Presentation on Mutual information using filterbank equalization for MIMO frequency selective channels presented at the National Conference on Communications 2011 held at IISc, Bangalore
1. Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective Mutual Information with filterbank
channels
Vijaya Krishna equalization for MIMO frequency selective
A, Shashank
V channels
Vijaya Krishna A Shashank V
Department of ECE
P E S Institute of Technology, Bangalore
NCC 2011
2. Outline
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
1 Motivation
Vijaya Krishna
A, Shashank
V
2 Signal model
3 Block processing
4 Filterbank framework
5 Mutual information with filterbank equalization
6 Conclusion
3. Motivation
Mutual
Information
with filterbank
equalization
for MIMO
frequency
MIMO systems: Higher rate, more reliability
selective
channels
Vijaya Krishna
Frequency selectivity: Equalization required at receiver
A, Shashank
V
Typically, block processing used:
Motivation
Zero padding or cyclic prefixing: Convert frequency
Signal model
selective fading to flat fading
Block
processing
Redundancy of the order of channel length required
Filterbank
framework
Mutual Lower data rates
information
Simulations
Additional processing required: coding, etc
Conclusion
4. Motivation
Mutual
Information
with filterbank
equalization
for MIMO
frequency
MIMO systems: Higher rate, more reliability
selective
channels
Vijaya Krishna
Frequency selectivity: Equalization required at receiver
A, Shashank
V
Typically, block processing used:
Motivation
Zero padding or cyclic prefixing: Convert frequency
Signal model
selective fading to flat fading
Block
processing
Redundancy of the order of channel length required
Filterbank
framework
Mutual Lower data rates
information
Simulations
Additional processing required: coding, etc
Conclusion
5. Motivation
Mutual
Information
with filterbank
equalization
for MIMO
frequency
MIMO systems: Higher rate, more reliability
selective
channels
Vijaya Krishna
Frequency selectivity: Equalization required at receiver
A, Shashank
V
Typically, block processing used:
Motivation
Zero padding or cyclic prefixing: Convert frequency
Signal model
selective fading to flat fading
Block
processing
Redundancy of the order of channel length required
Filterbank
framework
Mutual Lower data rates
information
Simulations
Additional processing required: coding, etc
Conclusion
6. Motivation
Mutual
Information
with filterbank
equalization
for MIMO
frequency
MIMO systems: Higher rate, more reliability
selective
channels
Vijaya Krishna
Frequency selectivity: Equalization required at receiver
A, Shashank
V
Typically, block processing used:
Motivation
Zero padding or cyclic prefixing: Convert frequency
Signal model
selective fading to flat fading
Block
processing
Redundancy of the order of channel length required
Filterbank
framework
Mutual Lower data rates
information
Simulations
Additional processing required: coding, etc
Conclusion
7. Mutual
Information
with filterbank
equalization Filterbank equalizers:
for MIMO
frequency
selective
Instead of converting to flat fading, view the channel as
channels FIR filter
Vijaya Krishna
A, Shashank
V Equalization: Inverse filtering
Motivation
Signal model
Block
processing
Filterbank
framework
Mutual
information
By adding no/minimal redundancy, we can find FIR
Simulations
inverse filters
Conclusion
8. Mutual
Information
with filterbank
equalization Filterbank equalizers:
for MIMO
frequency
selective
Instead of converting to flat fading, view the channel as
channels FIR filter
Vijaya Krishna
A, Shashank
V Equalization: Inverse filtering
Motivation
Signal model
Block
processing
Filterbank
framework
Mutual
information
By adding no/minimal redundancy, we can find FIR
Simulations
inverse filters
Conclusion
9. Mutual
Information
with filterbank
equalization Filterbank equalizers:
for MIMO
frequency
selective
Instead of converting to flat fading, view the channel as
channels FIR filter
Vijaya Krishna
A, Shashank
V Equalization: Inverse filtering
