# Basic Deﬁnitions and The Spectral Estimation Problem

Lecture 1

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997

Slide L1–1

Informal Deﬁnition of Spectral Estimation

Given: A ﬁnite record of a signal. Determine: The distribution of signal power over frequency.
signal t ω t=1, 2, ... ω ω+∆ω π spectral density

! = (angular) frequency in radians/(sampling interval) f = !=2 = frequency in cycles/(sampling interval)

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997

Slide L1–2

Applications Temporal Spectral Analysis Vibration monitoring and fault detection Hidden periodicity ﬁnding Speech processing and audio devices Medical diagnosis Seismology and ground movement study Control systems design Radar, Sonar

Spatial Spectral Analysis Source location using sensor arrays
Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997 Slide L1–3

Moses.i!t exists and is called the Discrete-Time Fourier Transform (DTFT) Lecture notes to accompany Introduction to Spectral Analysis by P.1 = t= If: discrete-time deterministic data sequence X 1 t=.1 jy(t)j2 < 1 X Then: Y (! ) = 1 t=. Prentice Hall. Stoica and R. 1997 Slide L1–4 .Deterministic Signals fy(t)g1 .1 y(t)e.

i!k (k) = 1 t=.Energy Spectral Density Parseval's Equality: jy(t)j2 = 21 .1 y(t)y (t . Stoica and R. S (!)d! t=. 1997 . Moses. Prentice Hall.1 X (k)e. k) Slide L1–5 Lecture notes to accompany Introduction to Spectral Analysis by P.1 X Z 1 where 4 S (!) = jY (!)j2 = Energy Spectral Density We can write S (!) = where X 1 k=.

Random Signals Random Signal random signal probabilistic statements about future variations t current observation time Here: X 1 t=. 1997 Slide L1–6 . Prentice Hall.1 n jy(t)j2 = 1 o But: E jy(t)j2 < 1 E f g = Expectation over the ensemble of realizations E jy(t)j2 n o = Average power in y (t) PSD = (Average) power spectral density Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses.

First Deﬁnition of PSD (! ) = where r X 1 k=. Moses. k)g r(k) = r (.1 r(k)e. Stoica and R. Prentice Hall. 1997 Slide L1–7 .i!k (k) is the autocovariance sequence (ACS) r(k) = E fy(t)y (t . (!)ei!k d! n o (Inverse DTFT) Interpretation: so r(0) = E jy(t)j2 = 21 . (!)d! Z (!)d! = inﬁnitesimal signal power in the band ! d! 2 Lecture notes to accompany Introduction to Spectral Analysis by P.k) r(0) jr(k)j Z Note that r(k) = 21 .

Stoica and R.Second Deﬁnition of PSD 1 N y(t)e. Moses. 1997 Slide L1–8 . Prentice Hall.i!t (!) = Nlim E N !1 t=1 X > : 8 > < 2 9 > = > Note that 1 jY (!)j2 (!) = Nlim E N N !1 YN (!) = X where N is the ﬁnite DTFT of fy(t)g. t=1 y(t)e.i!t Lecture notes to accompany Introduction to Spectral Analysis by P.

! ) P3: If y Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. we can restrict attention to !2 .! ) Otherwise: (! ) 6= (. Thus. Then: (! ) = (. 1997 Slide L1–9 . Moses.1=2 1=2] (!) 0 (t) is real.Properties of the PSD P1: (!) = (! + 2 ) for all !. P2: ] () f 2 . Prentice Hall.

k) hk e.1y(t) = y(t . 1) H (q ) y (t) 2 y (! ) = jH (! )j e( e(t) e (! ) Then !) y(t) = X 1 k=0 hk e(t . Moses. 1997 Slide L1–10 .Transfer of PSD Through Linear Systems System Function: where q . Stoica and R.k = unit delay operator: q.1 H (q) = X 1 k=0 hk q.i!k H (! ) = X 1 y (!) = jH (!)j2 e(!) k=0 Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall.

Stoica and R.The Spectral Estimation Problem The Problem: From a sample fy(1) : : : y(N )g (!): f ^(!) ! 2 . 1997 Slide L1–11 . ]g Find an estimate of Two Main Approaches : Nonparametric: – Derived from the PSD deﬁnitions. Parametric: – Assumes a parameterized functional form of the PSD Lecture notes to accompany Introduction to Spectral Analysis by P. Moses. Prentice Hall.

1997 Slide L2–1 . Prentice Hall. Moses.Periodogram and Correlogram Methods Lecture 2 Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R.

Moses. 1997 Slide L2–2 .Periodogram Recall 2nd deﬁnition of (! ): 1 N y(t)e. Prentice Hall. Lecture notes to accompany Introduction to Spectral Analysis by P.i!t N t=1 Natural estimator N 2 Used by Schuster ( 1900) to determine “hidden periodicities” (hence the name).i!t (!) = Nlim E N !1 t=1 X > : 8 > < 2 9 > = > Given : Drop “ fy(t)gN t=1 lim ” and “E N !1 f g” to get X ^p(!) = 1 y(t)e. Stoica and R.

1 X r(k)e.Correlogram Recall 1st deﬁnition of (! ): (! ) = P X 1 Truncate the “ ” and replace “r k=. Moses.i!k ^c(!) = k=.1 N .i!k (k)” by “^(k)”: r r ^(k)e.1) Lecture notes to accompany Introduction to Spectral Analysis by P.(N . Prentice Hall. Stoica and R. 1997 Slide L2–3 .

Covariance Estimators (or Sample Covariances) Standard unbiased estimate: 1 N y(t)y (t . 1997 Slide L2–4 . k) k 0 ^ r(k) = N . Moses. Prentice Hall.k) k < 0 r Lecture notes to accompany Introduction to Spectral Analysis by P. k t=k+1 X Standard biased estimate: 1 N y(t)y (t . k) k 0 r(k) = N ^ t=k+1 X For both estimators: r ^(k) = ^ (. Stoica and R.

(N . Prentice Hall. 1997 . Slide L2–5 Lecture notes to accompany Introduction to Spectral Analysis by P.1 X = ^c(!) k=.i!k ^p(!) = ^c(!) Consequence: Both p ! and ^( ) ^c(!) can be analyzed simultaneously.i!t N t=1 X 2 = N .Relationship Between ^p (! ) and ^c(! ) If: the biased ACS estimator r Then: ^(k) is used in ^c(!).1) r ^(k)e. Moses. Stoica and R. N ^p(!) = 1 y(t)e.

jk =0 there are “so many” that when summed in ! .1 are small. Prentice Hall.Statistical Performance of ^p(! ) and ^c(! ) Summary: Both are asymptotically (for large N ) unbiased: E ^p(!) ! (!) as N ! 1 n o Both have “large” variance. ^p(!) and ^c(!) have poor performance. the PSD error is large. even for large N . Intuitive explanation: r(k) . r(k) may be large for large jkj ^ Even if the errors r k r k Nj. ( )] Lecture notes to accompany Introduction to Spectral Analysis by P. p! f^( ) . ( )g ^ ( ) . 1997 Slide L2–6 . Thus. Moses. Stoica and R.

Slide L2–7 Lecture notes to accompany Introduction to Spectral Analysis by P.i!k k=. ( )WB (! . or triangular.(N . Stoica and R.i!k r k=. Moses. jNj jkj N .Bias Analysis of the Periodogram E ^p(!) = E ^c(!) = E f^(k)g e. N k=.1 jkj r(k)e.(N . Prentice Hall.1 X wB (k) = 8 < : = Thus. ) d Z Ideally: WB (!) = Dirac impulse (!).1 n o n o X ! X N . window E ^p(!) = 21 . 1997 . 1 jkj N Bartlett.1) 1 = wB (k)r(k)e.1) N . n o 0 k 1 .i!k = 1.

Stoica and R. 1997 . (!) may differ quite a bit from (!). WB 1=N . Slide L2–8 Lecture notes to accompany Introduction to Spectral Analysis by P.Bartlett Window WB (! ) 1 sin(!N=2) 2 WB (!) = N sin(!=2) " # 0 WB (!)=WB (0). Prentice Hall. Moses. for N = 25 −10 −20 dB −30 −40 −50 −60 −3 −2 −1 0 1 ANGULAR FREQUENCY 2 3 Main lobe 3dB width For “small” N .

1997 Slide L2–9 .Smearing and Leakage Main Lobe Width: smearing or smoothing Details in resolvable. Prentice Hall. φ(ω) smearing ω ω (!) separated in f by less than 1=N are not ^ φ(ω) <1/Ν Thus: Periodogram resolution limit = 1=N . Stoica and R. ^ φ(ω)≅W (ω) B Sidelobe Level: leakage φ(ω)≅δ(ω) leakage ω ω Lecture notes to accompany Introduction to Spectral Analysis by P. Moses.

