Spatial Array Processing

Signal and Image Processing Seminar

Murat Torlak

Telecommunications & Information Sys. Eng.

The University of Texas at Austin

1 Introduction

 A sensor array is a group of sensors located at spatially separated points

 Sensor array processing focuses on data collected at the sensors to carry out a given estimation task

 Application Areas

– Sonar

– Seismic exploration

– Anti-jamming communications

– YES! Wireless communications

2 Problem Statement

s1 (t) s2 (t)

θ 1 θ 2

∆ x (t) x (t) x (t) x (t) x1 (t) 2 3 x4 (t) 5 6

Find

1. Number of sources

2. Their direction-of-arrivals (DOAs)

3. Signal Waveforms

3 Assumptions

 Isotropic and nondispersive medium

– Uniform propagation in all directions

 Far-Field

– Radius of propogation >> size of array

– Plane wave propogation

 Zero mean white noise and signal, uncorrelated

 No coupling and perfect calibration

4 Antenna Array

Source θ

X X X X12X 3 45∆

 Array Response Vector–Far-Field Assumption

Narrowband 

- Delay = Phase Shift

Assumption

j2f 4 sin =c j2f 44 sin =c T

c c

a  = [1;e ;::: ;e ]

= xt

 Single Source Case

2 3 3 3 2 2

x t s t 1

1 1

6 7 7 7 6 6

j 2f 

c

6 7 7 7 6 6

x t s t   e

2 1

6 7 7 7 6 6

6 7 7 7 6 6

s t = a s t = 

6 7 7 7 6 6

1 1

. . . 1

6 7 7 7 6

. 6 . .

6 7 7 7 6 6

4 5 5 5 4

. 4 . .

j 2f M 1

c

x t s t M 1  e

1

M

 = 4 sin =c

where 1 .

5

General Model d

 By superposition, for signals,

xt = a s t+ + a s t

1 1

d d

d

X

= a s t

k k

k=1

 Noise

d

X

xt = a s t+ nt

k k

k =1

= ASt+ nt

where

A = [a ;::: ;a ]

1 d

and

T

St = [s t;::: ;s t] :

1 d

6 Low-Resolution Approach:Beamforming

 Basic Idea

d d

X X

i1j 2f 4 sin =c jw i1

c

k k

x t = = e s t = s te

i

k k

k =1 k =1

w = 2 4 sin =c i = 1;::: ;M

k

where k and .

 fw g

Use DFT (or FFT) to find the frequencies k

2 3

1 1  1

6 7

jw

jw jw

6 7

1 2 M

e e  e

6 7

6 7

F = [Fw    Fw ] =

6 7 1

M . . . . 6

. . . . 7 6

. 7 4

. . . 5

jM1w

jM 1w j M 1w

1 2 M

e e  e

 Look for the peaks in

 2

jF x tj = jF xtj i

 To smooth out noise

N

X

1

 2

jF xtj B w  =

i

N

t=1

7 Beamforming Algorithm

 Algorithm

P

N

1



R = xtx t

1. Estimate x

t=1

N



B w  = F w R Fw 

i x i

2. Calculate i

B w  w i

3. Find peaks of i for all possible ’s.

i = 1;::: ;d

4. Calculate k , .

 Advantage

- Simple and easy to understand

 Disadvantage

- Low resolution

8

Number of Sources

d < M

 Detection of number of signals for ,

xt = Ast+ nt

   

R = Efxtx tg = AEfsts tgA + E fntn tg

x

| {z } | {z }

2

R

s

 I

n

 2

I = A R A +

s

n

|{z} |{z}

|{z}

M d dM

dd 2

where  is the noise power.

n

 R d

No noise and rank of s is



R = AR A s

– Eigenvalues of x will be

f ;::: ; ;0; ::: ;0g:

1

d R

–Real positive eigenvalues because x is real, Hermition-symmetric

– rank d

 R

Check the rank of x or its nonzero eigenvalues to

detect the number of signals

2 

 Noise eigenvalues are shifted by

n

2 2 2 2

f +  ;::: ; +  ; ;::: ; g:

