),W(e jω ) ] = σ wnδ(ω ω ) ˆR w (e jω )= N W (ejω ) If W (e jω ) is a complex Gaussian RV with zero mean and variance σ w, then what is the distribution of W (e jω )? E [ W (e jω ) ] ] =E [ Re{W (e jω )} +Im{W (e jω )} ] = σ w where the real and imaginary components are both independent Gaussian RVs each with variance σw/ Note that Im{W (e jω )} =if ω =or ω = π WGN Distribution Thus the distribution of W (e jω ) is given by { W (e jω ) σwn χ ()/ ω lπ χ () ω = lπ where χ () is a chi-squared distribution with one degree of freedom. Thus { ˆR w (e jω σ w χ ()/ ω lπ ) σ w χ () ω = lπ Thus, we know what the distribution of ˆR w (e jω ) is in the WGN case, but what about non-white processes? J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8 Generalized Linear Processes x(n) =h(n) w(n) = R x (e jω )= H(e jω ) σ w k= h(k)w(n k) This is a standard and widely used model for wide-sense stationary processes Note that h(n) is two sided (non-causal) We require that h(n) has finite energy (a sufficient condition for stability) But how is ˆR x (e jω ) related to the known distribution of ˆR w (e jω )? Periodogram Distribution It s beyond the scope of this class to prove it (see [, pg. ]), but ˆR x (e jω )=R x (e jω ) ˆR σw w (e jω )+R ε (ω) where E[R ε (ω)] = O(/N ) Thus R ε (ω) as N If we assume N is large we can disregard it N must be large enough to make the bias due to windowing negligible This means that the periodogram estimate is χ distributed when the process is Gaussian We can then immediately write the distribution of ˆR x (e jω ) J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver..

11 Since Periodogram Distribution { ˆR x (e jω R x (e jω )χ ()/ ω lπ ) R x (e jω )χ () ω = lπ var [ χ ()/ ] = var [ χ () ] = The variance of the periodogram is given by { var{ ˆR x (e jω R )} x(e jω ) <ω<π Rx(e jω ) ω =,π which agrees with the text. Periodogram Covariance Depends on the true, unknown so is of limited value Can make some qualitative observations Estimates at two frequencies ω =(π/n)k and ω =(π/n)k where k and k are integers are approximately uncorrelated cov{ ˆR x (e jω ), ˆR x (e jω )} for k k The number of uncorrelated frequency pairs scales with N Thus, the periodogram at adjacent frequencies, say ω and ω + δ, becomes more variable as N increases J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. Periodogram Variance (Again) [ ( ) ] var{ ˆR sin(ωn) x (e jω )} Rx(e jω ) + N sin(ω) For large N this can be approximated as { [ ] var ˆRx (e jω R ) x(e jω ) Rx(e jω ) <ω<π ω =,π The variance depends on the true spectrum Perhaps not surprising since the variance of ˆr(l) depended on the true autocorrelation Note that the variance does not decrease as N Thus, the Periodogram is not a consistent estimator The estimate becomes more variable as a function of frequency, but the variance at a fixed frequency does not decrease var[] var[] x 5 Example : Periodogram Variance N:5 NA: :95% Estimated Variance Theoretical Variance J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver..

12 Example : Periodogram Variance Example : Periodogram Variance var[] x 5 N:5 NA: :95% Estimated Variance Theoretical Variance var[] x 5 N: NA: :95% Estimated Variance Theoretical Variance var[] var[] J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6 Example : Periodogram Variance Example : Periodogram Variance var[] x 5 N:5 NA: :95% Estimated Variance Theoretical Variance var[] x 5 N:5 NA: :95% Estimated Variance Theoretical Variance var[] var[] J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8

13 var[] var[] x 5 Example : Periodogram Variance N: NA: :95% Estimated Variance Theoretical Variance Periodogram Variance ˆR x (e jω ) N v(n)e jωn = V (e jω ) = F{ˆr(l)} N N n= Since the ensemble variance of the Periodogram does not decrease as N increases, it is considered a poor estimator We cannot fix this by choosing a better window This is perhaps surprising, because it is the natural estimator To obtain a better estimator, we need to reduce the ensemble variance J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 Fixing Problems with the Periodogram As N increases, the ensemble variation stays constant The estimator is not consistent (statistically) As N increases, the covariation at neighboring frequencies, ω and ω + δ, decreases Thus we are essentially able to estimate the true at more frequencies We have better frequency resolution as N increases Key assumption: the true is smooth Clearly true for all ARMA processes with poles and zeros not too close to the unit circle Key concept: If the is smooth and our estimate consists of uncorrelated high-resolution estimates, we can obtain a better estimate by averaging estimates at adjacent frequencies Periodogram Smoothing Smoothing the Periodogram can be thought of in several ways It reduces the frequency resolution It blurs the spectrum It increases the bias of the estimator It reduces the variance of the estimator It could either increase or decrease the MSE depending on the degree of smoothing and the variability of the true Thus, we can tradeoff some of the excessive variance of the Periodogram for increases bias by smoothing There are several ways to achieve this Average contiguous values of the periodogram Average periodograms obtained from multiple segments Both approaches produce about the same quality of estimators J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5

14 Smoothing the Periodogram ˆR x (PS) (e jω k ) M + M j= M ˆR x (e jω k j ) One approach to smoothing is to apply a moving average filter in the frequency domain In the definition above (borrowed from the book), only applies if ˆR x (e jω ) is known at discrete frequencies ω k (π/n)k This is usually the case because we estimate using the FFT Since the examples at these discrete frequencies are approximately uncorrelated var{ (PS) ˆR x (e jω k )} M + var{ ˆR x (e jω k )} However, we increase the bias, especially near sharp features in the frequency domain x 5 Example 5: Smoothed Periodogram Estimate M: N:5 NA: :95% True J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 x 5 Example 5: Smoothed Periodogram Estimate M: N:5 NA: :95% True x 5 Example 5: Smoothed Periodogram Estimate M:5 N:5 NA: :95% True J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 55 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 56

15 x 5 Example 5: Smoothed Periodogram Estimate M: N:5 NA: :95% True Example 5: MATLAB Code Q = [,,5,]; % MA filter order N = 5; % Length of signal segment NA = ; % No. averages P = 5; % Padding to eliminate transient cb = 95; % NZ = 8; b = poly([-.8,.97*exp(j *pi/),.97*exp(-j *pi/),....97*exp(j *pi/6),.97*exp(-j *pi/6)]); % Numerator a = poly([.8,.95*exp(j**pi/),.95*exp(-j**pi/),....95*exp(j*.5*pi/),.95*exp(-j*.5*pi/)]); % Denominator b = b*sum(a)/sum(b); % Set DC gain to [R,w] = freqz(b,a,nz); R = abs(r).^; f = w/(*pi); J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 57 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 58 Example 5: MATLAB Code Continued for c = :length(q), nx = P+N; Rh = zeros(na,nz/+); kh = :NZ/+; fh = (kh-)/nz; for c = :NA, w = randn(p+n,); x = filter(b,a,w); % System with known x = x(nx-n+:nx); % Eliminate transient (make stationary) rh = (/N)*abs(fft(x,NZ)).^; rh = [rh;rh;rh]; % Eliminate edge effects rhs = filter(ones(q(c),)/q(c),,rh); rhs = rhs((q(c)-)/+(nz+:*nz)); % Account for filter delay Rh(c,:) = rhs(kh). ; Rha = mean(rh); % Average Rhu = prctile(rh,-(-cb)/); % Upper confidence band Rhl = prctile(rh, (-cb)/); % Lower confidence band Example 5: MATLAB Code Continued figure; FigureSet(, LTX ); subplot(,,); h = plot(fh,rh(,:), g,f,r, r,fh,rhl, b,fh,rhu, b,fh,rha, k ); set(h(), LineWidth,.); set(h(), LineWidth,.); set(h(:), LineWidth,.5); set(h(5), LineWidth,.5); ylabel( ); title(sprintf( Q:%d N:%d NA:%d :%d%%,q(c),n,na,cb)); xlim([.5]); ylim([ max(r)*.5]); legend(h([ 5]),,, Average, True,); subplot(,,); h = semilogy(fh,rh(,:), g,f,r, r,fh,rhl, b,fh,rhu, b,fh,rha, k ); set(h(), LineWidth,.); set(h(), LineWidth,.); set(h(:), LineWidth,.5); set(h(5), LineWidth,.5); ylabel( ); xlabel( ); xlim([.5]); ylim([.*min(r) max(r)*5]); J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 59 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6