1997 Slide L2–10 . so ! is asymptotically unbiased. Moses. Prentice Hall. Stoica and R.Periodogram Bias Properties Summary of Periodogram Bias Properties: For “small” N . severe bias As N . !1 ^( ) ( )! ( ) Lecture notes to accompany Introduction to Spectral Analysis by P. WB ! !.

Moses. Lecture notes to accompany Introduction to Spectral Analysis by P. dev = φ(ω) too - Resolvability properties depend on both bias and variance. 1997 Slide L2–11 . (!2) 2(!1) !1 = !2 = 0 !1 6= !2 nh ih ( io Inconsistent estimate Erratic behavior ^ φ(ω) ^ φ(ω) asymptotic mean = φ(ω) ω + 1 st. Stoica and R.Periodogram Variance As N !1 E ^p(!1) . (!1) ^p(!2) . Prentice Hall.

Stoica and R.i 2 . Moses.i!t 2 k and W = e. 1997 Slide L2–12 . 1 Direct computation of O N 2 ﬂops ( ) .Discrete Fourier Transform (DFT) Finite DTFT: YN (!) = X N t=1 y(t)e.1 fY (k)gN=0 from fy(t)gN : t=1 k Lecture notes to accompany Introduction to Spectral Analysis by P. N Let ! = N Then YN ( 2 k) is the Discrete Fourier Transform (DFT): N X Y (k) = N t=1 y(t)W tk k = 0 ::: N . Prentice Hall.

1 for odd k Assume: X X X Let N For k ~ ~ = N=2 and W = W 2 = e.Radix–2 Fast Fourier Transform (FFT) N = 2m N=2 N Y (k) = y(t)W tk + y(t)W tk t=1 t=N=2+1 N=2 2 = y(t) + y(t + N=2)W Nk ]W tk t=1 Nk +1 for even k with W 2 = .i2 =N . 1997 . Stoica and R. Prentice Hall. ~ 4 = 0 2 4 : : : N . 2 = 2p: Y (2p) = For k X ~ N = 1 3 5 : : : N . y(t + N )]W tgW tp Lecture notes to accompany Introduction to Spectral Analysis by P. Moses. 1 = 2p + 1: X t=1 ~ ~ y(t) + y(t + N )]W tp Y (2p + 1) = Each is a N ~ N ~ = N=2-point DFT computation. Slide L2–13 t=1 ~ ~ f y(t) .

Lecture notes to accompany Introduction to Spectral Analysis by P.1 k2k ) ck = 2 ck = 1 N log2 N 2 Thus. Moses. Stoica and R. Prentice Hall.FFT Computation Count Let ck Then = number of ﬂops for N = 2k point FFT. 1997 Slide L2–14 . 2k + 2c ck = 2 k.

1 ! N k=0 ^ φ(ω) continuous curve Lecture notes to accompany Introduction to Spectral Analysis by P.1 N k=0 sampled. Stoica and R. 1997 Slide L2–15 . Prentice Hall. N=8 ω 2 k N .Zero Padding Append the given data by zeros prior to computing DFT (or FFT): fy(1) : : : y(N ) 0 : : : 0g | N {z } Goals: Apply a radix-2 FFT (so N Finer sampling of = power of 2) ^(!): 2 k N . Moses.

Moses. Stoica and R. Prentice Hall. 1997 Slide L3–1 .Improved Periodogram-Based Methods Lecture 3 Lecture notes to accompany Introduction to Spectral Analysis by P.

Blackman-Tukey Method Basic Idea: Weighted correlogram.1 X k=. with small weight applied to covariances r k with “large” k . Prentice Hall. Moses.(M . 1997 Slide L3–2 .1) w(k)^(k)e. ^( ) jj ^BT (!) = M . Stoica and R.i!k r fw(k)g = Lag Window w(k) 1 -M M k Lecture notes to accompany Introduction to Spectral Analysis by P.

Blackman-Tukey Method. con't ^BT (!) = 1 ^p( )W (! . Moses. )d 2 . Z W (!) = = Conclusion: Effect: DTFT fw(k)g Spectral Window ^BT (!) = “locally” smoothed periodogram Variance decreases substantially Bias increases slightly By proper choice of M : MSE = var + bias2 ! 0 as N ! 1 Slide L3–3 Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall. 1997 . Stoica and R.

Window Design Considerations Nonnegativeness: ^BT (!) = 1 ^p( ) W (! .(M (0) w Z w (k ) = equiv time width 1 2 . 1997 Slide L3–4 . Moses. )d 2 . Stoica and R. W (!)d! = equiv bandwidth e= W (0) Ne e = 1 Lecture notes to accompany Introduction to Spectral Analysis by P.1) Ne = k=. Prentice Hall. 0 Z | {z } If W (!) 0 (.1 X . w(k) is a psd sequence) ^BT (!) 0 (which is desirable) Then: Time-Bandwidth Product M .

Prentice Hall. e) is essentially ) Choose window shape to compromise between smearing (main lobe width) and leakage (sidelobe level).Window Design. ) Choose M as a tradeoff between variance and bias. Lecture notes to accompany Introduction to Spectral Analysis by P. The energy in the main lobe and in the sidelobes cannot be reduced simultaneously. 1997 Slide L3–5 . As M increases. Once M is given. Moses. con't e = 1=Ne = 0(1=M ) is the BT resolution threshold. once M is given. bias decreases and variance increases. Ne (and hence ﬁxed. Stoica and R.

M 0 = 25 −10 −20 dB −30 −40 −50 −60 −3 −2 −1 0 1 ANGULAR FREQUENCY 2 3 Rectangular Window. Prentice Hall. Stoica and R. 1997 Slide L3–6 . M 0 −10 = 25 −20 dB −30 −40 −50 −60 −3 −2 −1 0 1 ANGULAR FREQUENCY 2 3 Lecture notes to accompany Introduction to Spectral Analysis by P. Moses.Window Examples Triangular Window.

Bartlett Method Basic Idea:
1 Μ 2Μ ... Ν

^1(ω) ^2(ω) φ φ

... average

^L(ω) φ

^B(ω) φ
Mathematically:

yj (t) = y((j ; 1)M + t) t = 1 : : : M = the j th subsequence 4 (j = 1 : : : L = N=M ])

^j (!) = 1 yj (t)e;i!t M t=1
X

M

2

1 L ^ (!) ^B (!) = L j =1 j
X

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997

Slide L3–7

Comparison of Bartlett and Blackman-Tukey Estimates

1 L ^B (!) = L j =1
X

8 > < > :

M ;1 X

=

'
Thus:

k=;(M ;1) M ;1 X k=;(M ;1)

k=;(M ;1) 8 M ;1 < 1 X L X
:

r ^j (k)e;i!k
9 =

9 > = >

L j =1

r ^j (k) e;i!k

r ^(k)e;i!k

^B (!) ' ^BT (!)
Since B ! implicitly uses method has

with a rectangular lag window wR k

()

^( )

fwR(k)g, the Bartlett

High resolution (little smearing) Large leakage and relatively large variance
Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997 Slide L3–8

Welch Method Similar to Bartlett method, but allow overlap of subsequences (gives more subsequences, and thus “better” averaging) use data window for each periodogram; gives mainlobe-sidelobe tradeoff capability

1

2 . . . subseq subseq #1 #2 subseq #S

N

Let S # of subsequences of length M . (Overlapping means S > N=M “better averaging”.)

=

])

Additional ﬂexibility: The data in each subsequence are weighted by a temporal window Welch is approximately equal to non-rectangular lag window.

^BT (!) with a
Slide L3–9

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997

1997 Slide L3–10 . Stoica and R. f ^p(!j )g are (nearly) uncorrelated random variables for 2 j N . for N 1. Prentice Hall. Moses. k +J 1 ^D(!k ) = ^p (!j ) 2J + 1 j=k.1 ! = N j =0 Idea: “Local averaging” of (2J + 1) samples in the j (2 + 1) frequency domain should reduce the variance by about J .Daniell Method By a previous result.J X Lecture notes to accompany Introduction to Spectral Analysis by P.

Hence: ^D (! ) ' ^BT (! ) with a rectangular spectral Let Z window. ^D(!) ' 1 ^p(!)d! 2 . Prentice Hall. Stoica and R. con't As J increases: Bias increases (more smoothing) Variance decreases (more averaging) = 2J=N . for N 1. Lecture notes to accompany Introduction to Spectral Analysis by P. 1997 Slide L3–11 . Moses.Daniell Method. Then.

