E[x 2 ] = E[(an+b) 2 ] = E[a 2 n 2 +2nab+b 2 ] = n 2 E[a 2 ]+2nE[ab]+E[b 2 ] = n 2 E[a 2 ]+E[b 2 ] E[x 2 ] = n 2 σ 2 a+σ 2 b

Σχετικά έγγραφα
w o = R 1 p. (1) R = p =. = 1

Probability and Random Processes (Part II)

Homework 3 Solutions

CHAPTER 25 SOLVING EQUATIONS BY ITERATIVE METHODS

HW 3 Solutions 1. a) I use the auto.arima R function to search over models using AIC and decide on an ARMA(3,1)

HOMEWORK 4 = G. In order to plot the stress versus the stretch we define a normalized stretch:

ECE 308 SIGNALS AND SYSTEMS FALL 2017 Answers to selected problems on prior years examinations

DERIVATION OF MILES EQUATION FOR AN APPLIED FORCE Revision C

Second Order RLC Filters

4.6 Autoregressive Moving Average Model ARMA(1,1)

2 Composition. Invertible Mappings

6.1. Dirac Equation. Hamiltonian. Dirac Eq.

Matrices and Determinants

6.3 Forecasting ARMA processes

Example Sheet 3 Solutions

Module 5. February 14, h 0min

Problem Set 9 Solutions. θ + 1. θ 2 + cotθ ( ) sinθ e iφ is an eigenfunction of the ˆ L 2 operator. / θ 2. φ 2. sin 2 θ φ 2. ( ) = e iφ. = e iφ cosθ.

Statistical Inference I Locally most powerful tests

Jesse Maassen and Mark Lundstrom Purdue University November 25, 2013

ST5224: Advanced Statistical Theory II

Section 8.3 Trigonometric Equations

University of Illinois at Urbana-Champaign ECE 310: Digital Signal Processing

EE512: Error Control Coding

Forced Pendulum Numerical approach

Srednicki Chapter 55

Durbin-Levinson recursive method

Concrete Mathematics Exercises from 30 September 2016

3.4 SUM AND DIFFERENCE FORMULAS. NOTE: cos(α+β) cos α + cos β cos(α-β) cos α -cos β

( ) 2 and compare to M.

Areas and Lengths in Polar Coordinates

Appendix to On the stability of a compressible axisymmetric rotating flow in a pipe. By Z. Rusak & J. H. Lee

Introduction to Time Series Analysis. Lecture 16.

Second Order Partial Differential Equations

DESIGN OF MACHINERY SOLUTION MANUAL h in h 4 0.

Areas and Lengths in Polar Coordinates

CHAPTER 48 APPLICATIONS OF MATRICES AND DETERMINANTS

Απόκριση σε Μοναδιαία Ωστική Δύναμη (Unit Impulse) Απόκριση σε Δυνάμεις Αυθαίρετα Μεταβαλλόμενες με το Χρόνο. Απόστολος Σ.

Πανεπιστήµιο Κρήτης - Τµήµα Επιστήµης Υπολογιστών. ΗΥ-570: Στατιστική Επεξεργασία Σήµατος. ιδάσκων : Α. Μουχτάρης. εύτερη Σειρά Ασκήσεων.

F19MC2 Solutions 9 Complex Analysis

Introduction to the ML Estimation of ARMA processes

Inverse trigonometric functions & General Solution of Trigonometric Equations

Exercises 10. Find a fundamental matrix of the given system of equations. Also find the fundamental matrix Φ(t) satisfying Φ(0) = I. 1.