16 Periodogram Smoothing Comments ˆR x (PS) (e jω k ) { } var R x (PS) (e jω ) M + M j= M ˆR x (e jω k j ) M + var { R x (e jω ) } Smoothing the periodogram is a good idea Trades reduced frequency resolution and increased bias for reduced variance The moving average is just a lowpass filter There are many other lowpass filters that might perform better Examples: smoothing splines, local polynomial models, weighted averaging (kernel smoothing) Blackman-Tukey Spectral Estimation Idea [ ] E ˆRx (e jω ) = π R x (e ju ) π π N R w(e j(ω u) )du Recall that windowing the signal is equivalent to convolving the signal s with the window s, in expectation Despite the blurring and smoothing of this effect, the estimate is still has too high of variance We can increase the smoothing and reduce the variance by multiplying the autocorrelation by an even shorter window J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6 Blackman-Tukey Spectral Estimation ˆR x (BT) (e jω ) L l= (L ) ˆr x (l)w a (l)e jωl Here the window is user specified and has a finite duration of L samples Called the correlation or lag window In order for the estimate to be nonnegative, W (e jω ) for all ω Not all of the windows we described have this property In the frequency domain, the product ˆr x (l)w(l) is equivalent to convolving with W (e jω ) Assuming the window s spectrum is bumped shape, this convolution smooths the spectrum Note that the FFT can be used to calculate both ˆr x (l) and (e jω ) efficiently ˆR (BT) x Blackman-Tukey Bias & Variance ˆR x (BT) (e jω ) L l= (L ) ˆr x (l)w a (l)e jωl Conceptually, this is using an even more biased estimate of the autocorrelation: ˆr x (l)w a (l) The additional bias is towards zero (BT) Thus, it reduces the variance of both ˆr x (l) and ˆR x (e jω ) The user controls this tradeoff by picking the correlation window w a (l) The window can be any positive definite window with even symmetry (why?) The length of the window L has a larger impact than the shape J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6

17 x 5 Example 6: Blackman-Tukey Estimate N:5 L:5 NA: :95% d CIs True Coverage.8.6. Example 7: Blackman-Tukey Coverage N:5 L:5 NA: :95%. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 65 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 66 Example 7: Blackman-Tukey Estimate Example 8: Blackman-Tukey Coverage x 5 N:5 L:5 NA: :95% d CIs True Coverage.8.6. N:5 L:5 NA: :95%. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 67 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 68

18 x 5 Example 8: Blackman-Tukey Estimate N:5 L: NA: :95% d CIs True Coverage.8.6. Example 9: Blackman-Tukey Coverage N:5 L: NA: :95%. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 69 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 Example 9: Blackman-Tukey Estimate Example : Blackman-Tukey Coverage x 5 N:5 L:5 NA: :95% d CIs True Coverage.8.6. N:5 L:5 NA: :95%. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7

19 x 5 Example : Blackman-Tukey Estimate N:5 L:5 NA: :95% d CIs True Coverage.8.6. Example : Blackman-Tukey Coverage N:5 L:5 NA: :95%. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 Example : MATLAB Code L = [5,5,,5,5]; % MA filter order N = 5; % Length of signal segment NA = ; % No. averages P = 5; % Padding to eliminate transient cb = 95; % NZ = ; np = ^nextpow(*n-); % Figure out how much to pad the signal b = poly([-.8,.97*exp(j *pi/),.97*exp(-j *pi/),....97*exp(j *pi/6),.97*exp(-j *pi/6)]); % Numerator a = poly([.8,.95*exp(j**pi/),.95*exp(-j**pi/),....95*exp(j*.5*pi/),.95*exp(-j*.5*pi/)]); % Denominator b = b*sum(a)/sum(b); % Set DC gain to [R,w] = freqz(b,a,nz); R = abs(r).^; f = w/(*pi); Example : MATLAB Code Continued for c = :length(l), nx = P+N; Rh = zeros(na,nz/+); kh = :NZ/+; fh = (kh-)/nz; for c = :NA, w = randn(p+n,); x = filter(b,a,w); % System with known x = x(nx-n+:nx); % Eliminate transient (make stationary) xf = fft(x,np); % Calculate FFT of x r = ifft(xf.*conj(xf)); % Calculate biased autocorrelation r = real(r(:nx))/n; % Eliminate superfluous zeros no = (L(c)+)/; % Number of one-sided points to included r = [r(no:-:);r(:no)]; % Make two-sided wn = blackman(length(r)); % Create window wn = wn/max(wn); % Make maximum = rh = abs(fft(r.*wn,nz)); % Window autocorrelation Rh(c,:) = rh(kh). ; Rha = mean(rh); % Average Rhu = prctile(rh,-(-cb)/); % Upper confidence band Rhl = prctile(rh, (-cb)/); % Lower confidence band J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 75 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 76

20 Example : MATLAB Code Continued figure; subplot(,,); h = plot(fh,rh(,:), g,f,r, r,fh,rhl, b,fh,rhu, b,fh,rha, k ); set(h(), LineWidth,.); set(h(), LineWidth,.); set(h(:), LineWidth,.5); set(h(5), LineWidth,.5); ylabel( ); title(sprintf( N:%d L:%d NA:%d :%d%%,n,l(c),na,cb)); xlim([.5]); ylim([ max(r)*.5]); AxisSet(8); legend(h([ 5]),,, Average, True,); subplot(,,); h = semilogy(fh,rh(,:), g,f,r, r,fh,rhl, b,fh,rhu, b,fh,rha, k ); set(h(), LineWidth,.); set(h(), LineWidth,.); set(h(:), LineWidth,.5); set(h(5), LineWidth,.5); ylabel( ); xlabel( ); xlim([.5]); ylim([.*min(r) max(r)*5]); ˆR (BT) x (e jω ) [ ] E ˆR(BT) x (e jω ) Blackman-Tukey Mean = = L l= (L ) L l= (L ) L l= (L ) ˆr x (l)w a (l)e jωl E[ˆr x (l)] w a (l)e jωl [ r x (l) ] N r w(l) w a (l)e jωl = R x (e jω ) W (e jω ) W a (e jω ) Recall that the periodogram had an expected value given by the convolution of the Bartlett window s spectrum and the true Multiplication by another window causes another convolution Text assumes the data window w(l) is rectangular J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 77 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 78 Blackman-Tukey Mean [ ] E ˆR(BT) x (e jω ) = R x (e jω ) W (e jω ) W a (e jω ) Note that the convolution is circular since each term is a periodic function of ω The estimate is asymptotically unbiased if w a () = As L and N approach, both windows become impulse trains Blackman-Tukey Sampling Distribution ˆR x (BT) (e jω ) L l= (L ) ˆr x (l)w a (l)e jωl Sampling properties of the spectral estimates are extremely complicated Exact results can only be obtained in special circumstances Most expressions for mean, variance, covariance, and distributions are asymptotic and only apply when N is large The asymptotic expressions assume that L increases with N, but more slowly than N: L, N, L/N J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 79 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8