BT periodogram – Local smoothing/averaging of spectral window.Summary of Periodogram Methods Unwindowed periodogram – reasonable bias – unacceptable variance Modiﬁed periodograms – Attempt to reduce the variance at the expense of (slightly) increasing the bias. 1997 . Daniell Periodogram – Approximate interpretation: spectral window. ^BT (!) with a rectangular Slide L3–12 – Implemented by local averaging of periodogram values. Welch periodograms – Approximate interpretation: BT ! with a suitable lag window (rectangular for Bartlett. Moses. Lecture notes to accompany Introduction to Spectral Analysis by P. more general for Welch). Stoica and R. Prentice Hall. ^ () – Implemented by averaging subsample periodograms. ^p(!) by a suitably selected ^(k) using a lag – Implemented by truncating and weighting r window in c ! ^( ) Bartlett.

Parametric Methods for Rational Spectra Lecture 4 Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall. Stoica and R. 1997 Slide L4–1 . Moses.

Prentice Hall. ! can approximate arbitrarily well any continuous PSD. Stoica and R.θ) φ(ω) possibly revise assumption on φ(ω) Rational Spectra jkj m k e. 1997 Slide L4–2 ( ) .i!k (!) is a rational function in e. Moses. however: choice of m and n is not simple some PSDs are not continuous Lecture notes to accompany Introduction to Spectral Analysis by P. P By Weierstrass theorem. Note.θ) Estimate parameters in φ(ω.Basic Idea of Parametric Spectral Estimation Observed Data Assumed functional form of φ(ω. provided m and n are chosen sufﬁciently large.θ) ^ θ Estimate PSD ^ ^ =φ(ω.i!k (!) = P jkj n k e.i! .

n B (z) = 1 + b1z.g. Moses.1 + + bmz. Prentice Hall. e. a rational factored as (!) can be B (!) 2 2 (!) = A(!) A(z) = 1 + a1z. A(! ) = A(z )jz =ei! Signal Modeling Interpretation: e(t) e (! ) = 2 - B (q) A(q) white noise B (!) 2 2 A(!) ltered white noise y(t)y (! ) = ARMA: AR: MA: A(q)y(t) = B (q)e(t) A(q)y(t) = e(t) y(t) = B (q)e(t) Slide L4–3 Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R.m and. 1997 . MA.. and ARMA Models By Spectral Factorization theorem.AR.1 + + anz.

1997 Slide L4–4 . k) and take E f g to give: X r (k ) + X 2 m bj hj .k = j =0 = 0 for k > m B (q) = 1 h q. Stoica and R.ARMA Covariance Structure ARMA signal model: y(t)+ X n i=1 n aiy(t . j ) (b0 = 1) Multiply by y (t . k)g Lecture notes to accompany Introduction to Spectral Analysis by P. j )y (t . i) = X m j =0 bj e(t . Prentice Hall. i) = m j =0 bj E fe(t . Moses.k (h = 1) where H (q ) = A(q ) 0 k k=0 X X i=1 air(k .

r (. . Lecture notes to accompany Introduction to Spectral Analysis by P. an 0 3 7 7 7 5 2 6 6 6 4 Writing covariance equation in matrix form for k : : : n: =1 r(0) r(. . r(n) : : : r(0) 1 = 2 R 0 2 6 6 6 4 7 7 7 5 " # " 32 6 6 6 4 3 7 7 7 5 # These are the Yule–Walker (YW) Equations. 2 1 a1 = 0 . 1997 Slide L4–5 . . .1) . Stoica and R. . r(1) r(0) .1) : : : r(. .AR Signals: Yule-Walker Equations AR: m = 0. Prentice Hall.n) . Moses. .

.AR Spectral Estimation: YW Method Yule-Walker Method: Replace r 2 6 6 6 4 (k) by r(k) and solve for f^ig and ^2: ^ a r ^(0) ^(. ^(. r ^(n) : : : r(0) ^ ^n a 0 32 7 7 7 5 6 6 6 4 3 7 7 7 5 2 6 6 6 4 3 7 7 7 5 Then the PSD estimate is ^2 ^(!) = ^ 2 jA(!)j Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall. r ^(1) r(0) ^ ^. . r . Stoica and R. Moses.1) . . . 1997 Slide L4–6 .1 = 0 a .1) : : : ^(. .n) 1 r r ^2 . .

^ = . 1997 Slide L4–7 . 3 7 5 Y =4 y(n) y(n .(Y Y ). 2) Lecture notes to accompany Introduction to Spectral Analysis by P. . Find i=1 aiy(t . 1) y(n + 1) y(n) . 1) : : : y (t . Stoica and R.1(Y y) where 3 7 5 2 6 t=n+1 je(t)j2 y(1) y(2) y(N . Moses. 1) y(N . . Prentice Hall. n)]T . . . i) = y(t) + 'T (t) = a1 : : : an]T to minimize f( ) = X N This gives 2 y=6 4 y(n + 1) y(n + 2) y(N ) . . n) . y(N .AR Spectral Estimation: LS Method Least Squares Method: e(t) = y(t) + X n 4 = y(t) + y(t) ^ where '(t) = y (t . .

Moses. 0 2 n 3 7 7 7 5 O(n3) ﬂops = 1 : : : n: O(n4) ﬂops Levinson–Durbin Algorithm: Exploits the Toeplitz form of Rn+1 to obtain the solutions for k : : : n in O n2 ﬂops! =1 ( ) Lecture notes to accompany Introduction to Spectral Analysis by P. . . {z . . 1997 Slide L4–8 ...0 .. . Prentice Hall.1 1 . 32 7 7 7 5 } 6 6 6 4 1 n 3 7 7 7 5 2 6 6 6 4 ^ k = either r(k) or r(k). n .Levinson–Durbin Algorithm Fast. Stoica and R.n .. .1 . order-recursive solution to YW equations 2 6 6 6 4 | 0 . . Direct Solution: For one given value of n: For k Rn+1 1 0 = 0 ..

Levinson-Durbin Alg. Prentice Hall. con't Relevant Properties of R: Rx = y \$ Rx = y. . Stoica and R. n 3 7 5 1 Rn+2 4 1 n = 0 3 5 2 4 Rn+1 n+1 r ~n n+1 r ~n 0 32 54 1 n n = 0 0 n 3 5 2 4 2 3 5 where r n = n+1 + ~n n Lecture notes to accompany Introduction to Spectral Analysis by P. Moses. 2 4 Rn+1 n+1 r ~n n+1 r ~n 0 3 5 2 r ~n = 6 4 . 1997 Slide L4–9 . where x = xn : : : x1]T ~ ~ ~ Nested structure 2 Rn+2 = Thus.

Levinson-Durbin Alg. con't 2 Rn+2 4 82 < 4 +2 : 1 n n = 0 0 n 3 5 2 4 2 3 5 2 Rn+2 4 39 = 5 2 0 ~n = 0n 2 1 n 3 5 2 4 3 5 2 3 5 Combining these gives: Rn 0 1 n + kn ~n 0 1 3 5 2 4 = 4 n + kn n 2 0 n + kn 2 n = 4 n 2 +1 3 5 0 0 Thus. Stoica and R. 1997 . Lecture notes to accompany Introduction to Spectral Analysis by P. Moses. 2 kn = . Prentice Hall. jknj2) 2 2 n " n # " ~n 1 # Computation count: k ﬂops for the step k 2 ) n2 ﬂops !k+1 2 to determine f k k gn=1 k Slide L4–10 This is O (n2) times faster than the direct solution. n = n ) n+1 = 0 + kn 2+1 = n + kn n = n(1 .

1997 Slide L4–11 . Moses.i!k (!) = jB(!)j k=. Prentice Hall. r(k) = 0 for jkj > m and 2 2 = m r(k)e.m X Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. m) Thus.MA Signals MA: n =0 y(t) = B (q)e(t) = e(t) + b1e(t . 1) + + bme(t .

Moses. Lecture notes to accompany Introduction to Spectral Analysis by P. Estimate fbkg and 2 and insert them in (!) = jB(!)j2 2 nonlinear estimation problem ^(!) is guaranteed to be 0 2.m ^BT (!) with a rectangular lag window of 2m + 1. Prentice Hall. ^(!) is not guaranteed to be 0 =0 Both methods are special cases of ARMA methods described below. Stoica and R.i!k k=. Insert sample covariances (! ) = This is length X m f^(k)g in: r r(k)e.MA Spectrum Estimation Two main ways to Estimate (! ): 1. with AR model order n . 1997 Slide L4–12 .

k)] g = X n X n j =0 p=0 aj ap r(k + p .i!k k (!) = 2 A(!) jA(!)j2 P where k = E f B(q)e(t)] B(q)e(t . 1997 Slide L4–13 . Moses. Stoica and R.ARMA Signals ARMA models can represent spectra with both peaks (AR part) and valleys (MA part). k)] g = E f A(q)y(t)] A(q)y(t . j ) Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall. A(q)y(t) = B (q)e(t) B (!) 2 = m=.m k e.