1

d

n n n n

 > ::: >   >> 0 1

where d and

 Detect the number of principal (distinct) eigenvalues

9 MUSIC

 Subspace decomposition by performing eigenvalue

decomposition

M

X

  2

I =  e e R = AR A + 

x s

k k

n

k

k =1

e  k

where k is the eigenvector of the eigenvalue

 spanfAg = spanfe ;::: ;e g = spanfE g

1 d s

 a   spanfE g P a  A

Check which s or or

?

P a  P

, where A is a projection matrix

A

 Search for all possible such that

1

? 2

jP a j = 0 or M   = = 1

A

P a 

A

 R

After EVD of x

?  

P = I E E = E E

s n

A s n

where the noise eigenvector matrix

E = [e ;::: ;e ]

n d+1 M

10

Root-MUSIC

j 2f 4 sin =c

c

e

 For a true , is a root of

M

X

M1 T  1 M1

P z  = [1;z;::: ;z ] e e [1;z ;::: ;z ]:

k

k

k =d+1

 After eigenvalue decomposition,

d

fe g

- Obtain k

k =1

z 

-Formp

M 2 pz  - Obtain 2 roots by rooting

-Pickd roots lying on the unit circle

f g

- Solve for k

11 Estimation of Signal Parameters via

Rotationally Invariant Techniques (ESPRIT) M

 Decompose a uniform linear array of sensors into

1 two subarrays with M sensors

 Note the shift invariance property

2 3 3 2

jw

1 e

6 7 7 6

6 7 7 6

jw j2w

e e

6 7 7 6

jw 1 jw 2

6 7 7 6

= e = a e a   =

6 7 7 6 . .

. . 6 7 7

6 . .

4 5 5 4

jM 1w jM 1w

e e

 General form relating subarray (1) to subarray (2)

2 3

jw

1

e

6 7

6 7 1

2 .

= A1: = A

A . 7

6 .

4 5

jw

d

e

  f g

contains sufficient information of k

12

ESPRIT

 spanfE g = spanfAg E = AT s

s and

d  d - T is a nonsingular unitary matrix

- T comes from a Grahm-Schmit orthogonalization

of Ab in

 

R = E  E + E E

x s s n

s n

H 2

A R A +  I

s

n

2 1

2 1

 E = A T E = A T s

s and

2 1 1

E 2 = A T = A T = E 1T T

s s

1

 E

Multiply both sides by the pseudo inverse of s

1

1 1 1 1 1 1 1

E E 2 = E E  E E T T = T T

s s

where  means the pseudo-inverse

 sH 1 sH

A = A A A

1

 T  Eigenvalues of T are those of .

13

Superresolution Algorithms

P

N

1



R = xk x k 

1. Calculate x

k =1 N

2. Perform eigenvalue decomposition

f g d

3. Based on the distribution of k , determine

4. Use your favorite diraction-of-arrival estimation

algorithm:

  0

(a) MUSIC: Find the peaks of M for from to



180

d

^

f g d

- Find k corresponding the peaks of

k =1



M .

z  (b) Root-MUSIC: Root the polynomial p

- Pick the d roots that are closest to the unit

1

r c

d

k

^

fr g = sin k

circle k and .

k =1

2f 

c

2 1

E E s

(c) ESPRIT: Find the eigenvalues of s ,

f g

k

 c

1

k

^

= sin

- k

2f 4 c

14

Signal Waveform Estimation

A st xt  Given , recover from .

 Deterministic Method w

– No noise case: find k such that

w ? a ;i 6= k; w 6? a 

k i k k

 A

 can do the job

 

A xt = A Ast = st

nt

 With noise,

 

A xt = st + A nt  – Disadvantage = increased noise

15

Stocastic Approach

 w

Find k to minimize

2 

min E fjw xtj g = min w R w

k k k

k

 

a  w =1 a  w =1

k k k k

 Use the Langrange method

2  

min E fjw xtj g , min w R w + 2a  w 1

k k k k k

k



;w

a  w =1

k

k k

 Differentiating it, we obtain

1

R w = a ;orw = R a 

x

k k k k x

.