21 Blackman-Tukey Covariance (BT) cov{ ˆR x (e jω (BT) ), ˆR x (e jω )} π R πn x(e ju )W a (e j(ω u) )W a (e j(ω u) )du π Assumptions N is large so that W B (e jω ) behaves like an impulse train L is sufficiently large that W a (e jω ) is narrow and the product W a (e jω )W a (e jω ) is negligible Thus, covariance increases as The window s spectrum, W a (e j(ω u) ), grows wider The overlap between windows centered at different frequencies, ω and ω, increases For estimates to be uncorrelated, they must be separated in frequency by an amount greater than the bandwidth of the window Blackman-Tukey Variance (BT) var{ ˆR x (e jω )} π R πn x(e ju )Wa (e j(ω u) )du π If Rx(e jω ) is smooth over the width of the window and can be approximated as a constant, then π (BT) var{ ˆR x (e jω )} Rx(e jω ) Wa (e j(ω u) )du πn π = Rx(e jω ) E w N <ω<π E w = L l= (L ) w a(l) E w is the energy of the correlation window is called the variance reduction factor or variance ratio E w N J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8 log Blackman-Tukey Confidence Intervals (BT) ˆR x (e jω ) log χ ν( α/) ν log (BT) ˆR x (e jω ) < log ν = < log ˆR (BT) (BT) ˆR x (e jω ) + log N L l= (L ) w a(l) x (e jω ), ν χ ν( α/) Example : Blackman-Tukey Bias-Variance Tradeoff Plot the bias and variance at the frequencies of the poles and zeros of the LTI system for a range of window lengths ranging from 5 5 samples. These are only approximate confidence intervals Only works when Bias is small N L N such that L/N (this is why we don t obtain the Periodogram estimate when L = N) This estimate is a linear combination of the periodogram estimate, so it is also approximately has a χ distribution, but with more degrees of freedom J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8

22 Example : Blackman-Tukey Bias-Variance Frequencies Example : Blackman-Tukey Bias-Variance Tradeoff x 5 True ω =.785 rad/sample N:5 NA: x ω =.5 rad/sample x 9 ω =.56 rad/sample Variance Bias Variance + Bias MSE ω =.96 rad/sample Autocorrelation Window Length, L (samples) 5 Autocorrelation Window Length, L (samples) J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 85 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 86 Example : Blackman-Tukey Bias-Variance Tradeoff Example : MATLAB Code L = round(linspace(,5,5))*+; % Autocorrelation window length ef = [pi/ pi/6 *pi/.5*pi/]; % Evaluation frequencies N = 5; % Length of signal segment NA = ; % No. averages P = 5; % Padding to eliminate transient NZ = ^9; % No. zeros np = ^nextpow(*n-); % Figure out how much to pad the signal b = poly([-.8,.97*exp(j *pi/),.97*exp(-j *pi/),....97*exp(j *pi/6),.97*exp(-j *pi/6)]); % Numerator coefficients a = poly([.8,.95*exp(j**pi/),.95*exp(-j**pi/),....95*exp(j*.5*pi/),.95*exp(-j*.5*pi/)]); % Denominator coefficients b = b*sum(a)/sum(b); % Scale DC gain to [R,w] = freqz(b,a,nz); R = abs(r).^; R = R ; f = w; nl = length(l); Bias = zeros(nl,nz); Var = zeros(nl,nz); kh = :NZ; nx = P+N; J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 87 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 88

23 Example : MATLAB Code Continued for c = :nl, Rh = zeros(na,nz); no = (L(c)+)/; % Number of one-sided points to included in estimate wn = blackman(no*-+); % Create window wn = wn(:end-); % Trim MATLAB s zeros wn = wn/max(wn); % Make maximum = for c = :NA, w = randn(p+n,); x = filter(b,a,w); % System with known x = x(nx-n+:nx); % Eliminate start-up transient (make stationary) xf = fft(x,np); % Calculate FFT of x r = ifft(xf.*conj(xf)); % Calculate biased autocorrelation r = real(r(:nx))/n; % Eliminate superfluous zeros and nearly zero imaginar r = [r(no:-:);r(:no)]; % Make two-sided rh = abs(fft(r.*wn,*nz)); % Window autocorrelation and calculate the estimate Rh (c,:) = rh(kh). ; Rha = mean(rh); % Average Bias(c,:) = R-Rha; Var (c,:) = mean((rh - ones(na,)*rha).^); MSE (c,:) = mean((rh - ones(na,)*r ).^); drawnow; Example : MATLAB Code Continued figure; for c=:length(ef), [jnk,id] = min(abs(ef(c)-f)); subplot(,,c); h = plot(l,var(:,id),l,bias(:,id).^,l,var(:,id)+bias(:,id).^,l,mse(:,id)); set(h(), LineWidth,); set(h(), LineWidth,); xlim([min(l) max(l)]); ylim([.5*max(mse(:,id))]); box off; ylabel(sprintf( \\omega = %5.f rad/sample,ef(c))); if c==, xlabel( Autocorrelation Window Length, L (samples) ); elseif c==, title(sprintf( N:%d NA:%d,N,NA)); elseif c==, xlabel( Autocorrelation Window Length, L (samples) ); legend( Variance, Bias^, Variance + Bias^, MSE ); figure; smse = zeros(size(mse)); for c=:length(f), smse(:,c) = MSE(:,c)/max(MSE(:,c)); J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 89 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9 h = imagesc(f,l,smse,[ ]); [jnk,id] = min(smse); hold on; plot(f,l(id), w. ); hold off; Periodogram Averaging ˆR x (PA) (e jω ) K ˆR x (e jω ) K i= Consider K IID random variables The variance of the mean is /K times the variance of the random variables If we had different independent realizations of the process we could average them and obtain a similar reduction in the variance of the In most situations, only a single realization is available Can subdivide the record into K possibly overlapping segments of length L J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9

24 ˆR x (WB) (e jω k ) K K L i= Welch-Bartlett Spectral Estimation L v i (n)e jωn = KL n= K i= V i (e jω ) where v(n) x(n)w(n). Where w(n) is called the data window Window does not need to be positive definite Window trades sidelobe leakage (ripples) for mainlobe width (blurring) If no overlap between windows, called the Bartlett estimate If overlap, called Welch s method Example : Welch-Bartlett Spectral Estimation Generate the Welch-Bartlett estimated for a signal from a known LTI system with two sets of complex conjugate poles close to one another and near the unit circle. Repeat for various segment lengths. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 9 x 5 Example : Welch-Bartlett Estimate N:5 L: OL:5% K:99 NA: :95% True x 5 Example : Welch-Bartlett Estimate N:5 L:5 OL:5% K:9 NA: :95% True J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 95 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 96

25 x 5 Example : Welch-Bartlett Estimate N:5 L: OL:5% K:9 NA: :95% True x 5 Example : Welch-Bartlett Estimate N:5 L: OL:5% K: NA: :95% True J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 97 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 98 Example : MATLAB Code L = [,5,,]; % Segment lengths O =.5; % Overlap M = 5; % Padding to eliminate transient N = 5; % Length of signal segment NA = ; % No. averages cb = 95; % NZ = ; b = poly([-.8,.97*exp(j *pi/),.97*exp(-j *pi/),....97*exp(j *pi/6),.97*exp(-j *pi/6)]); % Numerator coefficients a = poly([.8,.95*exp(j**pi/),.95*exp(-j**pi/),....95*exp(j*.5*pi/),.95*exp(-j*.5*pi/)]); % Denominator coefficients b = b*sum(a)/sum(b); % Scale DC gain to [R,w] = freqz(b,a,nz); R = abs(r).^; f = w/(*pi); Example : MATLAB Code Continued for c = :length(l), K = floor((n-o*l(c))./(l(c)-o*l(c))); % No. segments ss = floor(l(c)-o*l(c)); % Step size nx = M+N; Rh = zeros(na,nz/+); kh = :NZ/+; fh = (kh-)/nz; for c = :NA, w = randn(m+n,); x = filter(b,a,w); % System with known x = x(nx-n+:nx); % Eliminate start-up transient (make stationary) rh = zeros(nz,); for c = :K, id = (:L(c)) + (c-)*ss; rh = rh + abs(fft(x(id),nz)).^; rh = rh/(k*l(c)); Rh(c,:) = rh(kh). ; Rha = mean(rh); % Average Rhu = prctile(rh,-(-cb)/); % Upper confidence band Rhl = prctile(rh, (-cb)/); % Lower confidence band J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 99 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver..