^(!) is not guaranteed to be 0 Lecture notes to accompany Introduction to Spectral Analysis by P. can use an approximate linear two-stage least squares method ^(!) is guaranteed to be 0 2.ARMA Spectrum Estimation Two Methods: 2g in (!) = 2 B(!) 2 1. Estimate fai bj A(!) nonlinear estimation problem. Prentice Hall. Estimate fai r(k)g in (!) = P m . 1997 Slide L4–14 .i!k k=. Moses.m k e jA(!)j2 linear estimation problem (the Modiﬁed Yule-Walker method). Stoica and R.

Two-Stage Least-Squares Method Assumption: The ARMA model is invertible: A e(t) = B(q) y(t) (q ) = y(t) + 1y(t . 1) + 2y(t . f kgK=1 by using AR k 1b) Estimate the noise sequence ^(t) = y(t) + ^1y(t . K ) 1a) Estimate the coefﬁcients modelling techniques. Moses. 1) + + K y(t . K t=K +1 X Lecture notes to accompany Introduction to Spectral Analysis by P. K ) ^2 = N 1 j^(t)j2 e N . for some large K e(t) ' y(t) + 1y(t . 2) + = AR(1) with j kj ! 0 as k ! 1 Step 1: Approximate. 1) + e and its variance + ^K y(t . Stoica and R. Prentice Hall. 1997 Slide L4–15 .

1) : : : . m)] e e = a1 : : : an b1 : : : bm]T Lecture notes to accompany Introduction to Spectral Analysis by P. Moses. 1997 Slide L4–16 .y(t . Prentice Hall. n) e ^(t . Stoica and R. y(t . 1) : : : ^(t . Note that the ai and bj coefﬁcients enter linearly in the above equation: y(t) . e A(q)y(t) ' B (q)^(t) e and obtain estimates of fai bj g by applying least squares Step 2: Replace techniques. con't fe(t)g by ^(t) in the ARMA equation. ^(t) ' .Two-Stage Least-Squares Method.

Prentice Hall. # 2 = r(m + M .. . 1) : : : r(m . n + M ) . r f^ g . n + 2) . . 1997 Slide L4–17 . If M > n overdetermined squares solution for ai . = f^ g Y W system of equations. . i) = 0 k > m 3 7 5 " r(m) r(m + 1) . r(m + M ) r(m + 1) r(m + 2) 3 7 If M n.. : : : r(m . Moses. n + 1) r(m . . Stoica and R. 5 . .Modiﬁed Yule-Walker Method ARMA Covariance Equation: r(k) + 2 6 4 X n In matrix form for k i=1 air(k .6 4 Replace fr(k)g by f^(k)g and solve for faig. . = m + 1 ::: m + M a1 an . least f^ g Note: For narrowband ARMA signals. the accuracy of ai is often better for M > n Lecture notes to accompany Introduction to Spectral Analysis by P. . . fast Levinson-type algorithms exist for obtaining ai .

Moses. Prentice Hall.Summary of Parametric Methods for Rational Spectra Method AR: YW or LS low low-medium medium-high medium-high medium low-medium No No Yes Computational Burden low Accuracy medium Guarantee ^(! ) 0 ? Yes MA: BT ARMA: MYW ARMA: 2-Stage LS Use for Spectra with (narrow) peaks but no valley Broadband spectra possibly with valleys but no peaks Spectra with both peaks and (not too deep) valleys As above Lecture notes to accompany Introduction to Spectral Analysis by P. 1997 Slide L4–18 . Stoica and R.

Prentice Hall.Parametric Methods for Line Spectra — Part 1 Lecture 5 Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. 1997 Slide L5–1 . Moses.

Line Spectra Many applications have signals with (near) sinusoidal components. sonar geophysical seismology ARMA model is a poor approximation Better approximation by Discrete/Line Spectrum Models 2 6(2 1) 2 6 (!) 6 (2 2) 2 6 (2 2) 3 ! Slide L5–2 . Examples: communications radar. Moses. Stoica and R. Prentice Hall. 1997 . !1 !2 !3 - An “Ideal” line spectrum Lecture notes to accompany Introduction to Spectral Analysis by P.

1997 . uniformly distributed on .Line Spectral Signal Model Signal Model: Sinusoidal components of frequencies !k and powers 2 . Moses. ] (realistic and mathematically convenient) A3: e(t) = circular white noise with variance 2 E fe(t)e (s)g = 2 t s E fe(t)e(s)g = 0 (can be achieved by “slow” sampling) Slide L5–3 Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. superimposed in white noise of k 2. power f g f g y(t) = x(t) + e(t) t = 1 2 : : : n x(t) = k ei(!k t+ k ) k=1 xk (t) X | {z } Assumptions: A1: A2: k>0 !k 2 . Prentice Hall. ] (prevents model ambiguities) f'k g = independent rv's.

i'j o = 1.i'j o e Z n 1 i' d' 2 = 0. !p) + 2 p Slide L5–4 Lecture notes to accompany Introduction to Spectral Analysis by P. k)g = n=1 2ei!pk + 2 k 0 p p P and (! ) = 2 X n p=1 2 (! . 1997 . Prentice Hall. r(k) = E fy(t)y (t . Stoica and R. for p = j e i'p e. for p 6= j =2 . e E xp(t)xj (t .Covariance Function and PSD Note that: E E n i'p e.i'j o = E nei'p o E ne. Moses. k) = 2 ei!pk p j p n o Hence.

Moses. 1997 Slide L5–5 .Parameter Estimation Estimate either: f!k k 'k gn=1 k 2 (Signal Model) f!k 2gn=1 2 k k Once (PSD Model) ^ Major Estimation Problem: f!k g f!kg are determined: ^ X f ^2g can be obtained by a least squares method from k n 2ei!pk + residuals ^ ^ r(k) = p p=1 OR: Both k and 'k can be derived by a least squares method from f^ g f^ g X y(t) = with n k= k=1 k ei'k . Prentice Hall. Stoica and R. ^ ei!k t + residuals k Lecture notes to accompany Introduction to Spectral Analysis by P.

. Moses. X n k=1 {z F (! ') k ei(!k t+'k ) 2 } = kei'k = 1 : : : n]T Y = y(1) : : : y(N )]T ei!1 ei!n . Stoica and R. Prentice Hall. . . 1997 Slide L5–6 .Nonlinear Least Squares (NLS) Method min' g f! Let: X N k k k t=1 | y(t) . B = eiN!1 eiN!n k 2 6 4 3 7 5 Lecture notes to accompany Introduction to Spectral Analysis by P.

1B Y This gives: ^ = (B B). Moses. con't Then: F = (Y .Nonlinear Least Squares (NLS) Method.1B Y ] +Y Y .1B Y !=^ ! and ! = arg max Y B (B B ). Prentice Hall. Stoica and R. 1997 Slide L5–7 . B k2 = . B ) = kY . (B B).1B Y ] B B] .1B Y ^ ! Lecture notes to accompany Introduction to Spectral Analysis by P. B ) (Y . Y B(B B). (B B).

NLS Properties Excellent Accuracy: 6 2 (for N var (^k ) = 3 2 ! N k Example: N = 300 SNRk = 2= 2 = 30 dB k q 1) Then var (^k ) 10. 1997 Slide L5–8 . it is difﬁcult to avoid convergence to local minima.5. Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. ! Difﬁcult Implementation: The NLS cost function F is multimodal. Moses. Prentice Hall.

Unwindowed Periodogram as an Approximate NLS Method

For a single (complex) sinusoid, the maximum of the unwindowed periodogram is the NLS frequency estimate: Assume: Then:

n=1
X

B B=N
N t=1

BY=

y(t)e;i!t = Y (!)

(ﬁnite DTFT)

1 Y B (B B );1B Y = N jY (!)j2 = ^p(!) = (Unwindowed Periodogram)
So, with no approximation,

! = arg max ^p(!) ^ !

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997

Slide L5–9

Unwindowed Periodogram as an Approximate NLS Method, con't

Assume: Then:

n>1

f!kgn=1 ' the ^ k
provided that

locations of the peaks of p !

