Chapter 7: Spectral Density 7-1 7-2 Introduction Relation of Spectral Density to the Fourier Transform Weiner-Khinchine Relationship 7-3 Properties of Spectral Density 7-4 Spectral Density and the Complex Frequency Plane 7-5 Mean-Square Values From Spectral Density 7-6 Relation of Spectral Density to the Autocorrelation Function 7-7 White Noise Noise Terminology: White Noise, Black Noise, Pink Noise Contour Integration – (Appendix I) 7-8 Cross-Spectral Density 7-9 Autocorrelation Function Estimate of Spectral Density 7-10 Periodogram Estimate of Spectral Density 7-11 Examples and Application of Spectral Density Concepts: Relation of Spectral Density to the Fourier Transform o Weiner-Khinchine Relationship Properties of Spectral Density Spectral Density and the Complex Frequency Plane Mean-Square Values From Spectral Density Relation of Spectral Density to the Autocorrelation Function White Noise, Black Noise, Pink Noise Contour Integration – (Appendix I) Cross-Spectral Density Autocorrelation Function Estimate of Spectral Density Periodogram Estimate of Spectral Density Wiener–Khinchin Theorem For WSS random processes, the autocorrelation function is time based and has a spectral decomposition given by the power spectral density. Also see: http://en.wikipedia.org/wiki/Wiener%E2%80%93Khinchin_theorem Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 1 of 24 ECE 3800 Chapter 7: Spectral Density Section 7-4 and 7-5 Discussion and important information. Setting up Chapter 8 material on linear system signal processing of a random process … ECE 3100 and ECE 3710 stuff … Spectral densities and linear systems Laplace domain Bode plots. The Laplace Domain descriptor “s” and in fact all transfer functions can be defined in terms of a frequency response to a real or complex input signal. As such we often describe the relationship: s j w or w j s A natural question is how do the spectral density plots from autocorrelation relate to the power-spectrum Bode plots of transfer functions. Power Spectral Density (from Autocorrelation): Inherently a power plot. No phase exists. Power Spectral Density (from Bode plots): Described the magnitude squared response of a transfer function or filter. A phase plot in frequency is expected to existed and be performed. Can we define transfer functions based on the autocorrelation’s PSD? Sure … The PSD may be expected to have a polynomial form as: S XX w S 0 w 2n a 2n 2 w 2n 2 a 2n 4 w 2n 4 a 2 w 2 a0 w 2m b2m 2 w 2m 2 b2m 4 w 2m 4 b2 w 2 b0 Substituting for w S XX s S 0 js 2n a 2 n2 js 2n2 a 2 n4 js 2n4 a 2 js 2 a0 js 2 m b2m2 js 2m2 b2 m4 js 2 m4 b2 js 2 b0 n n 1 1 s 2 n a 2 n 2 1 s 2 n 2 a 2 1 s 2 a 0 S XX s S 0 1n s 2m b2 m2 1m1 s 2 m2 b2 1 s 2 b0 This can be expressed as: S XX s H s H s cs c s d s d s Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 2 of 24 ECE 3800 Example: Section 7-4 p. 271. 10 w 2 5 S XX w 4 w 10 w 2 24 S XX s 10 js 5 js 4 2 10 js 24 S XX s 2 10 s 2 5 S XX s 4 s 10 s 2 24 10 s 5 s 5 10 s 5 s 5 2 2 s 4 s 6 s 2 s 2 s 6 s 6 S XX s H s H s 10 s 5 6 s 2 s 6 10 s 5 s 2 s The poles and zeros can be readily identified! 20 Pole-Zero Map 1 0 0.8 0.6 Imaginary Axis (seconds-1) Mag (dB) -20 -40 -60 -80 0.4 0.2 0 -0.2 -0.4 -0.6 -100 -0.8 -120 -1 10 0 10 1 10 Freq (w) 2 10 3 10 -1 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 Real Axis (seconds -1) Poles -2.4495 -2.0000 2.4495 2.0000 psd_sys = -10 s^2 + 50 ----------------s^4 - 10 s^2 + 24 Zeros 2.2361 -2.2361 Continuous-time transfer function. Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 3 of 24 ECE 3800 Example: Section 7-4 p. 272. w 2 w 2 25 S XX w 6 w 33 w 4 463 w 2 7569 S XX s s 2 s 2 25 s 6 33 s 4 463 s 2 7569 The poles and zeros can be found using MATLAB! -20 Pole-Zero Map 6 -30 -40 4 Imaginary Axis (seconds-1) -50 Mag (dB) -60 -70 -80 -90 2 0 -2 -100 -4 -110 -120 -1 10 0 10 1 2 10 Freq (w) 10 -6 -6 3 10 -4 -2 0 2 4 6 Real Axis (seconds -1) Poles -2.0000 + 5.0000i -2.0000 - 5.0000i 2.0000 + 5.0000i 2.0000 - 5.0000i -3.0000 + 0.0000i 3.0000 + 0.0000i psd_sys = -s^4 + 25 s^2 ----------------------------s^6 + 33 s^4 + 463 s^2 - 7569 Continuous-time transfer function. Zeros 0 0 5 -5 H s H s s s 5 s 2 j5 s 2 j5 s 3 s s 5 s s 5 s 2 j5 s 2 j5 s 3 s 2 j5 s 2 j5 s 3 Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 4 of 24 ECE 3800 Continuing Once the Laplace form has been defined, we can compute the Autocorrelation function using the expression S XX s H s H s cs c s d s d s To compute the autocorrelation function for 0 use a partial fraction expansion such that S XX w g s g s d s d s In this structure, all the s-domain poles in the left half-plane appear in the first term and all the poles in the right half-plane appear in the second term. and solve for 1 R XX 2 j g s d s exps ds j for determining 0 , use the RHP expansion, replace –s with s, perform the Laplace transform and replace t with –t. Extension of this concept … Once we have defined S XX s H s H s cs c s d s d s A transfer functions that provide the equivalent power spectrum can be defined as H s cs d s By construction, stability can be guaranteed; all the poles (and zeros) are in the left half-plane. Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 5 of 24 ECE 3800 Example: Find the autocorrelation S XX w A2 1 X 2 X 2 w 2 2 A2 2 w2 Substitute for s for w S XX s 2 A2 2 s2 2 A2 s s Partial fraction expansion k0 k 0 s k1 s k1 2 A2 S XX s s s s s s s k 0 k1 s 0 k 0 k1 2 A 2 S XX s k 0 k1 2k 0 2 A 2 A2 A2 s s Taking the LHP Laplace Transform A2 2 L A exp t s for t 0 Taking the RHP with –s and then –t. A2 A 2 exp t A 2 exp t A 2 expt L s for t 0 Combining we have R XX A 2 exp t Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 6 of 24 ECE 3800 7-5 Deriving the Mean-Square Values from the Power Spectral Density Using the Fourier transform relation between the Autocorrelation and PSD S XX w R XX exp iw d 1 R XX t 2 S XX w expiwt dw The mean squared value of a random process is equal to the 0th lag of the autocorrelation EX 2 1 R XX 0 2 R EX 2 XX 0 1 S XX w expiw 0 dw 2 S XX w dw S XX f expi2f 0 dw S XX f df Therefore, to find the second moment, integrate the PSD over all frequencies. As a note, since the PSD is real and symmetric, the integral can be performed as EX 2 1 R XX 0 2 2 S XX w dw 0 R EX 2 XX 0 2 S XX f df 0 For cases where the PSD is a complicated, the operation can be done in the Laplace Domain, as shown in the textbook. Procedure: (1) Form the PSD in the Laplace Domain. 1n s 2 n a 2 n2 1n1 s 2n2 a 2 1 s 2 a0 S XX s S 0 1n s 2m b2 m2 1m1 s 2 m2 b2 1 s 2 b0 Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 7 of 24 ECE 3800 (2) Factor the PSD in s into the transfer function elements S XX s H s H s cs c s d s d s (3) Perform the following integration (also not easy, based on text approach) 1 1 cs c s S XX s ds ds 2 j 2 j d s d s j X2 j A Table of Integrals has been provided on p. 276 for d(s) polynomials for n=1, 2, 3. An example application appear on p. 275-276 with a second on p. 277-278. One final method – Contour Integrals (Calculus 5? – Complex Variable Theory) A brief summary of the theory can be found in Appendix I. I am going to skip this material! Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 8 of 24 ECE 3800 White Noise Noise is inherently defined as a random process. You may be familiar with “thermal” noise, based on the energy of an atom and the mean-free path that it can travel. As a random process, whenever “white noise” is measured, the values are uncorrelated with each other, not matter how close together the samples are taken in time. Further, we envision “white noise” as containing all spectral content, with no explicit peaks or valleys in the power spectral density. As a result, we define “White Noise” as R XX S 0 t N S XX w S 0 0 2 This is an approximation or simplification because the area of the power spectral density is infinite! Nominally, noise is defined within a bandwidth to describe the power. For example, Thermal noise at the input of