Motivation
Signal model
Block
processing
Filterbank
framework
Mutual
information
By adding no/minimal redundancy, we can find FIR
Simulations
inverse filters
Conclusion
10. Mutual
Information
with filterbank
equalization
for MIMO
frequency Mutual information: Acheivable data rate
selective
channels I(X ; Y ) = H(X ) − H(X |Y )
Vijaya Krishna
A, Shashank
V
Aim: Quantify data rate: Mutual information for Filter
Motivation bank case
Signal model
Block Our Contribution:
processing
1 Derivation of expression for MI with filterbank
Filterbank
framework equalization for the MMSE criterion
Mutual
2 MI expression for the case of symbol by symbol
information detection
Simulations
Conclusion
11. Mutual
Information
with filterbank
equalization
for MIMO
frequency Mutual information: Acheivable data rate
selective
channels I(X ; Y ) = H(X ) − H(X |Y )
Vijaya Krishna
A, Shashank
V
Aim: Quantify data rate: Mutual information for Filter
Motivation bank case
Signal model
Block Our Contribution:
processing
1 Derivation of expression for MI with filterbank
Filterbank
framework equalization for the MMSE criterion
Mutual
2 MI expression for the case of symbol by symbol
information detection
Simulations
Conclusion
12. Mutual
Information
with filterbank
equalization
for MIMO
frequency Mutual information: Acheivable data rate
selective
channels I(X ; Y ) = H(X ) − H(X |Y )
Vijaya Krishna
A, Shashank
V
Aim: Quantify data rate: Mutual information for Filter
Motivation bank case
Signal model
Block Our Contribution:
processing
1 Derivation of expression for MI with filterbank
Filterbank
framework equalization for the MMSE criterion
Mutual
2 MI expression for the case of symbol by symbol
information detection
Simulations
Conclusion
13. Signal model
Mutual
Information
with filterbank
equalization
Consider M×N frequency selective LH tap MIMO
for MIMO
frequency
channel
selective
channels Signal model:
Vijaya Krishna LH −1
A, Shashank
V y [n] = H(k ) x(n − k ) + v (n)
Motivation k =0
Signal model
Block
Y(ejω ) = H(ejω )X(ejω ) + V(ejω )
processing
Filterbank Mutual information of channel:
framework
ˆ π
1 p0
log IN + 2 H∗ (ejω )H(ejω ) dω
Mutual
information I(H) = I(X ; Y ) =
2πN −π σv
Simulations
Conclusion
Difficult to evaluate
14. Signal model
Mutual
Information
with filterbank
equalization
Consider M×N frequency selective LH tap MIMO
for MIMO
frequency
channel
selective
channels Signal model:
Vijaya Krishna LH −1
A, Shashank
V y [n] = H(k ) x(n − k ) + v (n)
Motivation k =0
Signal model
Block
Y(ejω ) = H(ejω )X(ejω ) + V(ejω )
processing
Filterbank Mutual information of channel:
framework
ˆ π
1 p0
log IN + 2 H∗ (ejω )H(ejω ) dω
Mutual
information I(H) = I(X ; Y ) =
2πN −π σv
Simulations
Conclusion
Difficult to evaluate
15. Signal model
Mutual
Information
with filterbank
equalization
Consider M×N frequency selective LH tap MIMO
for MIMO
frequency
channel
selective
channels Signal model:
Vijaya Krishna LH −1
A, Shashank
V y [n] = H(k ) x(n − k ) + v (n)
Motivation k =0
Signal model
Block
Y(ejω ) = H(ejω )X(ejω ) + V(ejω )
processing
Filterbank Mutual information of channel:
framework
ˆ π
1 p0
log IN + 2 H∗ (ejω )H(ejω ) dω
Mutual
information I(H) = I(X ; Y ) =
2πN −π σv
Simulations
Conclusion
Difficult to evaluate
16. Signal model
Mutual
Information
with filterbank
equalization
Consider M×N frequency selective LH tap MIMO
for MIMO
frequency
channel
selective
channels Signal model:
Vijaya Krishna LH −1
A, Shashank
V y [n] = H(k ) x(n − k ) + v (n)
Motivation k =0
Signal model
Block
Y(ejω ) = H(ejω )X(ejω ) + V(ejω )
processing
Filterbank Mutual information of channel:
framework
ˆ π
1 p0
log IN + 2 H∗ (ejω )H(ejω ) dω
Mutual
information I(H) = I(X ; Y ) =
2πN −π σv
Simulations
Conclusion
Difficult to evaluate
17. Block processing
Mutual
Information Block processing: Zero padding scheme
with filterbank
equalization ˜ ˜ ˜
y (n) = HP x (n) + v (n)
for MIMO
frequency
selective
channels
H(0) . . . H(LH − 1) 0 . . . 0
Vijaya Krishna
A, Shashank
.. .. .. .
.
V
0 . . . .
HP =
.. .. .. .
.