^( )

n

largest

inf j!k ; !pj > 2 =N
which is the periodogram resolution limit. If better resolution desired then use a High/Super Resolution method.

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997

Slide L5–10

High-Order Yule-Walker Method Recall:

y(t) = x(t) + e(t) =

k ei(!k t+'k ) +e(t) | {z } k=1 xk (t)
X

n

“Degenerate” ARMA equation for y (t):
n

(1 ; ei!k q;1)xk (t) = k ei(!kt+'k) ; ei!k ei !k(t;1)+'k] = 0
o

Let

B (q) = 1 +

4 bk q;k = A(q)A(q) k=1 A(q) = (1 ; ei!1 q;1) (1 ; ei!n q;1) A(q) = arbitrary
X

L

Then B

(q)x(t) 0 ) B (q)y(t) = B (q)e(t)
Slide L5–11

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997

Moses.High-Order Yule-Walker Method. along with L . Prentice Hall. Stoica and R. 1997 Slide L5–12 . con't Estimation Procedure: Estimate f^igL=1 using an ARMA MYW technique b i ^( ) Roots of B q give “spurious” roots. n ^ k Lecture notes to accompany Introduction to Spectral Analysis by P. f!kgn=1.

. r(L + 2) . Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall. . .High-Order and Overdetermined YW Equations ARMA covariance: r (k ) + X L i=1 bir(k . . Moses. 1997 Slide L5–13 . 1) : : : r(M ) r(L + M ) 4 4 = = {z } | {z r(L) r(L + 1) This is a high-order (if L > n) and overdetermined (if M > L) system of YW equations. i) = 0 k > L In matrix form for k 2 6 6 6 4 | = L + 1 ::: L + M 3 7 7 7 5 2 6 6 6 4 3 7 7 7 5 } : : : r(1) r(L + 1) : : : r(2) b = . Stoica and R. r(L + M . .

Prentice Hall.1U Important property: The additional (L . Lecture notes to accompany Introduction to Spectral Analysis by P. y = . 1997 Slide L5–14 . Stoica and R. if the Minimum-Norm solution b is used. n) spurious zeros of B (q ) are located strictly inside the unit circle. diagonal and nonsingular Thus. Moses.High-Order and Overdetermined YW Equations. (U V )b = . The Minimum-Norm solution is b = .V . con't Fact: SVD of rank : ( )=n =U V U = (M n) with U U = In V = (n L) with V V = In = (n n).

Compute ^ = . Prentice Hall. Stoica and R. V be deﬁned similarly to U . . this approach may give spurious frequency estimates when L > n. this is the price paid for increased accuracy when L > n.V ^ .1U ^ ^ ^ b ^(q) that Then !k n=1 are found from the n zeroes of B k are closest to the unit circle. Practical Solution Let ^= ^ ^ ^ ^ but made from f^(k)g instead of fr(k)g.HOYW Equations. Moses. 1997 Slide L5–15 . f^ g When the SNR is low. r . SVD of . Lecture notes to accompany Introduction to Spectral Analysis by P. V from the Let U .

Prentice Hall.Parametric Methods for Line Spectra — Part 2 Lecture 6 Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. 1997 Slide L6–1 . Moses.

Stoica and R. = Ax(t) + ~(t) ~ e where x(t) = x1(t) : : : xn(t)]T ~ ~(t) = e(t) : : : e(t . m + 1)]T e 4 R = E fy(t)~ (t)g = APA + 2I ~ y Then with 2 P = E fx(t)~ (t)g = ~ x 6 4 2 1 0 .. 1) (for m 3 7 7 7 5 n) 2 46 y(t) = 6 ~ 6 4 y(t .. Prentice Hall. Moses. .i(m.The Covariance Matrix Equation Let: a(!) = 1 e. 1997 Slide L6–2 . m + 1) .i! : : : e.1)! ]T A = a(!1) : : : a(!n)] (m n) rank Note: Deﬁne (A) = n y(t) y(t . 0 2 n 3 7 5 Lecture notes to accompany Introduction to Spectral Analysis by P.

n)) Thus. Prentice Hall. Moses.ng: orthonormal eigenvectors associated with f n+1 : : : m g S = s1 : : : sn] (m n) G = g1 : : : gm.. 2 R = S G] 6 4 1 3 " .Eigendecomposition of R and Its Properties R = APA + 2I (m > n) Let: 1 2 ::: m: eigenvalues of R fs1 : : : sng: orthonormal eigenvectors associated with f 1 : : : ng fg1 : : : gm. Stoica and R.n] (m (m . m 7 5 S G # Lecture notes to accompany Introduction to Spectral Analysis by P.. 1997 Slide L6–3 .

1997 . Stoica and R.1) = AC S = A(PA S ( ) = rank(A) = n). 2 2 6 4 3 7 5 2 6 4 Note: RS = APA S + 2S = S 1 0 . Prentice Hall.. Therefore. 2 . Moses... since S G j j 6= 0 4 . 0 n 3 7 5 with C (since rank S .Eigendecomposition of R and Its Properties. . = = nonsingular 0 n. con't As rank (APA ) = n: k > 2 k = 1 ::: n k = 2 k = n + 1 ::: m 0 1. =0 A G=0 Slide L6–4 Lecture notes to accompany Introduction to Spectral Analysis by P.

Moses.MUSIC Method a (!1) . A G= G=0 a (!n) ) fa(!k )gn=1 ? R(G) k 2 6 4 3 7 5 Thus. Let: 1 N y(t)~ (t) ^ ~ y R = N t=m ^^ S G = S G made from the ^ eigenvectors of R X Lecture notes to accompany Introduction to Spectral Analysis by P. f!kgn=1 are the unique solutions of k a (!)GG a(!) = 0. 1997 Slide L6–5 . . Stoica and R. Prentice Hall.

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. a (!)G ^ Root MUSIC Method: ] f!kgn=1 = the angular positions of the n roots of: ^ k ^^ aT (z. Here. Moses.1)]T Note: Both variants of MUSIC may produce spurious frequency estimates. Prentice Hall. 1997 Slide L6–6 .1 : : : z. a(z) = 1 z.(m.Spectral and Root MUSIC Methods Spectral MUSIC Method: f!kgn=1 = the locations of the n highest peaks of the ^ k “pseudo-spectrum” function: 1 ^G a(!) ! 2 .1)GG a(z) = 0 that are closest to the unit circle.

1)^1 = 0 g no problem with spurious frequency estimates computationally simple (much) less accurate than MUSIC with m n+1 Lecture notes to accompany Introduction to Spectral Analysis by P.Pisarenko Method Pisarenko is a special case of MUSIC with m (the minimum possible value). Moses. 1997 Slide L6–7 . ) f!kgn=1 can be found from the roots of ^ k aT (z. Prentice Hall. Stoica and R. If: =n+1 m=n+1 Then: ^ g G = ^1.

but only one vector in G (as in Pisarenko). R( ) Let " 1 = the vector in R(G). Uses m n (as in MUSIC). Moses. g # Lecture notes to accompany Introduction to Spectral Analysis by P.Min-Norm Method Goals: Reduce computational burden. with ﬁrst element equal ^ ^ to one. that has minimum Euclidean norm. and reduce risk of false frequency estimates. 1997 Slide L6–8 . Stoica and R. Prentice Hall.

Moses.Min-Norm Method.1) 1 ^ g " # that are closest to the unit circle. Stoica and R. 1997 Slide L6–9 . Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall. con't Spectral Min-Norm f!gn=1 = ^k the locations of the n highest peaks in the “pseudo-spectrum” 2 1 = a (!) 1 ^ g " # Root Min-Norm f!gn=1 = ^k the angular positions of the n roots of the polynomial aT (z.

S(S S ). 1997 Slide L6–10 .) ^^ I=S S= Multiplying the above equation by gives: (1 . (S S ).1 exists iff = k k2 6= 1 (This holds. Moses. Stoica and R. Prentice Hall.S =(1 . k k2) ) ^ = . for N 1. g " # Min-Norm solution: As: ^ = .1 gm " # " # Min-Norm Method: Determining g ^ Then: 1 2 R(G) ) S 1 = 0 ^ ^ ^ ^ g g ) S ^= .1 g + S S .Let S ^= S g1 .1 = =(1 . at least. k k2) g Lecture notes to accompany Introduction to Spectral Analysis by P. k k2) = (S S ) ) (S S ).