C.S. 430 Assignment 6, Sample Solutions

SOLUTIONS TO MATH38181 EXTREME VALUES AND FINANCIAL RISK EXAM

Numerical Analysis FMN011

Section 7.6 Double and Half Angle Formulas

Approximation of distance between locations on earth given by latitude and longitude

derivation of the Laplacian from rectangular to spherical coordinates

Homework 8 Model Solution Section

Outline Analog Communications. Lecture 05 Angle Modulation. Instantaneous Frequency and Frequency Deviation. Angle Modulation. Pierluigi SALVO ROSSI

SCHOOL OF MATHEMATICAL SCIENCES G11LMA Linear Mathematics Examination Solutions

Partial Differential Equations in Biology The boundary element method. March 26, 2013

k A = [k, k]( )[a 1, a 2 ] = [ka 1,ka 2 ] 4For the division of two intervals of confidence in R +

The Simply Typed Lambda Calculus

Main source: "Discrete-time systems and computer control" by Α. ΣΚΟΔΡΑΣ ΨΗΦΙΑΚΟΣ ΕΛΕΓΧΟΣ ΔΙΑΛΕΞΗ 4 ΔΙΑΦΑΝΕΙΑ 1

ΗΜΥ 220: ΣΗΜΑΤΑ ΚΑΙ ΣΥΣΤΗΜΑΤΑ Ι Ακαδημαϊκό έτος Εαρινό Εξάμηνο Κατ οίκον εργασία αρ. 2

Other Test Constructions: Likelihood Ratio & Bayes Tests

Chapter 6: Systems of Linear Differential. be continuous functions on the interval

Notes on the Open Economy

Solutions to Exercise Sheet 5

Ordinal Arithmetic: Addition, Multiplication, Exponentiation and Limit

Math221: HW# 1 solutions

Finite Field Problems: Solutions


Econ Spring 2004 Instructor: Prof. Kiefer Solution to Problem set # 5. γ (0)

Quadratic Expressions

Solution Series 9. i=1 x i and i=1 x i.

Differentiation exercise show differential equation

Problem Set 3: Solutions

Math 6 SL Probability Distributions Practice Test Mark Scheme

Ψηφιακή Επεξεργασία Φωνής

= 0.927rad, t = 1.16ms

10.7 Performance of Second-Order System (Unit Step Response)

Lanczos and biorthogonalization methods for eigenvalues and eigenvectors of matrices

Assignment 1 Solutions Complex Sinusoids

SOLUTIONS TO MATH38181 EXTREME VALUES AND FINANCIAL RISK EXAM

Estimation for ARMA Processes with Stable Noise. Matt Calder & Richard A. Davis Colorado State University

CRASH COURSE IN PRECALCULUS

1. A fully continuous 20-payment years, 30-year term life insurance of 2000 is issued to (35). You are given n A 1

CHAPTER 101 FOURIER SERIES FOR PERIODIC FUNCTIONS OF PERIOD

ME 374, System Dynamics Analysis and Design Homework 9: Solution (June 9, 2008) by Jason Frye

Πανεπιστήµιο Κύπρου Πολυτεχνική Σχολή

ECE Spring Prof. David R. Jackson ECE Dept. Notes 2

Variational Wavefunction for the Helium Atom

Derivation of Optical-Bloch Equations

Section 9.2 Polar Equations and Graphs

ANSWERSHEET (TOPIC = DIFFERENTIAL CALCULUS) COLLECTION #2. h 0 h h 0 h h 0 ( ) g k = g 0 + g 1 + g g 2009 =?

( y) Partial Differential Equations

Econ 2110: Fall 2008 Suggested Solutions to Problem Set 8 questions or comments to Dan Fetter 1

= λ 1 1 e. = λ 1 =12. has the properties e 1. e 3,V(Y

Tridiagonal matrices. Gérard MEURANT. October, 2008

Τελικό Project Εργαστηρίου Ηλεκτρονικών Φίλτρων Χειµερινό Εξάµηνο

SCITECH Volume 13, Issue 2 RESEARCH ORGANISATION Published online: March 29, 2018

Strain gauge and rosettes

Reminders: linear functions

Multi-dimensional Central Limit Theorem

A Note on Intuitionistic Fuzzy. Equivalence Relation

EE101: Resonance in RLC circuits

2. THEORY OF EQUATIONS. PREVIOUS EAMCET Bits.

Phys460.nb Solution for the t-dependent Schrodinger s equation How did we find the solution? (not required)

Transcript:

Πανεπιστήµιο Κρήτης - Τµήµα Επιστήµης Υπολογιστών ΗΥ-57: Στατιστική Επεξεργασία Σήµατος 15 ιδάσκων : Α. Μουχτάρης Πρώτη Σειρά Ασκήσεων Λύσεις Ασκηση 1. 1. Since a and b are independent and have zero mean, we have E[a] = and E[b] = and E[ab] =. Therefore, E[x ] = E[(an+b) ] = E[a n +nab+b ] = n E[a ]+ne[ab]+e[b ] = n E[a ]+E[b ] E[x ] = n σ a+σ b. x(n) is not stationary since its variance E[x ] depends on the absolute time n. For this reason it is not ergodic since ergodicity depends on stationarity. Ασκηση. H(z) = 1 3.5z 1 + 6z + 4z 3 1 1 z 1 + 1 H(z) = z3 (1 3.5z 1 + 6z + 4z 3 ) 4 z z 3 (1 1 z 1 + 1 4 z ) H(z) = z3 3.5z + 6z + 4 z 3 1 z + 1 4 z H(z) = z3 3.5z + 6z + 4 z(z 1 z + 1 4 ) Poles: z(z 1 z + 1 4 ) = z 1 = and z 1 z + 1 4 = z =.3536 + j.3536, z 3 =.3536 j.3536 Zeros: z 3 3.5z + 6z + 4 = z 1 =.5, z = + j, z 3 = j. Zeros and poles can be found either using the matlab function roots() or solving algebraically the equations (for the zeros of a 3rd order linear equation see: http://en.wikipedia.org/wiki/cubic_function). So the pole-zero map for H(z) is shown in Fig. 1

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 1.5 1 Imaginary Part.5.5 1 1.5 1.5 1.5.5 1 1.5.5 3 Real Part Figure 1: Pole-zero map for H(z) Magnitude and Phase of H(z)(see Figs and 3) % matlab code for the magnitude and phase of H(z) B=[1, -3.5, 6, 4]; A=[1, -1/sqrt(), 1/4]; fr=-pi:pi/1:pi; plot(fr/pi, *log1(abs(freqz(b, A, fr))));%magnitude plot(fr/pi, angle(freqz(b, A, fr)));%phase

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 3 4 H(z) Magnitude (db) 18 16 14 1 1 1.8.6.4...4.6.8 1 normalized frequency Figure : Magnitude of H(z) 4 H(ω) 3 Phase (rads) 1 1 3 4 1.8.6.4...4.6.8 1 normalized frequency Figure 3: Phase of H(z)

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 4 H(z) as a cascade connection of a minimum phase and an all-pass system. H(z) = z3 3.5z + 6z + 4 z(z 1 z + 1 4 ) H(z) = (z +.5)(z j)(z + j) z(z.3536 j.3536)(z.3536 + j.3536) z +.5 (z j)(z + j) H(z) = z(z.3536 j.3536)(z.3536 + j.3536) 1 z +.5 (z j)(z + j) H(z) = z(z.3536 j.3536)(z.3536 + j.3536) ( 1 4 + j 1 4 )( 1 4 j 1 4 ) ( 1 4 + j 1 4 )(1 4 j 1 4 ) (z +.5)( 1 4 H(z) = + j 1 4 )( 1 4 j 1 4 ) (z j)(z + j) z(z.3536 j.3536)(z.3536 + j.3536) ( 1 4 + j 1 4 )( 1 4 j 1 4 ) where H min (z) = (z +.5)( 1 4 + j 1 4 )( 1 4 j 1 4 ) z(z.3536 j.3536)(z.3536 + j.3536) and H ap (z) = (z j)(z + j) ( 1 4 + j 1 4 )( 1 4 j 1 4 ) Pole-zero plot of H min (z) and H ap (z) (see Figs 4 and 5) % matlab code for Pole-zero plot of H_min(z) and H_ap(z) ZerosHmin=[-.5, 1/4+1j*1/4, 1/4-1j*1/4]; A=[1, -1/sqrt(), 1/4]; zplane(poly(zeroshmin), A); ZerosHap=[+1j*, -1j*]; PolesHap=[1/4+1j*1/4, 1/4-1j*1/4]; zplane(poly(zeroshap), poly(poleshap)); Magnitude and Phase of H min (z)(see Figs 6 and 7) % matlab code for the magnitude and phase of H_min(z) fr=-pi:pi/1:pi; plot(fr/pi, *log1(abs(freqz(poly(zeroshmin), A, fr)))); figure; plot(fr/pi, angle(freqz(poly(zeroshmin), A, fr)))

