polynomial matrix decompositions: applications and techniquesmopnet.cf.ac.uk/foster.pdf · 2009....
TRANSCRIPT
![Page 1: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/1.jpg)
Polynomial Matrix Decompositions: Algorithms and Applications
Joanne Foster – Loughborough University
John McWhirter – Cardiff University
Jonathon Chambers – Loughborough University
http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Email: [email protected]
![Page 2: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/2.jpg)
2 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Talk Outline
What are polynomial matrices?
Three types of polynomial matrix decompositions:
Eigenvalue decomposition (SBR2 Algorithm)
QR decomposition
Singular value decomposition
The potential applications of these decompositions to MIMO communication problems
![Page 3: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/3.jpg)
3 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
2x2 example:
General pxq polynomial matrix form:
A z =[a11 z a1q z
⋮ ⋮a p1 z ⋯ a pq z
] = ∑t=T
min
Tmax
A t z−t
T min≤Tmax
A z =[ 1 z−1 23z−2
2 z−1z−2 4z−1 ]
What is a Polynomial Matrix?
![Page 4: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/4.jpg)
4 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Used to describe convolutive mixing:
Source signals received at sensors over multiple paths and with different delays
Polynomial mixing matrix required, where each element is an FIR filter
More realistic mixing situation
…
How do polynomial matrices arise?
![Page 5: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/5.jpg)
5 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Polynomial Matrix Formulation
Two signals and two sensors:
Polynomial matrix form:
i.e.
x1 z =a11 z s1 z a12 z s2 z x 2 z =a21 z s1 z a22 z s2 z
[ x 1 z
x2 z ]=[a11 z a12 z
a21 z a22 z ][s 1 z
s2 z ]
x z = A z s z
![Page 6: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/6.jpg)
6 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Special Polynomial Matrices
Unimodular Matrix:
Paraconjugation:
Para-Hermitian Matrix:
Paraunitary matrix (defines multichannel all-pass filter):
A z = A z
H z H z = H z H z = I
A z = A¿T 1/ z
det [ A z ]=constant
![Page 7: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/7.jpg)
7 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Paraconjugate
Inverse
Holds for all values of z - time domain processing
A z =[1 z−1 z2
z−1 2 ]
A−1 z =[ 2 − z2 −z−1 1z−1 ] det [ A z ]=1
A z =[ 1z zz−12 2 ]
Polynomial Matrix Examples
![Page 8: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/8.jpg)
8 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Can be expressed as:
Calculate the decomposition of A:
Rearrange (*): then
This application could be extended to sets of polynomial equations, but would require new algorithms for formulating the decompositions
Motivation: the QR decomposition in narrowband signal processing
a11a21
a22
a12x1
x 2
y1
y2
[ y1
y2 ]=[a11 a12
a21 a22 ][x 1
x 2][a11 a12
a21 a22 ]=[q11 q12
q21 q22 ][r 11 r 12
0 r 22 ]y '=QH y=Rx
y '2=r22 x 2
y '1=r 11 x 1r 12 x 2
(*)
![Page 9: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/9.jpg)
9 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
The PQRD of a pxq polynomial matrix A(z) can be expressed as
where Q(z) is a pxp paraunitary matrix,
i.e.
and R(z) is a pxq upper triangular polynomial matrix.
Q z A z =R z
Q z Q z =Q z Q z = I
The Polynomial Matrix QR Decomposition (PQRD)
![Page 10: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/10.jpg)
10 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Gα , θ , φ , t z =[1 0
0 z−t ][cos θ eiα sin θ eiφ
−sin θ e−iφ cos θ e−iα ][1 0
0 z t ]
Rotate Delay
=[cos θ eiα sin θ eiφ z t
−sin θ e−iφ z−t cosθ e−iα ]This matrix is paraunitary.
Un-delay
Elementary Polynomial Givens Rotations (EPR)
![Page 11: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/11.jpg)
11 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Application of an EPGR on a Polynomial Matrix
Delay Rotate Un-delayz 0z−2
![Page 12: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/12.jpg)
12 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Example of an EPGR
A z =[ 1 22z−1 1 ]
[1 00 z1 ][ 1 2
2z−1 1 ]=[1 22 z1 ]
1
5 [1 2−2 1 ][1 2
2 z1 ]= 1
5 [52z−1 22z1
0 −4z1 ]
Delay:
Rotate:
Un-delay: 1
5 [1 00 z−1 ][52z−1 22z1
0 −4z1 ]= 1
5 [52z−1 22z1
0 −4z−11 ]
![Page 13: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/13.jpg)
13 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Step 1: To drive all elements beneath the diagonal of column 1 to zero. Iterative process – each iteration apply an EPGR (polynomial rotation matrix) to zero the largest coefficient (in magnitude) beneath the diagonal.