Stoica and R. Then S2 = A2C = A1DC = S1 C . where e. Moses. D= 0 e.1S1S2 Lecture notes to accompany Introduction to Spectral Analysis by P.i!n S1 = Im..1]S 2 6 4 | 3 7 5 Recall S = AC with jC j 6= 0.1 0]A A2 = 0 Im.i!1 0 ..1 0]S S2 = 0 Im. let = A1D.1DC {z } So has the same eigenvalues as D . 1997 Slide L6–11 . Prentice Hall.ESPRIT Method Let A1 = Im.1]A Then A2 Also. determined as is uniquely = (S1S1).

1S1S2 ^^ ^^ f!kgn=1 = . Stoica and R. Prentice Hall. then S1 and S2 . ^ ^ ^ ^ The frequency estimates are found by: where f^kgn=1 are the eigenvalues of k ^ = (S1S1). ﬁnd S . 1997 Slide L6–12 .ESPRIT Implementation From the eigendecomposition of R. Moses. arg(^k ) ^ k ESPRIT Advantages: computationally simple no extraneous frequency estimates (unlike in MUSIC or Min–Norm) accurate frequency estimates Lecture notes to accompany Introduction to Spectral Analysis by P.

Prentice Hall. Moses.Summary of Frequency Estimation Methods Method Periodogram Nonlinear LS Yule-Walker Pisarenko MUSIC Min-Norm ESPRIT Computational Accuracy / Risk for False Burden Resolution Freq Estimates small medium-high medium very high very high very high medium high medium small low none high high medium medium high small medium very high none Recommendation: Use Periodogram for medium-resolution applications Use ESPRIT for high-resolution applications Slide L6–13 Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. 1997 .

1997 Slide L7–1 . Moses.Filter Bank Methods Lecture 7 Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall. Stoica and R.

Nonparametric Approach: Implicitly smooth ! !=. but 2 requires to ensure that the number of estimated values = = ) is < N . Parametric Approach: Parameterize ﬁnite-dimensional model. 1997 Slide L7–2 . by assuming that ! is nearly constant over the bands f ( )g ( ) !. Stoica and R.Basic Ideas Two main PSD estimation approaches: 1. Moses. Prentice Hall. (!) by a 2. !+ ] N >1 1 2 is more general than 1. ( =2 2 =1 N > 1 leads to the variability / resolution compromise associated with all nonparametric methods. Lecture notes to accompany Introduction to Spectral Analysis by P.

Slide L7–3 Lecture notes to accompany Introduction to Spectral Analysis by P. are conﬂicting. c ( )d = (') (!) (a) consistent power calculation (b) Ideal passband ﬁlter with bandwidth (c) ( ) constant on 2 ! . Z 6 ~ (! ) @ -! (b) Z !c ^FB (!) ' 1 2 (a) .with varying ! Power Calculation Bandpass Filter c and xed bandwidth - Power in the band - Division by lter bandwidth ^(!c ) - 6 y y (! ) . !c @Q Q -! . as well as (b) and (c). Stoica and R. jH ( )j2 ( )d = ' 21 !+ !. Moses. Prentice Hall.Filter Bank Approach to Spectral Estimation H 6 (!) 1 !c -! Filtered Signal y (t) . 2 ! + 2 ] Note that assumptions (a) and (b). 1997 .

!) . Moses.t) t=1 X X = N where ( X 1 k=0 hk y(N .!) . Stoica and R. Prentice Hall.i!t 2 ~ ^p(~) = ! 4 N t=1 2 N 1 ~ = N y(t)ei!(N . 1 H (!) = hk e ! N ei(~. 1 ~ hk = N otherwise 0 1 center frequency of H 3dB bandwidth of H (!) = ! ~ (!) ' 1=N Slide L7–4 Lecture notes to accompany Introduction to Spectral Analysis by P.Filter Bank Interpretation of the Periodogram 1 N y(t)e.i!k = 1 eiN (~. 1 k=0 X 1 ei!k k = 0 : : : N . 1997 . k) 2 ! .

1997 Slide L7–5 . ^( ) Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. ! 0 −5 −10 −15 dB −20 −25 −30 −35 −40 −3 −2 −1 0 1 ANGULAR FREQUENCY 2 3 Conclusion: The periodogram p ! is a ﬁlter bank PSD estimator with bandpass ﬁlter as given above. Prentice Hall. for N = 50. Moses.Filter Bank Interpretation of the Periodogram. and: narrow ﬁlter passband. con't jH (!)j as a function of (~ . power calculation from only 1 sample of ﬁlter output. !).

Possible Improvements to the Filter Bank Approach 1. Both approaches compromise bias for variance. Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Each ﬁlter covers a small band centered on ! . Prentice Hall. and in fact are quite related to each other: splitting the data sample can be interpreted as a special form of windowing or ﬁltering. 2. Moses. Use several bandpass ﬁlters on the whole sample. Split the available sample. and bandpass ﬁlter each subsample. 1997 Slide L7–6 . This approach leads to Bartlett and Welch methods. This “multiwindow approach” is similar to the Daniell method. ~ provides several samples for power calculation. more data points for the power calculation stage.

im! ]T hk e.i!k = h a(!) Lecture notes to accompany Introduction to Spectral Analysis by P.i! : : : e. yF (t) = X m k=0 | hk y(t . Moses. k) 2 6 4 | . = h0 h1 : : : hm] . m) h y(t) ~ E jyF (t)j2 = h Rh R = E fy(t)~ (t)g ~ y {z } 7 5 {z } n o y(t) 3 H (! ) = where a X m (! ) = 1 k=0 e. Prentice Hall. 1997 Slide L7–7 . y(t .Capon Method Idea: Data-dependent bandpass ﬁlter design. Stoica and R.

con't Capon Filter Design Problem: min(h Rh) h Solution: subject to h a(!) = 1 h0 = R.1a=a R. Stoica and R.Capon Method. n The Bandwidth 1 ' m+1 = (ﬁlter 1 length) Conclusion Estimate PSD as: +1 ^(!) = m^.1a o The power at the ﬁlter output is: E jyF (t)j2 = h0Rh0 = 1=a (!)R. 1997 Slide L7–8 .1 a (!)R a(!) with N 1 ^ y(t)~ (t) ~ y R = N . Prentice Hall.m t=m+1 X Lecture notes to accompany Introduction to Spectral Analysis by P. Moses.1a(!) which should be the power of y (t) in a passband centered on ! .

. 1997 Slide L7–9 . bias decreases and variance increases. but it splits the N -data point sample into (N m) subsequences of length m with maximum overlap. Moses. Stoica and R.Capon Properties m is the user parameter that controls the compromise between bias and variance: – as m increases. Prentice Hall. Capon uses one bandpass ﬁlter only. Lecture notes to accompany Introduction to Spectral Analysis by P.

Relation between Capon and Blackman-Tukey Methods Consider ^BT (!) with Bartlett window: m m + 1 . Stoica and R. Prentice Hall.m m + 1 1 m m ^(t . Moses. jkj ^BT (!) = r(k)e. s)e.s) = m+1 r t=0 s=0 ^ a (!)Ra(!) R = ^(i .1a(!) Lecture notes to accompany Introduction to Spectral Analysis by P. 1997 Slide L7–10 .i!k ^ k=.i!(t. j )] ^ r = m+1 X X X Then we have ^ a (!)Ra(!) ^BT (!) = m+1 m+1 ^C (!) = ^ a (!)R.

AR Then ^C (!) = 1 1 m 1= ^AR(!) m + 1 k=0 k X Consequences: Due to the average over k. Moses. C ! generally has worse resolution and bias properties than the AR method. Stoica and R. C ! generally has less statistical variability than the AR PSD estimator. Due to the low-order AR terms in the average. 1997 Slide L7–11 . Prentice Hall.Relation between Capon and AR Methods Let 2 ^k ^k (!) = ^ 2 jAk(!)j be the kth order AR PSD estimate of y (t). ^( ) ^( ) Lecture notes to accompany Introduction to Spectral Analysis by P.

Moses.Spatial Methods — Part 1 Lecture 8 Lecture notes to accompany Introduction to Spectral Analysis by P. 1997 Slide L8–1 . Stoica and R. Prentice Hall.

communications. mechanical Receiving sensors: Hydrophones. @.. electromagnetic. v Source n @. antennas. sonar. Sensor 1 @. Emitted energy: Acoustic.The Spatial Spectral Estimation Problem Source 1 vSource 2 B B B B N B cccc ccc c c c c c v @ . 1997 Slide L8–2 . . @. Sensor 2 Sensor m Problem: Detect and locate n radiating sources by using an array of m passive sensors. Prentice Hall. @. . underwater surveillance Basic Approach: Determine energy distribution over space (thus the name “spatial spectral analysis”) Lecture notes to accompany Introduction to Spectral Analysis by P. .@ . @. Moses. seismometers Applications: Radar. Stoica and R.. . seismology.R .@ .. @.