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 5 Pole zero map of H min 1.8.6.4 Imaginary Part...4.6.8 1 1.5.5 1 Real Part Figure 4: Pole-zero map of H min (z) Pole zero map of H ap 1.5 1.5 Imaginary Part.5 1 1.5 1.5 1.5.5 1 1.5.5 3 Real Part Figure 5: Pole-zero map of H ap (z)

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 6 6 H min (ω) 4 magnitude (db) 4 6 8 1.8.6.4...4.6.8 1 normalized frequency Figure 6: Magnitude of H min (z).8 H min (ω).6.4. Phase (rads)..4.6.8 1.8.6.4...4.6.8 1 normalized frequency Figure 7: Phase of H min (z)

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 7 18.618 H ap (ω) 18.618 18.618 18.618 magnitude (db) 18.618 18.618 18.618 18.618 18.618 18.618 1.8.6.4...4.6.8 1 normalized frequency Figure 8: Magnitude of H ap (z) Magnitude and Phase of H ap (z)(see Figs 8 and 9) % matlab code for the magnitude and phase of H_ap(z) fr=-pi:pi/1:pi; plot(fr/pi, *log1(abs(freqz(poly(zeroshap), poly(poleshap), fr)))); figure; plot(fr/pi, angle(freqz(poly(zeroshap), poly(poleshap), fr)));

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 8 4 H ap (ω) 3 1 Phase (rads) 1 3 4 1.8.6.4...4.6.8 1 normalized frequency Figure 9: Phase of H ap (z) Ασκηση 3. 1. This is an AR process since the input (driving noise) is the signal ɛ(n) and the output y(n) is a linear combination of past output values and the input.. (a) For the first realization: This is an AR system of order M = 1, so from the Yule-Walker equations with M = 1 we get r()w 1 = r (1) w 1 = r (1) r() w 1 =.3785 3.186 w 1 =.9644 ˆα =.9654. Also for the variance of the driving white noise we have (see also Eq..71 Haykin): 1 σɛ = a k r(k) = 1 r() ˆαr(1) = 3.186.9654.3785 σɛ = 1.5764 k= (b) For the second realization: This is an AR system of order M = 1, so from the Yule-Walker equations with M = 1 we get r()w 1 = r (1) w 1 = r (1) r() w 1 =.71 3.671 w 1 =.949 ˆα =.949. Also for the variance of the driving white noise we have (see also Eq..71 Haykin): 1 σɛ = a k r(k) = 1 r() ˆαr(1) = 3.671.949.71 σɛ =.715 k=

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 9 4 First Realization 3 Power Spectrum (db) 1 1 Estimated Power Spectrum True Power Spectrum Periodogram 3 4.5 1 1.5.5 3 3.5 ω (rad) Figure 1: Power spectra for the first realization 3. Ŝ yy (ω) = ˆσ ɛ 1 âe jω, with ˆα =.9654 and σ ɛ = 1.5764 for the first realization and ˆα =.949 and σ ɛ =.715 for the second realization. For the plot see below. 4. The true power spectrum, with {a, σ ɛ } = {.95, 1}. 5. The periodogram spectrum ˆΣ yy (ω) = 1 N S yy (ω) = σ ɛ 1 ae jω, N 1 n= y ne jωn We plot all three spectra for the first realization in Fig. 1 and for the second realization in Fig. 11 6. For the results please refer to part 5. %Matlab code for parts -6