26 Example : MATLAB Code Continued figure; subplot(,,); h = plot(fh,rh(,:), g,f,r, r,fh,rhl, b,fh,rhu, b,fh,rha, k ); set(h(), LineWidth,.5); set(h(), LineWidth,.); set(h(:), LineWidth,.5); set(h(5), LineWidth,.8); ylabel( ); title(sprintf( N:%d L:%d OL:%d%% K:%d NA:%d :%d%%,n,l(c),o* xlim([.5]); ylim([ max(r)*.5]); legend(h([ 5]),, True,, Average Estimat subplot(,,); h = semilogy(fh,rh(,:), g,f,r, r,fh,rhl, b,fh,rhu, b,fh,rha, k ); set(h(), LineWidth,.5); set(h(), LineWidth,.); set(h(:), LineWidth,.5); set(h(5), LineWidth,.8); ylabel( ); xlabel( ); xlim([.5]); ylim([.*min(r) max(r)*5]); [ ] E ˆR(WB) x (e jω k ) Welch-Bartlett Mean = K = E K i= [ L [ E V i (e jω ) ] L Vi (e jω ) ] = R x (e jω ) L R w(e jω ) = πl π π R x (e jω )R w (e j(ω θ) )dθ As with the periodogram, if the window is normalized such that L w (n) = π R w (e jω )dω = L π π n= then the estimate is asymptotically unbiased J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. Welch-Bartlett Variance (WB) var{ ˆR x (e jω k )} K var{ ˆR x (e jω k )} (WB) var{ ˆR x (e jω k )} K Assumes segments are independent Not true, so estimate is optimistic Book fails to mention this ˆR x (WB) (e jω k ) K K L i= Welch-Bartlett Concepts L v i (n)e jωn n= = KL K i= Vi (e jω ) Thus, the user can once again trade variance for bias If N = KL is fixed, then can reduce variance at the expense of increased variance by increasing K and decreasing L The variance is insensitive to the shape of the window Increasing overlap up to 5% reduces variance without substantially impacting the bias Further overlap increases the computational requirements, but does not help bias or variance Confidence intervals listed in the text J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver..

27 Welch-Bartlett Comments ˆR (WB) x (e jω k )= KL K i= V i (e jω ) Consider the case when the overlap is L samples This would reduce variance the most In this case it becomes equivalent to periodogram smoothing [, pp ]! Bartlett originally proposed averaging periodograms of separate segments, but showed that this was essentially equivalent to the Blackman-Tukey method with a Bartlett autocorrelation window [, pp. 9-] When the overlap is less than L It is only approximately equivalent to periodogram smoothing It requires L FFTs Variance is increased Summary We discussed several different nonparametric methods of spectral estimation The natural estimate was the periodogram This estimate has unacceptable variance for most applications There are two fundamental approaches to reduce variance Periodogram smoothing Periodogram averaging They have comparable performance and ultimately are approximately equivalent Each enables the user to control the bias-variance tradeoff Typically the smoothing parameter is chosen to minimize bias Only approximate confidence intervals are available for each J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 5 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 6 MATLAB Resources The MATLAB implementation of these functions is constantly changing (improving?) The documentation is pretty good Open up the help window (helpwin) Signal Processing Toolbox Statistical Signal Processing Most of the techniques we have discussed are implemented Interface is not user-friendly References [] M. B. Priestley. Spectral Analysis and Time Series. Academic Press, 98. J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 7 J. McNames Portland State University ECE 58/68 Spectral Estimation Ver.. 8

Fast Fourier Transform Discrete-time windowing Discrete Fourier Transform Relationship to DTFT Relationship to DTFS Zero padding

Fast Fourier Transform Discrete-time windowing Discrete Fourier Transform Relationship to DTFT Relationship to DTFS Zero padding Fast Fourier Transform Discrete-time windowing Discrete Fourier Transform Relationship to DTFT Relationship to DTFS Zero padding J. McNames Portland State University ECE 223 FFT Ver. 1.03 1 Fourier Series

More information

Fast Fourier Transform Discrete-time windowing Discrete Fourier Transform Relationship to DTFT Relationship to DTFS Zero padding

Fast Fourier Transform Discrete-time windowing Discrete Fourier Transform Relationship to DTFT Relationship to DTFS Zero padding Fast Fourier Transform Discrete-time windowing Discrete Fourier Transform Relationship to DTFT Relationship to DTFS Zero padding Fourier Series & Transform Summary x[n] = X[k] = 1 N k= n= X[k]e jkω

More information

Spectrograms Overview Introduction and motivation Windowed estimates as filter banks Uncertainty principle Examples Other estimates

Spectrograms Overview Introduction and motivation Windowed estimates as filter banks Uncertainty principle Examples Other estimates Spectrograms Overview Introduction and motivation Windowed estimates as filter banks Uncertainty principle Examples Other estimates J. McNames Portland State University ECE 3 Spectrograms Ver. 1.1 1 Introduction

More information

Windowing Overview Relevance Main lobe/side lobe tradeoff Popular windows Examples

Windowing Overview Relevance Main lobe/side lobe tradeoff Popular windows Examples ing Overview Relevance Main lobe/side lobe tradeoff Popular windows Examples ing In practice, we cannot observe a signal x(n) for n = to n = Thus we must truncate, or window, the signal: x(n) w(n) ing

More information

Overview of Discrete-Time Fourier Transform Topics Handy Equations Handy Limits Orthogonality Defined orthogonal

Overview of Discrete-Time Fourier Transform Topics Handy Equations Handy Limits Orthogonality Defined orthogonal Overview of Discrete-Time Fourier Transform Topics Handy equations and its Definition Low- and high- discrete-time frequencies Convergence issues DTFT of complex and real sinusoids Relationship to LTI

More information

J. McNames Portland State University ECE 223 DT Fourier Series Ver

J. McNames Portland State University ECE 223 DT Fourier Series Ver Overview of DT Fourier Series Topics Orthogonality of DT exponential harmonics DT Fourier Series as a Design Task Picking the frequencies Picking the range Finding the coefficients Example J. McNames Portland

More information

Core Concepts Review. Orthogonality of Complex Sinusoids Consider two (possibly non-harmonic) complex sinusoids

Core Concepts Review. Orthogonality of Complex Sinusoids Consider two (possibly non-harmonic) complex sinusoids Overview of Continuous-Time Fourier Transform Topics Definition Compare & contrast with Laplace transform Conditions for existence Relationship to LTI systems Examples Ideal lowpass filters Relationship

More information

Practical Spectral Estimation

Practical Spectral Estimation Digital Signal Processing/F.G. Meyer Lecture 4 Copyright 2015 François G. Meyer. All Rights Reserved. Practical Spectral Estimation 1 Introduction The goal of spectral estimation is to estimate how the

More information

Terminology Suppose we have N observations {x(n)} N 1. Estimators as Random Variables. {x(n)} N 1

Terminology Suppose we have N observations {x(n)} N 1. Estimators as Random Variables. {x(n)} N 1 Estimation Theory Overview Properties Bias, Variance, and Mean Square Error Cramér-Rao lower bound Maximum likelihood Consistency Confidence intervals Properties of the mean estimator Properties of the

More information

II. Nonparametric Spectrum Estimation for Stationary Random Signals - Non-parametric Methods -

II. Nonparametric Spectrum Estimation for Stationary Random Signals - Non-parametric Methods - II. onparametric Spectrum Estimation for Stationary Random Signals - on-parametric Methods - - [p. 3] Periodogram - [p. 12] Periodogram properties - [p. 23] Modified periodogram - [p. 25] Bartlett s method

More information

Probability Space. J. McNames Portland State University ECE 538/638 Stochastic Signals Ver

Probability Space. J. McNames Portland State University ECE 538/638 Stochastic Signals Ver Stochastic Signals Overview Definitions Second order statistics Stationarity and ergodicity Random signal variability Power spectral density Linear systems with stationary inputs Random signal memory Correlation

More information

SPECTRUM. Deterministic Signals with Finite Energy (l 2 ) Deterministic Signals with Infinite Energy N 1. n=0. N N X N(f) 2

SPECTRUM. Deterministic Signals with Finite Energy (l 2 ) Deterministic Signals with Infinite Energy N 1. n=0. N N X N(f) 2 SPECTRUM Deterministic Signals with Finite Energy (l 2 ) Energy Spectrum: S xx (f) = X(f) 2 = 2 x(n)e j2πfn n= Deterministic Signals with Infinite Energy DTFT of truncated signal: X N (f) = N x(n)e j2πfn