Moses. Stoica and R. the array is calibrated. Prentice Hall.Simplifying Assumptions Far-ﬁeld sources in the same plane as the array of sensors Non-dispersive wave propagation Hence: The waves are planar and the only location parameter is direction of arrival (DOA) (or angle of arrival. The number of sources n is known. AOA).) Lecture notes to accompany Introduction to Spectral Analysis by P. (We do not treat the detection problem) The sensors are linear dynamic elements with known transfer characteristics and known locations (That is. 1997 Slide L8–3 .

Array Model — Single Emitter Case x(t) = k= the signal waveform as measured at a reference point (e. Lecture notes to accompany Introduction to Spectral Analysis by P.. thermal noise in sensor electronics. at the “ﬁrst” sensor) the delay between the reference point and the kth sensor the impulse response (weighting function) of sensor k “noise” at the kth sensor (e.) hk (t) = ek (t) = Note: t 2 R (continuous-time signals).g. etc. Then the output of sensor k is yk (t) = hk (t) x(t . () f kg with hk(t) This is a time-delay estimation problem in the unknown input case. 1997 Slide L8–4 . k) + ek (t) ( = convolution operator).g. Basic Problem: Estimate the time delays known but x t unknown. background noise. Prentice Hall. Moses. Stoica and R..

!c k ] hk (t) x(t . Then: x(t) = (t) cos !ct + '(t)] where (t) '(t) vary “slowly enough” so that (t . !c k + argfHk (!c)g] where Hk function (!) = Ffhk(t)g is the kth sensor's transfer Hence. Prentice Hall. 1997 . k ) ' jHk (!c)j (t)cos !ct + '(t) . Moses. the kth sensor output is yk (t) = jHk (!c)j (t) cos !ct + '(t) . k ) ' (t) cos !ct + '(t) . Stoica and R. !c k + arg Hk (!c)] + ek(t) Slide L8–5 Lecture notes to accompany Introduction to Spectral Analysis by P. k ) ' '(t) Time delay is now ' to a phase shift !c k : x(t . k) ' (t) '(t .Narrowband Assumption Assume: The emitted signals are narrowband with known carrier frequency !c .

by low-pass ﬁltering and sampling the signal yk (t)=2 = yk(t)e.i 2!ct+ (t)] g n o | {z } | {z } Lowpass ﬁlter 2z(t)e. Lecture notes to accompany Introduction to Spectral Analysis by P.i !ct+ (t)] 2 Demodulate z (t) (translate to baseband): 2z(t)e.i!c k + ek (t) where s(t) is the complex envelope of x(t). Stoica and R.i!ct ~ = yk(t) cos(!ct) .!ct = (t)f ei (t) + e. 1997 Slide L8–6 or . Prentice Hall. Moses.Complex Signal Representation The noise-free output has the form: z(t) = (t) cos !ct + (t)] = = (t) ei !ct+ (t)] + e.i! + ek (t) | s(t) {z } | Hk (!c) {z k c } c k yk (t) = s(t)Hk (!c) e. iyk(t) sin(!ct) we get the complex representation: (for t 2 Z ) yk (t) = (t) ei'(t) jHk (!c)j ei arg H (! )] e.i!ct to obtain (t)ei (t) lowpass highpass Hence.

Vector Representation for a Narrowband Source

Let

= m =
a( ) =
2 6 4

the emitter DOA the number of sensors
2 6 4

H1(!c) e;i!c
. .

1

3 7 5

Hm(!c) e;i!c m y1(t) e1(t) . . y(t) = e(t) = . . ym(t) em(t)
3 7 5 2 6 4

3 7 5

Then

y(t) = a( )s(t) + e(t)
NOTE: enters a via both For omnidirectional sensors the on .

()

f kg and fHk(!c)g. fHk(!c)g do not depend

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997

Slide L8–7

Analog Processing Block Diagram

Analog processing for each receiving array element
SENSOR DEMODULATION

A/D CONVERSION

-

-

Internal Noise External Noise XXXXX ? XX ll z l - Cabling and , Filtering : ,, Incoming Transducer Signal x(t ; k )

Re yk (t)] Lowpass Re yk (t)] - !! ~ Filter A/D cos(!ct) 6 y k (t) Oscillator -sin(!ct) ? ~ - Im yk (t)] Lowpass Im yk (t)] - !! Filter A/D

-

Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997

Slide L8–8

Multiple Emitter Case Given n emitters with received signals: DOAs:

fsk(t)gn=1 k

k

Linear sensors

) y(t) = a( 1)s1(t) +

+ a( n)sn(t) + e(t)

Let

A = a( 1) : : : a( n)] (m n) s(t) = s1(t) : : : sn(t)]T (n 1)
Then, the array equation is:

y(t) = As(t) + e(t)
Use the planar wave assumption to ﬁnd the dependence of k on .
Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. Moses, Prentice Hall, 1997 Slide L8–9

1) c Lecture notes to accompany Introduction to Spectral Analysis by P.Uniform Linear Arrays Source h JJ J J J J JJ J ^ J J JJ JJ+ JJ JJ 2 ] J d sin J ^ JJ J JJ 3 J J v 4 1 v ]? v d - v ppp m vSensor ULA Geometry Sensor #1 = time delay reference Time Delay for sensor k: where c = wave propagation speed d sin k = (k . Moses. Stoica and R. 1997 Slide L8–10 . Prentice Hall.

Prentice Hall.i!s : : : e. !s = spatial frequency The function !s 7! a( ) is one-to-one for j!sj \$ dj sin j 1 d =2 As =2 d =2 is a spatial Shannon sampling theorem. Slide L8–11 d = spatial sampling period Lecture notes to accompany Introduction to Spectral Analysis by P. Moses. Stoica and R.1)!s ]T By direct analogy with the vector a(! ) made from uniform samples of a sinusoidal time series.Spatial Frequency Let: sin 4 !s = !c d sin = 2 dc=f = 2 d sin c c = c=fc = signal wavelength a( ) = 1 e. 1997 .i(m.

Prentice Hall. Stoica and R. 1997 Slide L9–1 .Spatial Methods — Part 2 Lecture 9 Lecture notes to accompany Introduction to Spectral Analysis by P. Moses.

Stoica and R. Moses. Prentice Hall. Lecture notes to accompany Introduction to Spectral Analysis by P. 1997 Slide L9–2 .Spatial Filtering Spatial ﬁltering useful for DOA discrimination (similar to frequency discrimination of time-series ﬁltering) Nonparametric DOA estimation There is a strong analogy between temporal ﬁltering and spatial ﬁltering.

1)! ]T We can select h to enhance or attenuate signals with different frequencies ! .1] y(t) = u(t) : : : u(t . 1997 Slide L9–3 . m + 1)]T If u m. Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall.i(m. Moses.i! : : : e.1 X (t) = ei!t then yF (t) = h a(!)] u(t) | {z } ﬁlter transfer function a(!) = 1 e.Analogy between Temporal and Spatial Filtering Temporal FIR Filter: yF (t) = hk u(t . k) = h y(t) k=0 h = ho : : : hm. Stoica and R.

i!c m ]T yF (t) = h a( )] s(t) | {z } The corresponding ﬁlter output is ﬁlter transfer function We can select h to enhance or attenuate signals coming from different DOAs. Moses. Lecture notes to accompany Introduction to Spectral Analysis by P. Spatial FIR Filter output: yF (t) = X m k=1 hk yk (t) = h y(t) Narrowband Wavefront: The array's (noise-free) response to a narrowband ( sinusoidal) wavefront with complex envelope s t is: () y(t) = a( )s(t) a( ) = 1 e. 1997 Slide L9–4 .i!c 2 : : : e. Prentice Hall.Analogy between Temporal and Spatial Filtering Spatial Filter: fyk(t)gm=1 = k the “spatial samples” obtained with a sensor array. Stoica and R.