Step 2: To drive all elements beneath the diagonal of column 2 to zero.
Step 3: To drive all elements beneath the diagonal of column 3 to zero.
Repeat Steps 1 - 3: if any coefficients are larger beneath the diagonal are larger than a stopping criterion.
The PQRD By Columns Algorithm
A z =[a11 z a12 z a13 z
a21 z a22 z a23 z
a31 z a32 z a33 z
a41 z a42 z a43 z ]A z =[
a11 z a12 z a13 z
0 a22 z a23 z
0 0 a33 z
0 0 0]A z =[
a11 z a12 z a13 z
0 a22 z a23 z
0 a32 z a33 z
0 a42 z a43 z ]A z =[
a11 z a12 z a13 z
0 a22 z a23 z
0 0 a33 z
0 0 a43 z ]
![Page 14: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/14.jpg)
14 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Example 1
Paraunitary Upper triangular matrix
Input to PQRD: A z =[2 0 2z1
z 1 1 0
0 z−1 1 ]
Q z =[0 .8944 0 .4472 z−1 0
−0 . 2981 z1 0 .5963 0 .7454 z1
0 .3333 z1 −0 .6667 0 .6667 z2 ] R z =[2 .2361 0 . 4472 z−1 1 .7889 z 1
0 1. 3416 0 .7454 z 1−0 .5963 z 2
0 0 0 .6667 z 10 .6667 z 2 ]
![Page 15: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/15.jpg)
15 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Example 2
![Page 16: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/16.jpg)
16 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Upper Triangular Matrix R(z)
![Page 17: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/17.jpg)
17 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Paraunitary matrix Q(z)
![Page 18: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/18.jpg)
18 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Matrix obtained from inverse decomposition
A z =Q z R z R z = Q z A z Strictly upper
triangular
Input matrix
Inverse Decomposition
![Page 19: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/19.jpg)
19 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
The proportion of the Frobenius norm of matrix positioned beneath the diagonal of the matrix:
Performance Measure
∑t∑j=2
4
∑k=1
j−1
∣a jk' t ∣2
∑t=0
4
∑j=1
4
∑k=1
4
∣a jk t ∣2
Measure of the upper triangularity of the polynomial matrix.
![Page 20: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/20.jpg)
20 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Performance Measure
![Page 21: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/21.jpg)
21 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Order of both polynomial matrices grows at each iteration
Many outer layers consisting of small values
Find max value of T1 and min value of T2 such that:
Keep only coefficients of
This ensures that the accuracy of the decomposition is not significantly compromised, whilst reducing the computational load.
Truncation Method
∑t=T 2
b
∑j=1
p
∑k=1
p
∣a jk' t ∣2
∑t∑j=1
p
∑k=1
p
∣a jk t ∣2
¿μ2
∑t=a
T1
∑j=1
p
∑k=1
p
∣a jk' t ∣2
∑t∑j=1
p
∑k=1
p
∣a jk t ∣2
¿μ2
zT
11, , z
T21
![Page 22: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/22.jpg)
22 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
The PSVD of a pxq polynomial matrix A(z) can be expressed as
where
U(z) is a pxp paraunitary matrix,
V(z) is a qxq paraunitary matrix and
S(z) is a pxq diagonal polynomial matrix.
U z A z V z =Σ z
The Polynomial Matrix Singular Value Decomposition (PSVD)
![Page 23: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/23.jpg)
23 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
U N z U 1 z V 1 z V N z =Σ z
Yes
PSVD has been
calculated
A' ' z =[
a11¿ 1/ z a21
¿ 1/ z a31¿ 1/ z a 41
¿ 1/ z
0 a22¿ 1/ z a32
¿ 1/ z a 42¿ 1 / z
0 0 a33¿ 1/ z a43
¿ 1 / z ]
Set
Is A(z) diagonal?