More information

Nonparametric and Parametric Defined This text distinguishes between systems and the sequences (processes) that result when a WN input is applied

Nonparametric and Parametric Defined This text distinguishes between systems and the sequences (processes) that result when a WN input is applied Linear Signal Models Overview Introduction Linear nonparametric vs. parametric models Equivalent representations Spectral flatness measure PZ vs. ARMA models Wold decomposition Introduction Many researchers

More information

Discrete Fourier Transform

Discrete Fourier Transform Discrete Fourier Transform Virtually all practical signals have finite length (e.g., sensor data, audio records, digital images, stock values, etc). Rather than considering such signals to be zero-padded

More information

Laboratory Project 2: Spectral Analysis and Optimal Filtering

Laboratory Project 2: Spectral Analysis and Optimal Filtering Laboratory Project 2: Spectral Analysis and Optimal Filtering Random signals analysis (MVE136) Mats Viberg and Lennart Svensson Department of Signals and Systems Chalmers University of Technology 412 96

More information

LAB 6: FIR Filter Design Summer 2011

LAB 6: FIR Filter Design Summer 2011 University of Illinois at Urbana-Champaign Department of Electrical and Computer Engineering ECE 311: Digital Signal Processing Lab Chandra Radhakrishnan Peter Kairouz LAB 6: FIR Filter Design Summer 011

More information

Fourier Analysis of Signals Using the DFT

Fourier Analysis of Signals Using the DFT Fourier Analysis of Signals Using the DFT ECE 535 Lecture April 29, 23 Overview: Motivation Many applications require analyzing the frequency content of signals Speech processing study resonances of vocal

More information

6.435, System Identification

6.435, System Identification System Identification 6.435 SET 3 Nonparametric Identification Munther A. Dahleh 1 Nonparametric Methods for System ID Time domain methods Impulse response Step response Correlation analysis / time Frequency

More information

Change in Notation. e (i) (n) Note that this is inconsistent with the notation used earlier, c k+1x(n k) e (i) (n) x(n i) ˆx(n i) M

Change in Notation. e (i) (n) Note that this is inconsistent with the notation used earlier, c k+1x(n k) e (i) (n) x(n i) ˆx(n i) M Overview of Linear Prediction Terms and definitions Nonstationary case Stationary case Forward linear prediction Backward linear prediction Stationary processes Exchange matrices Examples Properties Introduction

More information

Overview of Sampling Topics

Overview of Sampling Topics Overview of Sampling Topics (Shannon) sampling theorem Impulse-train sampling Interpolation (continuous-time signal reconstruction) Aliasing Relationship of CTFT to DTFT DT processing of CT signals DT

More information

J. McNames Portland State University ECE 223 Sampling Ver

J. McNames Portland State University ECE 223 Sampling Ver Overview of Sampling Topics (Shannon) sampling theorem Impulse-train sampling Interpolation (continuous-time signal reconstruction) Aliasing Relationship of CTFT to DTFT DT processing of CT signals DT

More information

Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science : Discrete-Time Signal Processing

Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science : Discrete-Time Signal Processing Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science 6.34: Discrete-Time Signal Processing OpenCourseWare 006 ecture 8 Periodogram Reading: Sections 0.6 and 0.7

More information

Digital Signal Processing: Signal Transforms

Digital Signal Processing: Signal Transforms Digital Signal Processing: Signal Transforms Aishy Amer, Mohammed Ghazal January 19, 1 Instructions: 1. This tutorial introduces frequency analysis in Matlab using the Fourier and z transforms.. More Matlab

More information

Definition of Discrete-Time Fourier Transform (DTFT)

Definition of Discrete-Time Fourier Transform (DTFT) Definition of Discrete-Time ourier Transform (DTT) {x[n]} = X(e jω ) + n= {X(e jω )} = x[n] x[n]e jωn Why use the above awkward notation for the transform? X(e jω )e jωn dω Answer: It is consistent with

More information

Overview of Bode Plots Transfer function review Piece-wise linear approximations First-order terms Second-order terms (complex poles & zeros)

Overview of Bode Plots Transfer function review Piece-wise linear approximations First-order terms Second-order terms (complex poles & zeros) Overview of Bode Plots Transfer function review Piece-wise linear approximations First-order terms Second-order terms (complex poles & zeros) J. McNames Portland State University ECE 222 Bode Plots Ver.

More information

DFT & Fast Fourier Transform PART-A. 7. Calculate the number of multiplications needed in the calculation of DFT and FFT with 64 point sequence.

DFT & Fast Fourier Transform PART-A. 7. Calculate the number of multiplications needed in the calculation of DFT and FFT with 64 point sequence. SHRI ANGALAMMAN COLLEGE OF ENGINEERING & TECHNOLOGY (An ISO 9001:2008 Certified Institution) SIRUGANOOR,TRICHY-621105. DEPARTMENT OF ELECTRONICS AND COMMUNICATION ENGINEERING UNIT I DFT & Fast Fourier

More information

CHAPTER 2 RANDOM PROCESSES IN DISCRETE TIME

CHAPTER 2 RANDOM PROCESSES IN DISCRETE TIME CHAPTER 2 RANDOM PROCESSES IN DISCRETE TIME Shri Mata Vaishno Devi University, (SMVDU), 2013 Page 13 CHAPTER 2 RANDOM PROCESSES IN DISCRETE TIME When characterizing or modeling a random variable, estimates

More information

ADSP ADSP ADSP ADSP. Advanced Digital Signal Processing (18-792) Spring Fall Semester, Department of Electrical and Computer Engineering

ADSP ADSP ADSP ADSP. Advanced Digital Signal Processing (18-792) Spring Fall Semester, Department of Electrical and Computer Engineering Advanced Digital Signal rocessing (18-792) Spring Fall Semester, 201 2012 Department of Electrical and Computer Engineering ROBLEM SET 8 Issued: 10/26/18 Due: 11/2/18 Note: This problem set is due Friday,

More information

ECE 636: Systems identification

ECE 636: Systems identification ECE 636: Systems identification Lectures 3 4 Random variables/signals (continued) Random/stochastic vectors Random signals and linear systems Random signals in the frequency domain υ ε x S z + y Experimental

More information

Basic Design Approaches

Basic Design Approaches (Classic) IIR filter design: Basic Design Approaches. Convert the digital filter specifications into an analog prototype lowpass filter specifications. Determine the analog lowpass filter transfer function

More information

Statistical signal processing

Statistical signal processing Statistical signal processing Short overview of the fundamentals Outline Random variables Random processes Stationarity Ergodicity Spectral analysis Random variable and processes Intuition: A random variable

More information

Contents. Digital Signal Processing, Part II: Power Spectrum Estimation

Contents. Digital Signal Processing, Part II: Power Spectrum Estimation Contents Digital Signal Processing, Part II: Power Spectrum Estimation 5. Application of the FFT for 7. Parametric Spectrum Est. Filtering and Spectrum Estimation 7.1 ARMA-Models 5.1 Fast Convolution 7.2

More information

ECE 636: Systems identification

ECE 636: Systems identification ECE 636: Systems identification Lectures 7 8 onparametric identification (continued) Important distributions: chi square, t distribution, F distribution Sampling distributions ib i Sample mean If the variance

More information

Basics on 2-D 2 D Random Signal

Basics on 2-D 2 D Random Signal Basics on -D D Random Signal Spring 06 Instructor: K. J. Ray Liu ECE Department, Univ. of Maryland, College Park Overview Last Time: Fourier Analysis for -D signals Image enhancement via spatial filtering

More information

Problem Set 9 Solutions

Problem Set 9 Solutions Problem Set 9 Solutions EE23: Digital Signal Processing. From Figure below, we see that the DTFT of the windowed sequence approaches the actual DTFT as the window size increases. Gibb s phenomenon is absent

More information

Complement on Digital Spectral Analysis and Optimal Filtering: Theory and Exercises