  1

 a  w = a  R a  = 1

k k k

Since k , x

 Then

 1

 = a  R a 

k k x

 Capon’s Beamformer

1  1

w = R a =a  R a 

k k k k

x x

16 Subspace Framework for Sinusoid

Detection

d

P

 +j! t

k k

 xt = e

k

k =1 M

 Let us select a window of , i.e.,

T

xt = [xt;: : : ;xt M + 1]

 Then

2 3 3 2

 +j! t

k k

xt e

k

6 7 7 6

 +j! t1

6 7 7 6

k k

xt 1 e

d k

6 7 7 6

X

6 7 7 6

xt = =

6 7 7

. 6 .

6 7 7

. 6 .

6 7 7 6

k=1

4 5 5

. 4 .

 +j! tM +1

k k

xt M +1

e

k

3 2

1

7 6

 +j! 

7 6

k k

e

d

7 6

X

7 6

 +j! t

k k

e =

7 6

. k

7 6

{z }

. |

7 6

k =1 5

4 .

s t

k

 +j! M +1

k k

e

{z } |

a 

k

d

X

a s t = Ast; =

k k

k =1 d

where M is the window size, the number of sinusoids, and

+j!

k k

 = e

k .

17 Subspace Framework for Sinusoid Detection

 Therefore, the subspace methods can be applied to

f + j! g k find k

 Recall

d

X

 +j! t

k k

xt = e

k

k =1

 f g

Then finding k is a simple least squares problem.

18 Wireless Communications

co-channel interference

Multipaths

Direct Path

Cellular Telephony

Office Building

Residential Area

Personal Communications Outdoors Services (PCS)

To Networks

Direct Path

Direct Path

Multipath Wireless LAN

 Increasing Demand for Wireless Services

 Unique Problems compared to Wired communications

19 Problems in Wireless Communications

 Scarce Radio Spectrum and Co-channel Interference

1 2 4 3 1

4 1 3 2 1

 Multipath

Multipath Direct Path Multipath

Base Station

Time

Desired Signal Reflected Signal

 Coverage/Range

20 Systems

 Employ more than one antenna element and exploit the spatial dimension in to improve some system operating parameter(s):

- Capacity, Quality, Coverage, and Cost.

User One User Two

Multiple RF Module

Advanced Signal Processing Algorithms

Conventional Communication Module

21 Experimental Validation of Smart Uplink Algorithm

 Comparison of constellation before (upper) and after smart uplink processing (middle and lower)

real axis Antenna Output

real axis Equalized Signal 1 imaginary axis imaginary axis imaginary axis real axis Equalized Signal 2

22 Selective Transmission Using DOAs

 Beamforming results for two sources separated by

 20

1

0.8

0.6

0.4

Power Spectrum 0.2

0 0.5 1 1.5 2 Frequency [Hz], User #1 4 x 10 1

0.8

0.6

0.4

Power Spectrum 0.2

0 0.5 1 1.5 2 Frequency [Hz], User #2 4 x 10

23 Selective Transmission Using DOAs

 Beamforming results for two sources separated by

 3

1

0.8

0.6

0.4

Power Spectrum 0.2

0 0.5 1 1.5 2 Frequency [Hz], User #1 4 x 10 1

0.8

0.6

0.4

Power Spectrum 0.2

0 0.5 1 1.5 2 Frequency [Hz], User #2 4 x 10

24 Future Directions

 Adapt the theoretical methods to fit the particular demands in specific applications

– Smart Antennas

– Synthetic aperture radar

– Underwater acoustic imaging

– Chemical sensor arrays

 Bridge the gap between theoretical methods and real-time applications

25