Motivation 0 . . . .
. .
Signal model . .. .. .. .
. . .
Block
. .
processing 0 ... H(0) · · · H(LH − 1)
Filterbank
framework M(P+LH -1) by NP Block Toeplitz matrix
Mutual
information P: no of input symbols per block
Simulations x (n) = [x T (Pn), x T (Pn − 1), ....., x T (P(n − 1) − 1)]T
˜
Conclusion
Results of flat fading channels can be used for block
processing
18. Block processing
Mutual
Information Block processing: Zero padding scheme
with filterbank
equalization ˜ ˜ ˜
y (n) = HP x (n) + v (n)
for MIMO
frequency
selective
channels
H(0) . . . H(LH − 1) 0 . . . 0
Vijaya Krishna
A, Shashank
.. .. .. .
.
V
0 . . . .
HP =
.. .. .. .
.
Motivation 0 . . . .
. .
Signal model . .. .. .. .
. . .
Block
. .
processing 0 ... H(0) · · · H(LH − 1)
Filterbank
framework M(P+LH -1) by NP Block Toeplitz matrix
Mutual
information P: no of input symbols per block
Simulations x (n) = [x T (Pn), x T (Pn − 1), ....., x T (P(n − 1) − 1)]T
˜
Conclusion
Results of flat fading channels can be used for block
processing
19. Block processing
Mutual
Information Block processing: Zero padding scheme
with filterbank
equalization ˜ ˜ ˜
y (n) = HP x (n) + v (n)
for MIMO
frequency
selective
channels
H(0) . . . H(LH − 1) 0 . . . 0
Vijaya Krishna
A, Shashank
.. .. .. .
.
V
0 . . . .
HP =
.. .. .. .
.
Motivation 0 . . . .
. .
Signal model . .. .. .. .
. . .
Block
. .
processing 0 ... H(0) · · · H(LH − 1)
Filterbank
framework M(P+LH -1) by NP Block Toeplitz matrix
Mutual
information P: no of input symbols per block
Simulations x (n) = [x T (Pn), x T (Pn − 1), ....., x T (P(n − 1) − 1)]T
˜
Conclusion
Results of flat fading channels can be used for block
processing
20. Mutual
Information
with filterbank
equalization For flat fading channel with channel matrix H, mutual
for MIMO
frequency information:
selective
channels 1 P0
Vijaya Krishna I(H) = log2 I + 2 H∗ H
A, Shashank N σv
V
Motivation Mutual information with zero padding:
Signal model
1 P0 ∗
Block IB (H) = log2 I + 2 HP HP
processing N(P + LH − 1) σv
Filterbank
framework
lim IB (HP ) = I(H)
Mutual P→∞
information
Simulations
Can be realized using joint ML detection at receiver
Conclusion
21. Mutual
Information
with filterbank
equalization For flat fading channel with channel matrix H, mutual
for MIMO
frequency information:
selective
channels 1 P0
Vijaya Krishna I(H) = log2 I + 2 H∗ H
A, Shashank N σv
V
Motivation Mutual information with zero padding:
Signal model
1 P0 ∗
Block IB (H) = log2 I + 2 HP HP
processing N(P + LH − 1) σv
Filterbank
framework
lim IB (HP ) = I(H)
Mutual P→∞
information
Simulations
Can be realized using joint ML detection at receiver
Conclusion
22. Mutual
Information
with filterbank
equalization For flat fading channel with channel matrix H, mutual
for MIMO
frequency information:
selective
channels 1 P0
Vijaya Krishna I(H) = log2 I + 2 H∗ H
A, Shashank N σv
V
Motivation Mutual information with zero padding:
Signal model
1 P0 ∗
Block IB (H) = log2 I + 2 HP HP
processing N(P + LH − 1) σv
Filterbank
framework
lim IB (HP ) = I(H)
Mutual P→∞
information
Simulations
Can be realized using joint ML detection at receiver
Conclusion
23. Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Optionally,
Vijaya Krishna
A, Shashank 1. Successive interference cancellation (MMSE-SIC)
V
Motivation 2. Eigenmode precoding
Signal model
Block
processing
May not be feasible. Suboptimal MMSE with symbol by
Filterbank
symbol detection used.
framework
Mutual
information
Simulations
Conclusion
24. Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Optionally,
Vijaya Krishna
A, Shashank 1. Successive interference cancellation (MMSE-SIC)
V
Motivation 2. Eigenmode precoding
Signal model
Block
processing
May not be feasible. Suboptimal MMSE with symbol by
Filterbank
symbol detection used.