Complement on Digital Spectral Analysis and Optimal Filtering: Theory and Exercises Complement on Digital Spectral Analysis and Optimal Filtering: Theory and Exercises Random Processes With Applications (MVE 135) Mats Viberg Department of Signals and Systems Chalmers University of Technology

More information

x n C l 1 u n C m 1 y n C p 1 v n C p 1

x n C l 1 u n C m 1 y n C p 1 v n C p 1 Kalman Filter Derivation Examples Time and Measurement Updates x u n v n, State Space Model Review x n+ = F n x n + G n u n x u k v k y n = H n x n + v n Π = Q n δ nk S n δ nk Snδ nk R n δ nk F n C l l

More information

Multimedia Signals and Systems - Audio and Video. Signal, Image, Video Processing Review-Introduction, MP3 and MPEG2

Multimedia Signals and Systems - Audio and Video. Signal, Image, Video Processing Review-Introduction, MP3 and MPEG2 Multimedia Signals and Systems - Audio and Video Signal, Image, Video Processing Review-Introduction, MP3 and MPEG2 Kunio Takaya Electrical and Computer Engineering University of Saskatchewan December

More information

Chirp Transform for FFT

Chirp Transform for FFT Chirp Transform for FFT Since the FFT is an implementation of the DFT, it provides a frequency resolution of 2π/N, where N is the length of the input sequence. If this resolution is not sufficient in a

More information

HW13 Solutions. Pr (a) The DTFT of c[n] is. C(e jω ) = 0.5 m e jωm e jω + 1

HW13 Solutions. Pr (a) The DTFT of c[n] is. C(e jω ) = 0.5 m e jωm e jω + 1 HW3 Solutions Pr..8 (a) The DTFT of c[n] is C(e jω ) = = =.5 n e jωn + n=.5e jω + n= m=.5 n e jωn.5 m e jωm.5e jω +.5e jω.75 =.5 cos(ω) }{{}.5e jω /(.5e jω ) C(e jω ) is the power spectral density. (b)

More information

8 The Discrete Fourier Transform (DFT)

8 The Discrete Fourier Transform (DFT) 8 The Discrete Fourier Transform (DFT) ² Discrete-Time Fourier Transform and Z-transform are de ned over in niteduration sequence. Both transforms are functions of continuous variables (ω and z). For nite-duration

More information

Time series models in the Frequency domain. The power spectrum, Spectral analysis

Time series models in the Frequency domain. The power spectrum, Spectral analysis ime series models in the Frequency domain he power spectrum, Spectral analysis Relationship between the periodogram and the autocorrelations = + = ( ) ( ˆ α ˆ ) β I Yt cos t + Yt sin t t= t= ( ( ) ) cosλ

More information

Discrete Time Fourier Transform (DTFT)

Discrete Time Fourier Transform (DTFT) Discrete Time Fourier Transform (DTFT) 1 Discrete Time Fourier Transform (DTFT) The DTFT is the Fourier transform of choice for analyzing infinite-length signals and systems Useful for conceptual, pencil-and-paper

More information

Adaptive Filtering. Squares. Alexander D. Poularikas. Fundamentals of. Least Mean. with MATLABR. University of Alabama, Huntsville, AL.

Adaptive Filtering. Squares. Alexander D. Poularikas. Fundamentals of. Least Mean. with MATLABR. University of Alabama, Huntsville, AL. Adaptive Filtering Fundamentals of Least Mean Squares with MATLABR Alexander D. Poularikas University of Alabama, Huntsville, AL CRC Press Taylor & Francis Croup Boca Raton London New York CRC Press is

More information

Parametric Signal Modeling and Linear Prediction Theory 1. Discrete-time Stochastic Processes

Parametric Signal Modeling and Linear Prediction Theory 1. Discrete-time Stochastic Processes Parametric Signal Modeling and Linear Prediction Theory 1. Discrete-time Stochastic Processes Electrical & Computer Engineering North Carolina State University Acknowledgment: ECE792-41 slides were adapted

More information

Review of Fundamentals of Digital Signal Processing

Review of Fundamentals of Digital Signal Processing Solution Manual for Theory and Applications of Digital Speech Processing by Lawrence Rabiner and Ronald Schafer Click here to Purchase full Solution Manual at http://solutionmanuals.info Link download

More information

Complement on Digital Spectral Analysis and Optimal Filtering: Theory and Exercises

Complement on Digital Spectral Analysis and Optimal Filtering: Theory and Exercises Complement on Digital Spectral Analysis and Optimal Filtering: Theory and Exercises Random Signals Analysis (MVE136) Mats Viberg Department of Signals and Systems Chalmers University of Technology 412

More information

EE538 Final Exam Fall 2007 Mon, Dec 10, 8-10 am RHPH 127 Dec. 10, Cover Sheet

EE538 Final Exam Fall 2007 Mon, Dec 10, 8-10 am RHPH 127 Dec. 10, Cover Sheet EE538 Final Exam Fall 2007 Mon, Dec 10, 8-10 am RHPH 127 Dec. 10, 2007 Cover Sheet Test Duration: 120 minutes. Open Book but Closed Notes. Calculators allowed!! This test contains five problems. Each of

More information

Digital Signal Processing Lecture 9 - Design of Digital Filters - FIR

Digital Signal Processing Lecture 9 - Design of Digital Filters - FIR Digital Signal Processing - Design of Digital Filters - FIR Electrical Engineering and Computer Science University of Tennessee, Knoxville November 3, 2015 Overview 1 2 3 4 Roadmap Introduction Discrete-time

More information

A=randn(500,100); mu=mean(a); sigma_a=std(a); std_a=sigma_a/sqrt(500); [std(mu) mean(std_a)] % compare standard deviation of means % vs standard error

A=randn(500,100); mu=mean(a); sigma_a=std(a); std_a=sigma_a/sqrt(500); [std(mu) mean(std_a)] % compare standard deviation of means % vs standard error UCSD SIOC 221A: (Gille) 1 Reading: Bendat and Piersol, Ch. 5.2.1 Lecture 10: Recap Last time we looked at the sinc function, windowing, and detrending with an eye to reducing edge effects in our spectra.

More information

Solutions for examination in TSRT78 Digital Signal Processing,

Solutions for examination in TSRT78 Digital Signal Processing, Solutions for examination in TSRT78 Digital Signal Processing, 2014-04-14 1. s(t) is generated by s(t) = 1 w(t), 1 + 0.3q 1 Var(w(t)) = σ 2 w = 2. It is measured as y(t) = s(t) + n(t) where n(t) is white

More information

Review of Fundamentals of Digital Signal Processing

Review of Fundamentals of Digital Signal Processing Chapter 2 Review of Fundamentals of Digital Signal Processing 2.1 (a) This system is not linear (the constant term makes it non linear) but is shift-invariant (b) This system is linear but not shift-invariant

More information

Estimators as Random Variables

Estimators as Random Variables Estimation Theory Overview Properties Bias, Variance, and Mean Square Error Cramér-Rao lower bound Maimum likelihood Consistency Confidence intervals Properties of the mean estimator Introduction Up until

More information

LAB 2: DTFT, DFT, and DFT Spectral Analysis Summer 2011

LAB 2: DTFT, DFT, and DFT Spectral Analysis Summer 2011 University of Illinois at Urbana-Champaign Department of Electrical and Computer Engineering ECE 311: Digital Signal Processing Lab Chandra Radhakrishnan Peter Kairouz LAB 2: DTFT, DFT, and DFT Spectral

More information

Discrete-Time Fourier Transform (DTFT)

Discrete-Time Fourier Transform (DTFT) Discrete-Time Fourier Transform (DTFT) 1 Preliminaries Definition: The Discrete-Time Fourier Transform (DTFT) of a signal x[n] is defined to be X(e jω ) x[n]e jωn. (1) In other words, the DTFT of x[n]

More information

UNIVERSITY OF OSLO. Faculty of mathematics and natural sciences. Forslag til fasit, versjon-01: Problem 1 Signals and systems.