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 1 3 Second Realization 1 Power Spectrum (db) 1 Estimated Power Spectrum True Power Spectrum Periodogram 3 4.5 1 1.5.5 3 3.5 ω (rad) Figure 11: Power spectra for the second realization y1=[3.848, 3.5, 5.55, 4.976, 6.599,6.17,6.57, 6.388, 6.5, 5.564,... 5.683, 5.55, 4.53, 3.95, 3.668,3.668, 3.6, 1.945,.4,.14].'; R1=xcorr(y1, 'biased'); a_hat1=r1(1)/r1(); sigma_square1=r1()-r1(1)/r1()*r1(1); y=[5.431, 5.55, 4.873, 5.1, 5.7, 5.86, 6.133, 5.68, 6.479, 4.31,... 5.181, 4.79, 5.469, 5.87, 3.819,.968,.751, 3.36, 3.13, 3.694].'; R=xcorr(y, 'biased'); a_hat=r(1)/r(); sigma_square=r()-r(1)/r()*r(1); omega=linspace(, pi, ); Spectrum1=sigma_square1./(abs(1-a_hat1*exp(-1i*omega)).ˆ); Spectrum1true=1./(abs(1-.95*exp(-1i*omega)).ˆ);

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 11 n=(:length(y1)-1).'; for iomeg=1:length(omega) Periodogram1(iomeg)=1/(length(y1))*abs(sum(y1.*exp(-1i*omega(iomeg)*n))).ˆ; end Spectrum=sigma_square./(abs(1-a_hat*exp(-1i*omega)).ˆ); Spectrumtrue=1./(abs(1-.95*exp(-1i*omega)).ˆ); for iomeg=1:length(omega) Periodogram(iomeg)=1/(length(y))*abs(sum(y.*exp(-1i*omega(iomeg)*n))).ˆ; end figurea(1,8,4); hold on; box('on'); plot(omega, 1*log1(Spectrum1)); plot(omega, 1*log1(Spectrum1true), 'r'); plot(omega, 1*log1(Periodogram), 'g'); title('first Realization'); xlabel('\omega (rad)'); ylabel('power Spectrum (db)'); legend({'estimated Power Spectrum'; 'True Power Spectrum'; 'Periodogram'}, 'Location', 'Best'); saveas(gcf, 'FirstRealization.fig','fig'); print(gcf,'-depsc',['firstrealization.eps']); hold off; close; figurea(,8,4); hold on; box('on'); plot(omega, 1*log1(Spectrum)); plot(omega, 1*log1(Spectrumtrue), 'r'); plot(omega, 1*log1(Periodogram), 'g');

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 1 3 N=56 Power Spectrum (db) 1 1 Estimated Power Spectrum True Power Spectrum Periodogram 3.5 1 1.5.5 3 3.5 ω (rad) Figure 1: Power spectra for N = 56 samples title('second Realization'); xlabel('\omega (rad)'); ylabel('power Spectrum (db)'); legend({'estimated Power Spectrum'; 'True Power Spectrum'; 'Periodogram'}, 'Location', 'Best'); saveas(gcf, 'SecondRealization.fig','fig'); print(gcf,'-depsc',['secondrealization.eps']); hold off; close; 7. For N = 56 ˆα =.934 and σ ɛ =.9146. For N = 14 ˆα =.959 and σ ɛ =.9499. The corresponding plots can be seen in Fig. 1 and Fig. 13. Ασκηση 4. 1. The filter output is

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 13 3 N=14 Power Spectrum (db) 1 1 Estimated Power Spectrum True Power Spectrum Periodogram 3.5 1 1.5.5 3 3.5 ω (rad) Figure 13: Power spectra for N = 14 samples x(n) = w H u(n), where u(n) is the tap-input vector. The average power of the filter output is therefore E[ x(n) ] = E[w H u(n)u H (n)w] = w H E[u(n)u H (n)]w = w H Rw.. If u(n) is extracted from a zero mean white noise of variance σ, we have R = σ I, where I is the identity matrix. Hence, E[ x(n) ] = σ w H w