Step 2:
Calculate the PQRD of :A z
U i z A z =A' z V i z A
' z = A' ' z
A z = A' ' z
Step 1:
Calculate the PQRD of A(z): A' z
The PSVD by PQRD Algorithm
A z =[a11 z a12 z a13 z
a21 z a22 z a23 z
a31 z a32 z a33 z
a41 z a42 z a43 z ] A' z =[
a11 z a12 z a13 z
0 a22 z a23 z
0 0 a33 z
0 0 0]A
' z =[
a11¿ 1 / z 0 0 0
a12¿ 1 / z a22
¿ 1/ z 0 0
a13¿ 1 / z a23
¿ 1/ z a33¿ 1/ z 0 ]
No
A z
i=i1
i=1
Assuming the algorithm requires N iterations to convergence
![Page 24: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/24.jpg)
24 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Example 1
![Page 25: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/25.jpg)
25 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Diagonal Matrix Σ(z)
![Page 26: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/26.jpg)
26 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Paraunitary matrix U(z)
![Page 27: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/27.jpg)
27 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Paraunitary matrix V(z)
![Page 28: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/28.jpg)
28 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Matrix obtained from inverse decomposition
A z = U z Σ z V z Σ z =U z A z V z Strictly diagonal
Input matrix
Inverse Decomposition
![Page 29: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/29.jpg)
29 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Convergence Measures
![Page 30: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/30.jpg)
30 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
x t =∑k=0
N
A k s t−k n t
U z A z V z =Σ z
Convolutive Mixing Model
Or expressed in polynomial form x z = A z s z n z
Assume channel matrix is known, then
U z x z x ' z
=U z U z Σ z V z A z
[ V z s' z ]
s z
U z n z n ' z
Rearranging
Application to broadband MIMO communications
![Page 31: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/31.jpg)
31 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Potential Application of the PSVD to MIMO Systems
A z U z V z
¿ Σ z
s1 z s 2 z
s p z
x1 z x 2 z
x q z
![Page 32: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/32.jpg)
32 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
V z A z U z s ' z x z
n z
x ' z
Σ z
n ' z
x ' z Equivalent system:
Implement:
The PSVD in a MIMO system:
s ' z
![Page 33: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/33.jpg)
33 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
[ x 1' z
x2' z ]=[σ11 z 0
0 σ 22 z ][s '1 z
s'2 z ][n1' z
n2' z ]
x1' z =σ 11 z s
'1 z n1
' z
x 2' z =σ22 z s
'2 z n2
' z
1. Estimate source 1
2. Estimate source 2
Single channel equalisation
problems – solve using a
maximum likelihood sequence estimator
2x2 Example (PSVD)
![Page 34: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/34.jpg)
34 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Comparative Bit Error Rate Simulations
Benchmark scheme: Dominant mode MIMO Orthogonal Frequency Division Multiplexing (OFDM) SVD [1].
Our scheme: Dominant mode PSVD with Turbo equalisation [2].
Simulation parameters:
5 x 5 exponentially decaying MIMO channels Matrix order (delay spread) four. Additive complex Gaussian noise with variance chosen
to obtain a range of signal to noise values.
[1] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless Communications, 1st ed. University Press, Cambridge: Cambridge University Press, 2003.
[2] M. Davies, S. Lambotharan, J. Chambers and J. G. McWhirter, Broadband MIMO Beamforming For Frequency Selective Channels Using the Sequential Best Rotation Algorithm, 67th Vehicular Technology Conference, Singapore, 2008.
![Page 35: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/35.jpg)
35 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
M. Davies, Polynomial Matrix Decomposition Techniques for Frequency Selective MIMO Channels, PhD Thesis, Department of Electronic and Electrical Engineering, Loughborough University, submitted 2009.
![Page 36: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/36.jpg)
36 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
x t =∑k=0
N
A k s t−k n t
A z =Q z R z
Convolutive Mixing Model
Or expressed in polynomial form x z = A z s z n z
Assume channel matrix is known, then
Q z x z x ' z
=R z s z Q z n z n ' z
Rearranging
Application to broadband MIMO communications
![Page 37: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/37.jpg)
37 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
A z Q z s z x z
n z
x ' z
R z s z
n ' z
x ' z
Implement:
Equivalent system:
The PQRD in a MIMO system:
![Page 38: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/38.jpg)
38 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
[ x 1' z
x2' z ]=[ r 11 z r12 z
0 r 22 z ][s1 z
s2 z ][n1' z
n2' z ]
x 2' z =r 22 z s2 z n2
' z
x1' z −r12 z s2 z =r11 z s1 z n1
' z
Use back substitution to...