UNIVERSITY OF OSLO. Faculty of mathematics and natural sciences. Forslag til fasit, versjon-01: Problem 1 Signals and systems. UNIVERSITY OF OSLO Faculty of mathematics and natural sciences Examination in INF3470/4470 Digital signal processing Day of examination: December 1th, 016 Examination hours: 14:30 18.30 This problem set

More information

L6: Short-time Fourier analysis and synthesis

L6: Short-time Fourier analysis and synthesis L6: Short-time Fourier analysis and synthesis Overview Analysis: Fourier-transform view Analysis: filtering view Synthesis: filter bank summation (FBS) method Synthesis: overlap-add (OLA) method STFT magnitude

More information

Stochastic Processes. M. Sami Fadali Professor of Electrical Engineering University of Nevada, Reno

Stochastic Processes. M. Sami Fadali Professor of Electrical Engineering University of Nevada, Reno Stochastic Processes M. Sami Fadali Professor of Electrical Engineering University of Nevada, Reno 1 Outline Stochastic (random) processes. Autocorrelation. Crosscorrelation. Spectral density function.

More information

Summary notes for EQ2300 Digital Signal Processing

Summary notes for EQ2300 Digital Signal Processing Summary notes for EQ3 Digital Signal Processing allowed aid for final exams during 6 Joakim Jaldén, 6-- Prerequisites The DFT and the FFT. The discrete Fourier transform The discrete Fourier transform

More information

Lecture 14: Windowing

Lecture 14: Windowing Lecture 14: Windowing ECE 401: Signal and Image Analysis University of Illinois 3/29/2017 1 DTFT Review 2 Windowing 3 Practical Windows Outline 1 DTFT Review 2 Windowing 3 Practical Windows DTFT Review

More information

where r n = dn+1 x(t)

where r n = dn+1 x(t) Random Variables Overview Probability Random variables Transforms of pdfs Moments and cumulants Useful distributions Random vectors Linear transformations of random vectors The multivariate normal distribution

More information

Discrete-Time Fourier Transform

Discrete-Time Fourier Transform Discrete-Time Fourier Transform Chapter Intended Learning Outcomes: (i) (ii) (iii) Represent discrete-time signals using discrete-time Fourier transform Understand the properties of discrete-time Fourier

More information

! Spectral Analysis with DFT. ! Windowing. ! Effect of zero-padding. ! Time-dependent Fourier transform. " Aka short-time Fourier transform

! Spectral Analysis with DFT. ! Windowing. ! Effect of zero-padding. ! Time-dependent Fourier transform.  Aka short-time Fourier transform Lecture Outline ESE 531: Digital Signal Processing Spectral Analysis with DFT Windowing Lec 24: April 18, 2019 Spectral Analysis Effect of zero-padding Time-dependent Fourier transform " Aka short-time

More information

6.003: Signal Processing

6.003: Signal Processing 6.003: Signal Processing Discrete Fourier Transform Discrete Fourier Transform (DFT) Relations to Discrete-Time Fourier Transform (DTFT) Relations to Discrete-Time Fourier Series (DTFS) October 16, 2018

More information

ECGR4124 Digital Signal Processing Final Spring 2009

ECGR4124 Digital Signal Processing Final Spring 2009 ECGR4124 Digital Signal Processing Final Spring 2009 Name: LAST 4 NUMBERS of Student Number: Do NOT begin until told to do so Make sure that you have all pages before starting Open book, 2 sheet front/back

More information

DETECTION theory deals primarily with techniques for

DETECTION theory deals primarily with techniques for ADVANCED SIGNAL PROCESSING SE Optimum Detection of Deterministic and Random Signals Stefan Tertinek Graz University of Technology turtle@sbox.tugraz.at Abstract This paper introduces various methods for

More information

(i) Understanding the characteristics and properties of DTFT

(i) Understanding the characteristics and properties of DTFT Discrete-Time Fourier Transform (DTFT) Chapter Intended Learning Outcomes: (i) Understanding the characteristics and properties of DTFT (ii) Ability to perform discrete-time signal conversion between the

More information

Discrete Time Systems

Discrete Time Systems Discrete Time Systems Valentina Hubeika, Jan Černocký DCGM FIT BUT Brno, {ihubeika,cernocky}@fit.vutbr.cz 1 LTI systems In this course, we work only with linear and time-invariant systems. We talked about

More information

Voiced Speech. Unvoiced Speech

Voiced Speech. Unvoiced Speech Digital Speech Processing Lecture 2 Homomorphic Speech Processing General Discrete-Time Model of Speech Production p [ n] = p[ n] h [ n] Voiced Speech L h [ n] = A g[ n] v[ n] r[ n] V V V p [ n ] = u [

More information

Errata and Notes for

Errata and Notes for Errata and Notes for Statistical and Adaptive Signal Processing Dimitris G. Manolakis, Vinay K. Ingle, and Stephen M. Kogon, Artech Housre, Inc., c 2005, ISBN 1580536107. James McNames March 15, 2006 Errata

More information

Advanced Digital Signal Processing -Introduction

Advanced Digital Signal Processing -Introduction Advanced Digital Signal Processing -Introduction LECTURE-2 1 AP9211- ADVANCED DIGITAL SIGNAL PROCESSING UNIT I DISCRETE RANDOM SIGNAL PROCESSING Discrete Random Processes- Ensemble Averages, Stationary

More information

ECG782: Multidimensional Digital Signal Processing

ECG782: Multidimensional Digital Signal Processing Professor Brendan Morris, SEB 3216, brendan.morris@unlv.edu ECG782: Multidimensional Digital Signal Processing Filtering in the Frequency Domain http://www.ee.unlv.edu/~b1morris/ecg782/ 2 Outline Background

More information

-Digital Signal Processing- FIR Filter Design. Lecture May-16

-Digital Signal Processing- FIR Filter Design. Lecture May-16 -Digital Signal Processing- FIR Filter Design Lecture-17 24-May-16 FIR Filter Design! FIR filters can also be designed from a frequency response specification.! The equivalent sampled impulse response

More information

x(n) = a k x(n k)+w(n) k=1

x(n) = a k x(n k)+w(n) k=1 Autoregressive Models Overview Direct structures Types of estimators Parametric spectral estimation Parametric time-frequency analysis Order selection criteria Lattice structures? Maximum entropy Excitation

More information

Lecture 4: FT Pairs, Random Signals and z-transform

Lecture 4: FT Pairs, Random Signals and z-transform EE518 Digital Signal Processing University of Washington Autumn 2001 Dept. of Electrical Engineering Lecture 4: T Pairs, Rom Signals z-transform Wed., Oct. 10, 2001 Prof: J. Bilmes

More information

ECE-700 Review. Phil Schniter. January 5, x c (t)e jωt dt, x[n]z n, Denoting a transform pair by x[n] X(z), some useful properties are

ECE-700 Review. Phil Schniter. January 5, x c (t)e jωt dt, x[n]z n, Denoting a transform pair by x[n] X(z), some useful properties are ECE-7 Review Phil Schniter January 5, 7 ransforms Using x c (t) to denote a continuous-time signal at time t R, Laplace ransform: X c (s) x c (t)e st dt, s C Continuous-ime Fourier ransform (CF): ote that:

More information

ELEG 3143 Probability & Stochastic Process Ch. 6 Stochastic Process

ELEG 3143 Probability & Stochastic Process Ch. 6 Stochastic Process Department of Electrical Engineering University of Arkansas ELEG 3143 Probability & Stochastic Process Ch. 6 Stochastic Process Dr. Jingxian Wu wuj@uark.edu OUTLINE 2 Definition of stochastic process (random

More information

How to manipulate Frequencies in Discrete-time Domain? Two Main Approaches

How to manipulate Frequencies in Discrete-time Domain? Two Main Approaches How to manipulate Frequencies in Discrete-time Domain? Two Main Approaches Difference Equations (an LTI system) x[n]: input, y[n]: output That is, building a system that maes use of the current and previous

More information

Computer Engineering 4TL4: Digital Signal Processing

Computer Engineering 4TL4: Digital Signal Processing Computer Engineering 4TL4: Digital Signal Processing Day Class Instructor: Dr. I. C. BRUCE Duration of Examination: 3 Hours McMaster University Final Examination December, 2003 This examination paper includes

More information

CCNY. BME I5100: Biomedical Signal Processing. Stochastic Processes. Lucas C. Parra Biomedical Engineering Department City College of New York