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 14 1 u(n) N=3 5 5 v(n) N=3 1 1 3 n (samples) 4 1 3 n (samples) 4 u(n) N=56 1 1 v(n) N=56 5 1 15 5 n (samples) 4 5 1 15 5 n (samples) 4 4 u(n) N=48 v(n) N=48 4 5 1 15 n (samples) 4 5 1 15 n (samples) Figure 14: Driving noise and 4th order AR process for N = {3, 56, 48} samples Ασκηση 5. 1. For all N cases plots of the driving noise and the AR process are shown in Fig. 14. The sample autocorrelation is given by: ˆr u (k) = 1 N N k 1 n= u (n)u(n + k), k =, 1,,..., N 1 For the true autocorrelation sequence, r u (k), from he definition of the autocorrelation and the definition of the 4th order AR process we are given, it holds: r u (k) = E[u(n)u (n k)] = E[(a 1 u(n 1) + a u(n ) + a 3 u(n 3) + a 4 u(n 4) + v(n)) u(n k)] = a 1 E[u(n 1)u(n k)]+a E[u(n )u(n k)]+a 3 E[u(n 3)u(n k)]+a 4 E[u)n 4)u(n k)]+e[v(n)u(n k)] The term E[v(n)u(n k)] is zero for k >, since u(n k) involves samples of the white noise process up to time n k, that is u(n k) is uncorrelated to v(n), so r u (k) becomes:

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 15 r u (k) = a 1 E[u(n 1)u(nk)]+a E[u(n )u(n k)]+a 3 E[u(n 3)u(n k)]+a 4 E[u)n 4)u(n k)] r u (k) = a 1 r u (k 1) + a r u (k ) + a 3 r u (k 3) + a 4 r u (k 4). (1) Since the autocorrelation function is a recurrence relation we have to estimate the initial conditions, i.e., r u (k) for k =, 1,, 3, 4. The autocorrelation at lag k = is equal to : r u () = E[u(n)u (n)] = E[(a 1 u(n 1) + a u(n ) + a 3 u(n 3) + a 4 u(n 4) + v(n)) u(n)] = a 1 r u ( 1) + a r u ( ) + a 3 r u ( 3) + a 4 r u ( 4) + E[v(n)u(n)] = a 1 r u (1) + a r u () + a 3 r u (3) + a 4 r u (4) + E[v(n) (a 1 u(n 1) + a u(n ) + a 3 u(n 3) + a 4 u(n 4) + v(n))] = a 1 r u (1) + a r u () + a 3 r u (3) + a 4 r u (4)+ a 1 E[v(n)u(n 1)] + a E[v(n)u(n )] + a 3 E[v(n)u(n 3)] + a 4 E[v(n)u(n 4)] + E[v(n)v(n)] = a 1 r u (1) + a r u () + a 3 r u (3) + a 4 r u (4) + σv r u () = a 1 r u (1) + a r u () + a 3 r u (3) + a 4 r u (4) + σ v r u () a 1 r u (1) a r u () a 3 r u (3) a 4 r u (4) = σ v, () since terms of the form a k E[v(n)u(n k)], k > are equal to zero. From Eq. (1) The autocorrelation at lag k = 1 is equal to : r u (1) = a 1 r u () + a r u ( 1) + a 3 r u ( ) + a 4 r u ( 3) r u (1) = a 1 r u () + a r u (1) + a 3 r u () + a 4 r u (3) a 1 r u () + (a 1)r u (1) + a 3 r u () + a 4 r u (3) + r u (4) =. (3) The autocorrelation at lag k = is equal to : r u () = a 1 r u (1) + a r u () + a 3 r u ( 1) + a 4 r u ( ) r u () = a 1 r u (1) + a r u () + a 3 r u (1) + a 4 r u () a r u () + (a 1 + a 3 )r u (1) + (a 4 1)r u () + r u (3) + r u (4) =. (4)