1. Estimate source 2
2. Estimate source 1
Single channel equalisation
problem – solve using a
maximum likelihood sequence estimator
2x2 Example (PQRD)
![Page 39: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/39.jpg)
39 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Comparative Bit Error Rate Simulations
Benchmark scheme: Horizontal Bell Laboratories Layered Space Time (H-BLAST) MIMO-OFDM QRD [1,2].
Our scheme: H-BLAST PQRD.
Simulation parameters:
3 x 3 quasi static MIMO channels (polynomial) Matrix order (delay spread) four. Additive complex Gaussian noise with variance chosen
to obtain a range of signal to noise values.
[1] A. Paulraj, R. Nabar, and D. Gore, Introduction to Space-Time Wireless Communications, 1st ed. University Press, Cambridge: Cambridge University Press, 2003.
[2] G. Foschini, Layered space-time architecture for wireless communication in a fading environment when using multi-element antennas, Bell Labs Technical Journal (Autumn), pp. 41–59, 1996.
![Page 40: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/40.jpg)
40 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
M. Davies, Polynomial Matrix Decomposition Techniques for Frequency Selective MIMO Channels, PhD Thesis, Department of Electronic and Electrical Engineering, Loughborough University, submitted 2009.
SNR
![Page 41: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/41.jpg)
41 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
Algorithms developed for calculating the following decompositions of a polynomial matrix:
QR decomposition Eigenvalue decomposition Singular value decomposition
Decompositions have been shown to converge, the proof of which is given in [1].
Potential applications (so far) are to broadband MIMO channel equalisation and convolutive blind source separation (strong decorrelation).
Conclusions
[1] J.A. Foster, J.G. McWhirter, M. Davies and J.A Chambers, An Algorithm for Calculating the QR and Singular Value Decompsotion of a Polynomial Matrix, accepted for publication to IEEE Transactions on Signal Processing.
![Page 42: Polynomial Matrix Decompositions: Applications and Techniquesmopnet.cf.ac.uk/foster.pdf · 2009. 10. 1. · Polynomial Matrix Formulation Two signals and two sensors: Polynomial matrix](https://reader033.vdocument.in/reader033/viewer/2022060822/609b7ae4c30804454049d284/html5/thumbnails/42.jpg)
42 http://www.lboro.ac.uk/departments/el/research/asp/ProjSite/index.htm
J.G. McWhirter, P.D. Baxter, T. Cooper, S. Redif and J. Foster, An EVD Algorithm for
Para-Hermitian Polynomial Matrices, IEEE Transactions on Signal Processing, Vol. 55(6),
pp. 2158-2169, 2007.
J. A. Foster, J.G. McWhirter and J. Chambers, A Polynomial Matrix QR Decomposition
with Application to MIMO Channel Equalisation, Proc. 41st Asilomar Conference on
Signals, Systems and Computers, Asilomar (Invited Talk), California, November, 2007. M. Davies, S. Lambotharan, J. Chambers and J. G. McWhirter, Broadband MIMO
Beamforming For Frequency Selective Channels Using the Sequential Best Rotation Algorithm, 67th Vehicular Technology Conference, Singapore, 2008.
J. A. Foster, Algorithms and Techniques for Polynomial Matrix Decompositions, PhD Thesis, School of Engineering, Cardiff University, UK, 2008.
M. Davies, S. Lambotharan, J.A. Foster, J. Chambers and J. G. McWhirter, Polynomial Matrix QR Decomposition and Iterative Decoding of Frequency Selective MIMO Channels, IEEE Wireless Communications and Networking Conference, Budapest, Hungary, 2009.
J.A. Foster, J.G. McWhirter and J. Chambers, A Novel Algorithm for Calculating the QR Decomposition of a Polynomial Matrix, ICASSP, 2009.
J.A. Foster, J.G. McWhirter, M. Davies and J.A Chambers, An Algorithm for Calculating the QR and Singular Value Decompsotion of a Polynomial Matrix, accepted for publication to IEEE Transactions on Signal Processing.
References