framework
Mutual
information
Simulations
Conclusion
25. Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Optionally,
Vijaya Krishna
A, Shashank 1. Successive interference cancellation (MMSE-SIC)
V
Motivation 2. Eigenmode precoding
Signal model
Block
processing
May not be feasible. Suboptimal MMSE with symbol by
Filterbank
symbol detection used.
framework
Mutual
information
Simulations
Conclusion
26. Mutual
Information
with filterbank
equalization
for MIMO
Symbol by symbol detection:
frequency
selective
channels For the k th symbol, rate is
Vijaya Krishna
A, Shashank
V
B 1 1
Ik ,MMSE = log2
−1
Motivation N(P + LH − 1) p0 ∗
Signal model
I+ 2H H
σv P P k ,k
Block
processing
Filterbank Total rate is
framework
MP−1
Mutual
B 1 B
information IMMSE = Ik ,MMSE
Simulations
N(P + LH − 1)
k =0
Conclusion
27. Mutual
Information
with filterbank
equalization
for MIMO
Symbol by symbol detection:
frequency
selective
channels For the k th symbol, rate is
Vijaya Krishna
A, Shashank
V
B 1 1
Ik ,MMSE = log2
−1
Motivation N(P + LH − 1) p0 ∗
Signal model
I+ 2H H
σv P P k ,k
Block
processing
Filterbank Total rate is
framework
MP−1
Mutual
B 1 B
information IMMSE = Ik ,MMSE
Simulations
N(P + LH − 1)
k =0
Conclusion
28. Mutual
Information
with filterbank
equalization
for MIMO
Symbol by symbol detection:
frequency
selective
channels For the k th symbol, rate is
Vijaya Krishna
A, Shashank
V
B 1 1
Ik ,MMSE = log2
−1
Motivation N(P + LH − 1) p0 ∗
Signal model
I+ 2H H
σv P P k ,k
Block
processing
Filterbank Total rate is
framework
MP−1
Mutual
B 1 B
information IMMSE = Ik ,MMSE
Simulations
N(P + LH − 1)
k =0
Conclusion
29. Filterbank framework
Mutual
Information
with filterbank
equalization y(z) = H(z)x(z) + v(z)
for MIMO
frequency
selective
channels
˜ ˜ ˜
y (n) = Hx (n) + v (n)
Vijaya Krishna
A, Shashank
V
H(0) . . . H(LH − 1) 0 . . . 0
Motivation .. .. .. .
.
Signal model
0 . . . .
Block H=
.. .. .. .
.
processing 0 . . . .
. .
Filterbank
. .. .. .. .
. . .
framework
. .
Mutual 0 ... H(0) · · · H(LH − 1)
information
Simulations
MLF by N(LF +LH -1) block Toeplitz matrix
Conclusion
LF : Length of FIR filter used for equalization
30. Filterbank framework
Mutual
Information
with filterbank
equalization
z −d X(z) = F(z)Y(z)
for MIMO
frequency
selective
channels ˆ ˜ ˜
x (n − d) = FHx (n) + Fv (n)
Vijaya Krishna
A, Shashank
V MMSE inverse: FMMSE = Rxx Jd H∗ (HRx x H∗ + Rv v )−1
¯¯ ¯¯
Motivation Jd = [0N×Nd IN×N 0N×N(LH +LF −d−2) ]
Signal model
Block
processing 0 ··· 0 1 ··· 0 0 ··· 0
. .. . .. .. .. .. .. .
Filterbank
framework
Jd = .
. . .
. . . . . . .
.
Mutual 0 ··· 0 0 ··· 1 0 ··· 0
information
Simulations
2
If Rxx = I and Rv v = σv I
Conclusion
¯¯
FMMSE = Jd H∗ (HH∗ + σv I)−1
2
31. Filterbank framework
Mutual
Information
with filterbank
equalization
z −d X(z) = F(z)Y(z)
for MIMO
frequency
selective
channels ˆ ˜ ˜
x (n − d) = FHx (n) + Fv (n)
Vijaya Krishna
A, Shashank
V MMSE inverse: FMMSE = Rxx Jd H∗ (HRx x H∗ + Rv v )−1
¯¯ ¯¯
Motivation Jd = [0N×Nd IN×N 0N×N(LH +LF −d−2) ]
Signal model
Block
processing 0 ··· 0 1 ··· 0 0 ··· 0
. .. . .. .. .. .. .. .