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 16 5 4 3 Estimated r u (k) True r u (k) 1 N=3 1 3 4 5 1 15 5 3 35 n (samples) Figure 15: Estimated and true autocorrelation sequence of the 4th order AR process for N = 3 samples The autocorrelation at lag k = 3 is equal to : r u (3) = a 1 r u () + a r u (1) + a 3 r u () + a 4 r u ( 1) r u (3) = a 1 r u () + a r u (1) + a 3 r u () + a 4 r u (1) a 3 r u () + (a + a 4 )r u (1) + a 1 r u () r u (3) + r u (4) =. (5) Finally, the autocorrelation at lag k = 4 is equal to : r u (4) = a 1 r u (3) + a r u () + a 3 r u (1) + a 4 r u () a 4 r u () + a 3 r u (1) + a r u () + a 1 r u (3) r u (4) =. (6) Equations () to (6) form a system of linear equations. Solving the system we get: r u () = 36.861, r u (1) = 5.178, r u () = 3, 7887, r u (3) = 1, 5576 and r u (4) = 3, 6389 and from Eq. (1) r u (k) = a 1 r u (k 1) + a r u (k ) + a 3 r u (k 3) + a 4 r u (k 4), k = 1,,..., N 1. The estimate and true autocorrelation functions for N = {3, 56, 48} can be seen in Fig. 15-17

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 17 4 3 1 N=56 1 3 Estimated r u (k) True r u (k) 4 5 1 15 5 n (samples) Figure 16: Estimated and true autocorrelation sequence of the 4th order AR process for N = 56 samples 4 3 1 N=48 1 3 Estimated r u (k) True r u (k) 4 4 6 8 1 1 14 16 18 n (samples) Figure 17: Estimated and true autocorrelation sequence of the 4th order AR process for N = 48 samples

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 18 1 PSD (db) 3 4 5 Blackman & Tukey PSD Parametric PSD True PSD 6.5 1 1.5.5 3 3.5 ω (rad) Figure 18: Blackman-Tukey, Parametric AR and True Power Spectral Density for the 4th order AR process for N = 3 samples 3. For the Blackman-Tukey method a hamming window of size N/5 samples has been used. All spectra have been normalized to 1 ( db in db scale). We see the comparative plots in Fig. 18- Ασκηση 6. We are given x(n) = v(n) +.75v(n 1) +.5v(n ). Taking the z-transform of both sides: X(z) = (1 +.75z 1 +.5z )V (z). Hence, the transfer function of the MA model is: X(z) V (z) = 1 +.75z 1 +.5z = 1 (1 +.75z 1 +.5z 1. (7) )

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 19 PSD (db) 5 1 15 5 3 35 4 45 Blackman & Tukey PSD Parametric PSD True PSD 5.5 1 1.5.5 3 3.5 ω (rad) Figure 19: Blackman-Tukey, Parametric AR and True Power Spectral Density for the 4th order AR process for N = 56 samples

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 PSD (db) 5 1 15 5 3 35 4 45 Blackman & Tukey PSD Parametric PSD True PSD 5.5 1 1.5.5 3 3.5 ω (rad) Figure : Blackman-Tukey, Parametric AR and True Power Spectral Density for the 4th order AR process for N = 48 samples

ΗΥ-57: Στατιστική Επεξεργασία Σήµατος Πρώτη Σειρά Ασκήσεων Λύσεις 15 1 Using long division, we may perform the following expansion of the denominator in Eq. (7): ( 1 +.75z 1 +.5z ) 1 = 1 3 4 z 1 + 5 16 z 3 64 z 3 11 56 z 4 + 45 14 z 5 91 496 z 6 + 93 1683 z 7 + 85 65536 z 8 67 6144 z 9 + 1541 148576 z 1 +... 1.75z 1 +.315z.469z 3.43z 4 +.439z 5.z 6 +.57z 7 +.13z 8.4z 9 +.15z 1. (8) 1. M= Retaining terms in Eq. (8) up to z, we may approximate the MA model with an AR model of order two as follows: X(z) V (z) 1 1.75z 1 +.315z. Retaining terms in Eq. (8) up to z 5, we obtain the following approximation in the form of an AR model of order five: X(z) V (z) 1 1.75z 1 +.315z.469z 3.43z 4 +.439z 5 3. M=1 Finally, retaining terms in Eq. (8) up to z 1, we obtain the following approximation in the form of an AR model of order ten: X(z) V (z) 1 D(z), where D(z) = 1.75z 1 +.315z.469z 3.43z 4 +.439z 5.z 6 +.57z 7 +.13z 8.4z 9 +.15z 1.