a receiver is defined in terms of kT, Boltzmann’s constant times absolute temperature, in terms of Watts/Hz. Thus there is kT Watts of noise power in every Hz of bandwidth. For communications, this is equivalent to –174 dBm/Hz or –144 dBW/Hz. For typical applications, we are interested in Band-Limited White Noise where N0 S 0 S XX w 2 0 f W W f The equivalent noise power is then: R EX 2 XX 0 W S 0 dw 2 W S 0 2 W W N0 N 0 W 2 For communications, we use kTB where W=B and N0=kT. How much noise power, in dBm, would I say that there is in a 1 MHz bandwidth? dBkTB dBkT dBB 174 60 114 dBm Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 9 of 24 ECE 3800 Receiver Sensitivity What does it mean when you buy a radio receiver? For a great receiver (spectrum analyzer grade), assume a 200 kHz FM radio bandwidth. Noise Power kT -174. dBm/Hz Equivalent Noise Bandwidth B 53. dB Hz Receiver Noise Figure NF 10. dB Signal Detection Threshold D 8. dB Minimum Detectable Signal MDS -103. dBm FM radio stations can transmit up to 1 Megawatt +90 dBm Why doesn’t your receiver get blasted? Path loss, distance, higher noise figure, receiving antenna inefficiency, etc. But notice that your commercial; receiver is in microvolts, where 2.0 uV is very good. Power into 50 ohms is V^2/R or 8e-14 W = -130.97 dBW -101 dBm. An MDS of -103 dBm 1.6 uV Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 10 of 24 ECE 3800 Band Limited White Noise S 0 S XX w 0 f W W f The equivalent noise power is then: R EX 2 XX 0 W S 0 dw 2 W S 0 W but what about the autocorrelation? W R XX t S 0 expi2ft df W expi 2ft expi 2Wt exp i 2Wt S0 R XX t S 0 i 2t i 2t i 2t W W R XX t S 0 For sinc xt 2 i sin i 2Wt i 2t xt xt R XX t 2 W S 0 sinc2Wt Using the concept of correlation, for what values will the autocorrelation be zero? (At these delays in time, sampled data would be uncorrelated with previous samples!) 2Wt k k t 2W for k 1, 2, Sampling at 1/2W seems to be a good idea, but isn’t that the Nyquist rate!! Also note, noise passed through a filter becomes band-limited, and the narrower the filter the smaller the noise power … but the wider is the sinc autocorrelation function. Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 11 of 24 ECE 3800 Noise and Filtered Noise Matlab Simulation Based on HW Problem 6-4.6. x=randn(N,1); % zero mean, unit power random signal [b,a] = butter(4,20/500); y=filter(b,a,x); % applying a digital filter y=y/std(y); % normalizing the output power Rxx=xcorr(x)/(N+1); Ryy=xcorr(y)/(N+1); DFTx = fftshift(fft(x))/N; DFTy = fftshift(fft(y))/N; DFTRxx = fftshift(fft(Rxx,2*N))/N; DFTRyy = fftshift(fft(Ryy,2*N))/N; Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 12 of 24 ECE 3800 Sec7_7_White_Pink_Noise Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 13 of 24 ECE 3800 The Cross-Spectral Density Why not form the power spectral response of the cross-correlation function? The Fourier Transform in w S XY w R XY exp iw d and SYX w 1 R XY t 2 RYX exp iw d 1 S XY w expiwt dw and RYX t 2 SYX w expiwt dw Properties of the functions S XY w conjSYX w Since the cross-correlation is real (for X and Y real functions), the real portion of the spectrum is even the imaginary portion of the spectrum is odd There are no other important (assumed) properties to describe Note: The trick or method of using the Laplace transform to form first the positive (poles in LHP) and then the negative portions (poles in RHP) of the “time-based” cross-correlation is required. See example in P. 290-291. In general, a “correlation receiver” can be used when detecting pulses (radar, sonar, etc.) and communication symbols (PSK, FSK, QAM, etc.). Later we will see that his has “matched filter” characteristics … a form of optimal filtering/signal reception. If a signal plus noise is received and the “correlater” contains a copy of the transmitted signal … the “signal” is being auto-correlated (with a time delay) and the noise is being cross-correlated or “filtered” by the transmitted signal. You can estimate the “filtered noise power” if you know the “equivalent noise bandwidth” of the transmitted signal! (An M-FSK correlation was demonstrated previously in class without noise added.) Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 14 of 24 ECE 3800 Autocorrelation Function Estimation of Spectral Density An estimate of the autocorrelation was defined as: 1 Rˆ XX T T xt xt dt 0 An unbiased estimate was given for sample data systems as Rˆ XX kt Rˆ XX k 1 N 1 k N k xi xi k i 0 where it is assumed that values of k<<N are of interest. For the continuous time estimate, we should insure that there is significant information integrated; a general restriction on the estimate may be defined as T 1 xt xt dt Rˆ XX T 0 0 0 m m In fact this can be thought of as windowing the estimate as T 1 xt xt dt ˆ ˆ r R XX wr R XX T 0 0 0 m m where a time window defined as follows is applied 1 wr 0 0 m m Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 15 of 24 ECE 3800 As might be expected we want to take the Fourier Transform of this window autocorrelation function … (remember that multiplication in the time-domain is equivalent to a convolution in frequency-domain) ˆ r S XX w r Rˆ XX exp iw d wr Rˆ XX exp iw d ˆ r S XX w Wr w Sˆ XX w While this function appears to be defined for all w, remember that Wr(w) is present and the estimate of the PSD may be highly inaccurate if sufficient time sample are not available for integration. As a note and another potential problem, the Fourier Transform of the rectangular window function is Wr f wr exp i2f d 2 m sinc2 m f The sinc function possesses negative sidelobes that, when convolved, could cause the estimate of the PSD to become negative. This would violate one of the properties of the PSD! Therefore, the structure of the window function is of importance when performing this estimate. Alternate windows are proposed to “adjust” for the sinc sidelobes Hamming Window Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 16 of 24 ECE 3800 0.54 0.46 cos wham m 0 0 m m Wham f wham wr exp i2f d 0.46 1 1 f Wr f Wham f 0.54 f f 2 2 m 2 m Applying this window function would result in ˆ h S XX w wh wr Rˆ XX exp iw d ˆ ˆ h S XX w Wham w S XX w ˆ w W f 0.54 f 0.23 f 1 f 1 Sˆ w S h XX r XX 2 m 2 m 1 ˆ 1 ˆ ˆ ˆ S XX w f h S XX w Wr f 0.54 S XX w 0.23 S XX w f 2 2 m m 1 ˆ 1 ˆ ˆ ˆ r S XX f h S XX w 0.54 r S XX w 0.23 r S XX f 2 2 m m This function will modify the sidelobes present in the initial estimate to a level that may be acceptable. Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 17 of 24 ECE 3800 Whenever the Fourier transform is applied, the use of a window should be considered. There are entire tables of window functions for FFT’s that show various performance factors! For example: For an exhaustive analysis, see Fred Harris F. J. Harris, “On the use of Windows for harmonic analysis with the Discrete Fourier Transform,” Proc. IEEE, vol. 66, (no. 1), pp. 51–83, Jan 1978. http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=1455106&isnumber=31261 or for a less exhaustive review see Rapuano, S.; Harris, F.J., "An introduction to FFT and time domain windows," Instrumentation & Measurement Magazine, IEEE , vol.10, no.6, pp.32,44, December 2007. http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4428580&isnumber=4428566 Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 18 of 24 ECE 3800 Example: Applying a Hamming Window to the previous example see Sec7_7_FilteredNoise_RxxWin Rxx=xcorr(x)/(N+1); Ryy=xcorr(y)/(N+1); Rxy=xcorr(x,y)/(N+1); Rxx=Rxx.*hamming(2*N-1); Ryy=Ryy.*hamming(2*N-1); Rxy=Rxy.*hamming(2*N-1); Filtered Unfiltered Autocorrelation Autocorrelation 1.2 1.2 Rxx Ryy 1 Rxx Ryy 1 0.8 0.8 Magnitude Magnitude 0.6 0.4 0.6 0.4 0.2 0.2 0 0 -0.2 -0.4 -2 -1.5 -1 -0.5 0 Lag (sec) 0.5 1 1.5 -0.2 -2 2 -1.5 -1 -0.5 DFT of Autocorrelation 0 Lag (sec) 0.5 1 1.5 2 DFT of Autocorrelation 0 0 -10 -10 -20 -20 PSD PSD -30 -30 -40 -40 -50 -50 -60 -60 -70 -500 -70 White Noise Filtered Noise -400 -300 -200 -100 0 100 Frequency 200 300 400 500 -80 -500 White Noise Filtered Noise -400 -300 -200 -100 0 100 Frequency 200 300 400 500 The estimated autocorrelation has been reduced at larger lags. The estimated PSD has been smoothed (not as “noisy”). For a “finite” time estimate (simulation), which one would you prefer working with? Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 19 of 24 ECE 3800 Discrete Implementations The discrete Fourier Transform X D k N 1 WNnk xn n0 1 xˆ k N N 1 WN nk X D k n0 j 2nk where WNnk exp N Equivalence of the continuous and discrete transforms … (1) Discretization of the input x n xn t for t 1 f sample k (2) Spectral bins “actual center freq” X C X C k f X D k N t f where k=0, +/-1, … +/- N/2, and f sample N Get it right at the frequency sample points … based on the applied or actual window used. f sample (3) Spectral bin concept 1 f 2N f sample k XC df X D k N t 2N This is a concept; reality requires that the discrete point consist of the complete “windowed” signal response. (see the Matlab windows). This can be thought of as a piecewise linear estimate of the bin energy. Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 20 of 24 ECE 3800 DFT Symmetry for real numbers X k N 1 X N k WNnk xn n 0 N 1 WNn N k xn n 0 * N 1 WNnN WN nk xn n 0 X N k X k Y k i Y N k i Y N k i N 1 WNnk yn n 0 N 1 WNn N k y n i n 0 N 1 N 1 WNnN WN nk yn n 0 WN nk yn n 0 N 1 Y * N k i WNnk yn n0 * Y N k Y k Notice the spectral symmetries involved. Purely real inputs are conjugate symmetric about N, while imaginary inputs are conjugate anti-symmetric about N. For real input, bins 0 and N/2 are special, while X N 1 X * 1 X N 2 X * 2 N N X 1 X * 1 2 2 Note that k=0 to N/2-1 represent the positive spectrum, while N/2+1 to N-1 represent the negative spectrum. Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 21 of 24 ECE 3800 DFT Symmetry for even symmetric sequences xn x2 N 2 n , n 0 : N 1 x0 x2 N 2 x1 x2 N 3 x N 2 x N x N 1 x N 1 Note that x(N-1) is singular … it is symmetric to itself X k 2 N 2 n 0 X k N 2 W nk 2 N 2 n 0 X k X k xn W nk 2 N 2 xn W2 NN 12k x N 1 N 2 N 2 n 0 n 0 2 N 2 W n N nk 2 N 2 xn W2nkN 2 xn W2k xN 1 W22NN22n k x2 N 2 n N 2 W n 0 nk 2 N 2 X k N 2 xn W2k x N 1 W22NN2 2 k W2Nnk 2 x2 N 2 n n 0 N 2 N 2 n 0 n 0 W2nkN 2 xn W2k xN 1 W2Nnk2 xn X k W N 2 n 0 X k N 2 nk 2 N 2 W2Nnk 2 xn W2k x N 1 nk 2 cos 2 2 N 2 xn 1 k xN 1 n 0 Therefore, X(k) is purely real … no phase components as we would require. Since x(n) is real, we must also have … “conjugate symmetry” but there is no imaginary part! X k X N 2 k Therefore, only k=0 to N-1 must be computed! Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 22 of 24 ECE 3800 Periodograms and Windowing Data We can window the data prior to autocorrelation. this will cause a change in the energy that must be compensated T 1 2 AvgEnergy xt wt dt T 0 T T 1 1 2 2 2 2 E AvgEnergy E xt wt dt E xt wt dt T 0 T 0 For a rectangular window, this becomes 2 E AvgEnergy E xt If an alternate window is used, a power scaling will be necessary to maintain the same estimated energy. For a given window, function we can apply: T scale 1 2 wt dt T 0 So that wscale t wt wt scale T 1 2 wt dt T 0 This also applies to the sampled data computations where wscale n wn wn scale 1 N 1 2 wm N m 0 see Sec7_7_Filtered NoiseWindowed.m Now, multiple smaller results may be processed to generate an average. The spectral resolution (number of frequency bins) necessarily decreases, but now multiple spectra and can be averaged together based on “unique” time sample sets of the data. It is useful for observing time varying phenomena and capturing the values when the desired signal is present. See WaterFallDemo.m for a spectral periodogram …. Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 23 of 24 ECE 3800 Note: Some homework problems to review …. in class 7-6.3 and 7-5.1 Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, 1999. ISBN: 0-19-512354-9. B.J. Bazuin, Spring 2015 24 of 24 ECE 3800