Filterbank
framework
Jd = .
. . .
. . . . . . .
.
Mutual 0 ··· 0 0 ··· 1 0 ··· 0
information
Simulations
2
If Rxx = I and Rv v = σv I
Conclusion
¯¯
FMMSE = Jd H∗ (HH∗ + σv I)−1
2
32. Filterbank framework
Mutual
Information
with filterbank
equalization
z −d X(z) = F(z)Y(z)
for MIMO
frequency
selective
channels ˆ ˜ ˜
x (n − d) = FHx (n) + Fv (n)
Vijaya Krishna
A, Shashank
V MMSE inverse: FMMSE = Rxx Jd H∗ (HRx x H∗ + Rv v )−1
¯¯ ¯¯
Motivation Jd = [0N×Nd IN×N 0N×N(LH +LF −d−2) ]
Signal model
Block
processing 0 ··· 0 1 ··· 0 0 ··· 0
. .. . .. .. .. .. .. .
Filterbank
framework
Jd = .
. . .
. . . . . . .
.
Mutual 0 ··· 0 0 ··· 1 0 ··· 0
information
Simulations
2
If Rxx = I and Rv v = σv I
Conclusion
¯¯
FMMSE = Jd H∗ (HH∗ + σv I)−1
2
33. Mutual information
Mutual
Information
with filterbank Idea is that error vector is orthogonal to the estimate
equalization
for MIMO and
frequency
selective ˆ
X = Axy Y + X⊥Y = X + E
channels
Vijaya Krishna
A, Shashank
V ˆ −1
X = X|Y = Axy Y = Rxy Ryy Y
Motivation
Signal model
|Rxx |
Block IF (H) = log2
processing |Ree |
Filterbank
framework
Mutual
Theorem
information
Simulations
1 |Rxx |
Conclusion IF (H) = log2 ∗ (HR H∗ + R )−1 HJ R |
N |Rxx − Rxx Jd H ¯¯
xx ¯¯
vv d xx
34. Mutual information
Mutual
Information
with filterbank Idea is that error vector is orthogonal to the estimate
equalization
for MIMO and
frequency
selective ˆ
X = Axy Y + X⊥Y = X + E
channels
Vijaya Krishna
A, Shashank
V ˆ −1
X = X|Y = Axy Y = Rxy Ryy Y
Motivation
Signal model
|Rxx |
Block IF (H) = log2
processing |Ree |
Filterbank
framework
Mutual
Theorem
information
Simulations
1 |Rxx |
Conclusion IF (H) = log2 ∗ (HR H∗ + R )−1 HJ R |
N |Rxx − Rxx Jd H ¯¯
xx ¯¯
vv d xx
35. Proof
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
I(X ; Y ) = h(X ) − h(X |Y )
channels
Vijaya Krishna
A, Shashank
For the MMSE equalizer, h(X |Y ) = h(E), the entropy of
V
the error vector
Motivation 1 |Rxx |
IF (H) = h(X ) − h(E) = log2
Signal model
N |Ree |
Block
processing
Filterbank Ree = E{xx ∗ } − E{x y ∗ }E{y y ∗ }E{y x ∗ }
˜ ˜˜ ˜
framework
Mutual
information
Ree = Rxx − Rxx Jd H∗ (HRx x H∗ + Rv v )−1 HJd Rxx
¯¯ ¯¯
Simulations
Conclusion
36. Proof
Mutual
Information
with filterbank 2
If Rxx = p0 I and Ree = σv I then
equalization
for MIMO
frequency
selective
Ree = p0 I − p0 Jd H∗ (p0 HH∗ + σv I)−1 HJd p0
2
channels
Vijaya Krishna
A, Shashank
Using matrix inversion lemma,
p
V Ree = p0 Jd (I + σ0 H∗ H)−1 J∗
2 d
v
Motivation
1 1
Signal model IF (H) = log2 p0 ∗ −1 J∗ |
N |Jd (I + 2 H H) d
Block σv
processing
Filterbank
framework For the case of symbol by symbol detection,
Mutual N−1
information 1 1
Simulations
IF (H) = log2
N p0 ∗ −1 J∗
Conclusion k =0 Jd (I + 2 H H)
σv d k ,k
37. Proof
Mutual
Information
with filterbank 2
If Rxx = p0 I and Ree = σv I then
equalization
for MIMO
frequency
selective
Ree = p0 I − p0 Jd H∗ (p0 HH∗ + σv I)−1 HJd p0
2
channels
Vijaya Krishna
A, Shashank
Using matrix inversion lemma,
p
V Ree = p0 Jd (I + σ0 H∗ H)−1 J∗
2 d
v
Motivation
1 1
Signal model IF (H) = log2 p0 ∗ −1 J∗ |
N |Jd (I + 2 H H) d
Block σv
processing
Filterbank
framework For the case of symbol by symbol detection,
Mutual N−1