.@ 0 B 1 C h @@ C B B e. ? qq 1 ! (Spatial sampling) (b) Spatial ﬁlter Lecture notes to accompany Introduction to Spectral Analysis by P.? B -P C B C B B B .i! C @ h a(!)]u(t) R @ C .1 q .. Prentice Hall. s qq h - ? 1 ? - m.i! m | e {z a( ) aa m! point ( ) qq ( ) 1 C C C C C Cs(t) C C C C C A } - ? @ ? @ @ h - 1 @ R @ - P h a( )]s(t) - hm. 1997 Slide L9–5 .1 ? - 1 . Cu(t) C B . C C B C B C B C B C B A @ .i! Be 2 B !! B B B B .Analogy between Temporal and Spatial Filtering u(t) = ei!t z 1 s q. ! hm. B . B B @ .B . Moses.i m. | e {z } ? a(!) 1 ? 0 qq ( 1) 1 (Temporal sampling) (a) Temporal ﬁlter narrowband source with DOA= Z z Z Z Z ~ aa 1reference !! h 0 0 1 B aa 2. Stoica and R.

h a 0 .Spatial Filtering. Moses. where 0 j ( )j = ( ) = 25 0 30 −30 Th et a (d eg ) 60 −60 90 0 2 4 6 Magnitude 8 10 −90 Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall. 1997 Slide L9–6 . Stoica and R. con't Example: The response magnitude h a of a spatial ﬁlter (or beamformer) for a 10-element ULA. Here.

Prentice Hall. To locate an emitter in the ﬁeld of view. hence ﬁltering out interferences located outside the ﬁlter's beam (but possibly having the same temporal characteristics as the signal). 1997 Slide L9–7 . Stoica and R. Moses. Lecture notes to accompany Introduction to Spectral Analysis by P. by sweeping the ﬁlter through the DOA range of interest (“goniometer”).Spatial Filtering Uses Spatial Filters can be used To pass the signal of interest only.

Nonparametric Spatial Methods A Filter Bank Approach to DOA estimation. Prentice Hall. 1997 . Basic Ideas Design a ﬁlter h ( ) such that for each – It passes undistorted the signal with DOA = – It attenuates all DOAs 6= Sweep the ﬁlter through the DOA range of interest. Stoica and R. n o n o The (dominant) peaks of h DOAs of the sources. ( )Rh( ) give the Slide L9–8 Lecture notes to accompany Introduction to Spectral Analysis by P. Moses. and evaluate the powers of the ﬁltered signals: E jyF (t)j2 = E jh ( )y(t)j2 = h ( )Rh( ) with R = E fy (t)y (t)g.

y t can be considered as impinging on the array from all DOAs. Moses.Beamforming Method Assume the array output is spatially white: Then: R = E fy(t)y (t)g = I E jyF (t)j2 = h h n o Hence: In direct analogy with the temporally white assumption for ﬁlter bank methods. () Filter Design: min (h h) h Solution: subject to h a( ) = 1 h = a( )=a ( )a( ) = a( )=m E jyF (t)j2 = a ( )Ra( )=m2 n o Lecture notes to accompany Introduction to Spectral Analysis by P. Prentice Hall. Stoica and R. 1997 Slide L9–9 .

Stoica and R. but Slide L9–10 Lecture notes to accompany Introduction to Spectral Analysis by P. ( )^ ( ) This is the direct spatial analog of the Blackman-Tukey periodogram. Moses. pj > = wavelength array length array beamwidth Inconsistency problem: Beamforming DOA estimates are consistent if n inconsistent if n > . Prentice Hall. 1 = 1. 1997 . Resolution Threshold: inf j k .Implementation of Beamforming 1 N y(t)y (t) ^ R=N t=1 X The beamforming DOA estimates are: f^kg = the locations of the n largest peaks of a Ra .

1a( ): Performance: Slightly superior to Beamforming. Prentice Hall. Lecture notes to accompany Introduction to Spectral Analysis by P. Stoica and R. 1997 Slide L9–11 . They do not make assumptions on the covariance properties of the data (and hence do not depend on them).1a( ) n o Implementation: f^kg = the locations of the n largest peaks of ^ 1=a ( )R. Moses.Capon Method Filter design: min(h Rh) subject to h a( ) = 1 h Solution: h = R. Both Beamforming and Capon are nonparametric approaches.1a( )=a ( )R.1a( ) E jyF (t)j2 = 1=a ( )R.

Moses. MIN-NORM and ESPRIT methods of frequency estimation can be used. Stoica and R. Prentice Hall. Lecture notes to accompany Introduction to Spectral Analysis by P. MUSIC. for DOA estimation.Parametric Methods Assumptions: The array is described by the equation: y(t) = As(t) + e(t) The noise is spatially white and has the same power in all sensors: E fe(t)e (t)g = 2I The signal covariance matrix P = E fs(t)s (t)g is nonsingular. Then: R = E fy(t)y (t)g = APA + 2I Thus: The NLS. almost without modiﬁcation. 1997 Slide L9–12 . YW.

A(A A).1A y(t) t = 1 : : : N s X Then 1 N k I . f ^k g = arg max trf A(A A).1A ]Rg ^ Thus.1A ]Rg X f kg For N . As(t)k2 min(t)g N f k g fs t=1 X | Minimizing f over s gives f ( s) {z } ^(t) = (A A).Nonlinear Least Squares Method 1 N ky(t) . Lecture notes to accompany Introduction to Spectral Analysis by P. A(A A).1A ]y(t) = N t=1 ^ = trf I . this is precisely the form of the NLS method of frequency estimation.1A ]y(t)k2 f ( ^) = N s t=1 1 N y (t) I . 1997 Slide L9–13 =1 . Moses. Prentice Hall. Stoica and R. A(A A).

Stoica and R. 1997 Slide L9–14 .Nonlinear Least Squares Method Properties of NLS: Performance: high Computational complexity: high Main drawback: need for multidimensional search. Prentice Hall. Lecture notes to accompany Introduction to Spectral Analysis by P. Moses.

n ~ ~ Properties: A V2 = 0 V1 2 R(A) Then: rank |{z} |{z} Lecture notes to accompany Introduction to Spectral Analysis by P. Moses.n 0 V2 g L n M .) = n. and the SVD of . Stoica and R.= y (M L) Also assume: M > n L > n ( ) m = M + L > 2n) rank ~ (A) = rank(A) = n " # " # (.Yule-Walker Method e y(t) = y(t) = A s(t) + ~(t) ~ y(t) ~ e(t) A " # " # " # Assume: Then: E fe(t)~ (t)g = 0 e 4 E fy(t)~ (t)g = AP A ~ . 1997 Slide L9–15 . is . Prentice Hall. = U1 U2 ] 0 n n 0 V1 g n .

Moses. Prentice Hall.=N y t=1 X Properties: Computational complexity: medium Performance: satisfactory if m 2n Main advantages: – weak assumption on e t – the subarray A need not be calibrated f ( )g Lecture notes to accompany Introduction to Spectral Analysis by P.YW-MUSIC DOA Estimator f^kg = where the n largest peaks of 1=~ ( )V2V2 ~( ) a ^^a ~( ). is the “array transfer vector” for y(t) a ~ at DOA ^ V2 is deﬁned similarly to V2. Stoica and R. (L 1). using 1 N y(t)~ (t) ^ . 1997 Slide L9–16 .

Prentice Hall. that is. if rank (P ) = rank(E fs(t)s (t)g) < n Modiﬁcations that apply in the coherent case exist. Spectral forms of MUSIC and Min-Norm can be used for arbitrary arrays Root forms can be used only with ULAs MUSIC and Min-Norm break down if the source signals are coherent. Moses. 1997 Slide L9–17 . Stoica and R.MUSIC and Min-Norm Methods Both MUSIC and Min-Norm methods for frequency estimation apply with only minor modiﬁcations to the DOA estimation problem. Lecture notes to accompany Introduction to Spectral Analysis by P.

D= 0 e..ESPRIT Method Assumption: The array is made from two identical subarrays separated by a known displacement vector. Moses. Lecture notes to accompany Introduction to Spectral Analysis by P..i!c ( n) ( ) = the time delay from subarray 1 to 2 6 4 3 7 5 subarray 2 for a signal with DOA = : where d is the subarray separation and is measured from the perpendicular to the subarray displacement vector. 1997 Slide L9–18 ( ) = d sin( )=c . Stoica and R. Prentice Hall. where e.i!c ( 1) 0 . Let m = # sensors in each subarray A1 = Im 0]A (transfer matrix of subarray 1) A2 = 0 Im]A (transfer matrix of subarray 2) Then A2 = A1D.

so m m and .1 0]A = .1 A1 = Im.1]A Lecture notes to accompany Introduction to Spectral Analysis by P. 1997 Slide L9–19 . Prentice Hall.1 . Moses.ESPRIT Method. con't ESPRIT Scenario source θ known subarray 1 displac ement v ector subarray 2 Properties: Requires special array geometry Computationally efﬁcient No risk of spurious DOA estimates Does not require array calibration Note: For a ULA. the two subarrays are often the ﬁrst m and last m array elements. Stoica and R.1 A2 = 0 Im.