CCNY. BME I5100: Biomedical Signal Processing. Stochastic Processes. Lucas C. Parra Biomedical Engineering Department City College of New York BME I5100: Biomedical Signal Processing Stochastic Processes Lucas C. Parra Biomedical Engineering Department CCNY 1 Schedule Week 1: Introduction Linear, stationary, normal - the stuff biology is not

More information

Extended Kalman Filter Derivation Example application: frequency tracking

Extended Kalman Filter Derivation Example application: frequency tracking Extended Kalman Filter Derivation Example application: frequency tracking Nonlinear State Space Model Consider the nonlinear state-space model of a random process x n+ = f n (x n )+g n (x n )u n y n =

More information

EE123 Digital Signal Processing. M. Lustig, EECS UC Berkeley

EE123 Digital Signal Processing. M. Lustig, EECS UC Berkeley EE123 Digital Signal Processing Today Last time: DTFT - Ch 2 Today: Continue DTFT Z-Transform Ch. 3 Properties of the DTFT cont. Time-Freq Shifting/modulation: M. Lustig, EE123 UCB M. Lustig, EE123 UCB

More information

7.17. Determine the z-transform and ROC for the following time signals: Sketch the ROC, poles, and zeros in the z-plane. X(z) = x[n]z n.

7.17. Determine the z-transform and ROC for the following time signals: Sketch the ROC, poles, and zeros in the z-plane. X(z) = x[n]z n. Solutions to Additional Problems 7.7. Determine the -transform and ROC for the following time signals: Sketch the ROC, poles, and eros in the -plane. (a) x[n] δ[n k], k > 0 X() x[n] n n k, 0 Im k multiple

More information

Lecture Notes 7 Stationary Random Processes. Strict-Sense and Wide-Sense Stationarity. Autocorrelation Function of a Stationary Process

Lecture Notes 7 Stationary Random Processes. Strict-Sense and Wide-Sense Stationarity. Autocorrelation Function of a Stationary Process Lecture Notes 7 Stationary Random Processes Strict-Sense and Wide-Sense Stationarity Autocorrelation Function of a Stationary Process Power Spectral Density Continuity and Integration of Random Processes

More information

Assignment 4 Solutions Continuous-Time Fourier Transform

Assignment 4 Solutions Continuous-Time Fourier Transform Assignment 4 Solutions Continuous-Time Fourier Transform ECE 3 Signals and Systems II Version 1.01 Spring 006 1. Properties of complex numbers. Let c 1 α 1 + jβ 1 and c α + jβ be two complex numbers. a.

More information

EEO 401 Digital Signal Processing Prof. Mark Fowler

EEO 401 Digital Signal Processing Prof. Mark Fowler EEO 4 Digital Signal Processing Pro. Mark Fowler ote Set # Using the DFT or Spectral Analysis o Signals Reading Assignment: Sect. 7.4 o Proakis & Manolakis Ch. 6 o Porat s Book /9 Goal o Practical Spectral

More information

Generalizing the DTFT!

Generalizing the DTFT! The Transform Generaliing the DTFT! The forward DTFT is defined by X e jω ( ) = x n e jωn in which n= Ω is discrete-time radian frequency, a real variable. The quantity e jωn is then a complex sinusoid

More information

EEM 409. Random Signals. Problem Set-2: (Power Spectral Density, LTI Systems with Random Inputs) Problem 1: Problem 2:

EEM 409. Random Signals. Problem Set-2: (Power Spectral Density, LTI Systems with Random Inputs) Problem 1: Problem 2: EEM 409 Random Signals Problem Set-2: (Power Spectral Density, LTI Systems with Random Inputs) Problem 1: Consider a random process of the form = + Problem 2: X(t) = b cos(2π t + ), where b is a constant,

More information

Elec4621 Advanced Digital Signal Processing Chapter 11: Time-Frequency Analysis

Elec4621 Advanced Digital Signal Processing Chapter 11: Time-Frequency Analysis Elec461 Advanced Digital Signal Processing Chapter 11: Time-Frequency Analysis Dr. D. S. Taubman May 3, 011 In this last chapter of your notes, we are interested in the problem of nding the instantaneous

More information

Department of Electrical and Computer Engineering Digital Speech Processing Homework No. 6 Solutions

Department of Electrical and Computer Engineering Digital Speech Processing Homework No. 6 Solutions Problem 1 Department of Electrical and Computer Engineering Digital Speech Processing Homework No. 6 Solutions The complex cepstrum, ˆx[n], of a sequence x[n] is the inverse Fourier transform of the complex

More information

EE123 Digital Signal Processing

EE123 Digital Signal Processing EE123 Digital Signal Processing Lecture 2B D. T. Fourier Transform M. Lustig, EECS UC Berkeley Something Fun gotenna http://www.gotenna.com/# Text messaging radio Bluetooth phone interface MURS VHF radio

More information

Discrete-Time Gaussian Fourier Transform Pair, and Generating a Random Process with Gaussian PDF and Power Spectrum Mark A.

Discrete-Time Gaussian Fourier Transform Pair, and Generating a Random Process with Gaussian PDF and Power Spectrum Mark A. Discrete-Time Gaussian Fourier Transform Pair, and Generating a Random Process with Gaussian PDF and Power Spectrum Mark A. Richards October 3, 6 Updated April 5, Gaussian Transform Pair in Continuous

More information

SF2943: TIME SERIES ANALYSIS COMMENTS ON SPECTRAL DENSITIES

SF2943: TIME SERIES ANALYSIS COMMENTS ON SPECTRAL DENSITIES SF2943: TIME SERIES ANALYSIS COMMENTS ON SPECTRAL DENSITIES This document is meant as a complement to Chapter 4 in the textbook, the aim being to get a basic understanding of spectral densities through

More information

TAKEHOME FINAL EXAM e iω e 2iω e iω e 2iω

TAKEHOME FINAL EXAM e iω e 2iω e iω e 2iω ECO 513 Spring 2015 TAKEHOME FINAL EXAM (1) Suppose the univariate stochastic process y is ARMA(2,2) of the following form: y t = 1.6974y t 1.9604y t 2 + ε t 1.6628ε t 1 +.9216ε t 2, (1) where ε is i.i.d.

More information

The Discrete Fourier Transform

The Discrete Fourier Transform In [ ]: cd matlab pwd The Discrete Fourier Transform Scope and Background Reading This session introduces the z-transform which is used in the analysis of discrete time systems. As for the Fourier and

More information

Responses of Digital Filters Chapter Intended Learning Outcomes:

Responses of Digital Filters Chapter Intended Learning Outcomes: Responses of Digital Filters Chapter Intended Learning Outcomes: (i) Understanding the relationships between impulse response, frequency response, difference equation and transfer function in characterizing

More information

DISCRETE-TIME SIGNAL PROCESSING

DISCRETE-TIME SIGNAL PROCESSING THIRD EDITION DISCRETE-TIME SIGNAL PROCESSING ALAN V. OPPENHEIM MASSACHUSETTS INSTITUTE OF TECHNOLOGY RONALD W. SCHÄFER HEWLETT-PACKARD LABORATORIES Upper Saddle River Boston Columbus San Francisco New

More information

System Identification

System Identification System Identification Lecture 4: Transfer function averaging and smoothing Roy Smith 28-- 4. Averaging Multiple estimates Multiple experiments: u r pk, y r pk, r,..., R, and k,..., K. Multiple estimates

More information

INTRODUCTION TO THE DFS AND THE DFT

INTRODUCTION TO THE DFS AND THE DFT ITRODUCTIO TO THE DFS AD THE DFT otes: This brief handout contains in very brief outline form the lecture notes used for a video lecture in a previous year introducing the DFS and the DFT. This material

More information

The Discrete-Time Fourier

The Discrete-Time Fourier Chapter 3 The Discrete-Time Fourier Transform 清大電機系林嘉文 cwlin@ee.nthu.edu.tw 03-5731152 Original PowerPoint slides prepared by S. K. Mitra 3-1-1 Continuous-Time Fourier Transform Definition The CTFT of

More information