information 1 1
Simulations
IF (H) = log2
N p0 ∗ −1 J∗
Conclusion k =0 Jd (I + 2 H H)
σv d k ,k
38. Mutual
Information
with filterbank
equalization
for MIMO 2
If Rxx = p0 I and Ree = σv I then
frequency
selective
channels
Vijaya Krishna 1 1
A, Shashank
V
IF (H) = log2 p0 ∗ −1 J∗ |
N |Jd (I + 2 H H) d
σv
Motivation
Signal model
Block
For the case of symbol by symbol detection,
processing
N−1
Filterbank 1 1
framework IF (H) = log2
N p0 ∗ −1 J∗
Mutual k =0 Jd (I + 2 H H)
σv d k ,k
information
Simulations
Conclusion
39. Mutual
Information
with filterbank
equalization
for MIMO 2
If Rxx = p0 I and Ree = σv I then
frequency
selective
channels
Vijaya Krishna 1 1
A, Shashank
V
IF (H) = log2 p0 ∗ −1 J∗ |
N |Jd (I + 2 H H) d
σv
Motivation
Signal model
Block
For the case of symbol by symbol detection,
processing
N−1
Filterbank 1 1
framework IF (H) = log2
N p0 ∗ −1 J∗
Mutual k =0 Jd (I + 2 H H)
σv d k ,k
information
Simulations
Conclusion
40. Mutual
Information
with filterbank
equalization Observation
for MIMO
frequency 1 1
selective IF (H) = log2 p0 ∗ −1 J∗ |
channels N |Jd (I + 2 H H) d
Vijaya Krishna
σv
A, Shashank
V 1 1
IB (H) = log2 −1
Motivation N(P + LH − 1) p0 ∗
I+ 2H H
σv P P
Signal model
Block
processing
Filterbank Remark
framework
The MI for filterbank equalization depends on the
Mutual p
information determinant of N by N submatrix of (I + σ0 H∗ H)−1 . So we
2
v
Simulations
can choose the delay so as to maximize MI
Conclusion
41. Mutual
Information
with filterbank
equalization Observation
for MIMO
frequency 1 1
selective IF (H) = log2 p0 ∗ −1 J∗ |
channels N |Jd (I + 2 H H) d
Vijaya Krishna
σv
A, Shashank
V 1 1
IB (H) = log2 −1
Motivation N(P + LH − 1) p0 ∗
I+ 2H H
σv P P
Signal model
Block
processing
Filterbank Remark
framework
The MI for filterbank equalization depends on the
Mutual p
information determinant of N by N submatrix of (I + σ0 H∗ H)−1 . So we
2
v
Simulations
can choose the delay so as to maximize MI
Conclusion
42. Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Vijaya Krishna
A, Shashank
V
Motivation
Signal model
Block
processing
Filterbank
framework
Mutual
information
Simulations
Conclusion p0 ∗ −1
Choose submatrix of (I + 2 H H)
σv
with lowest
determinant
43. Simulations
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Vijaya Krishna
A, Shashank 4×3 Rayleigh fading channels of length LH = 8
V
Block processing case: no of inputs symbols per block
Motivation
P = 20
Signal model
Block Filterbank case: Length of equalizer LF = 21
processing
Filterbank
framework
Mutual
information
Simulations
Conclusion
44. Simulations
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Vijaya Krishna
A, Shashank
V
Motivation
Signal model
Block
processing
Filterbank
framework
Mutual
information
Simulations
Conclusion Figure: Comparison between block processing and Filterbank
equalizers
45. Simulations
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Vijaya Krishna
A, Shashank
V
Motivation
Signal model
Block
processing
Filterbank
framework
Mutual
information
Simulations
Conclusion Figure: MI with variation in delay. SNR=15 dB
46. Simulations
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Vijaya Krishna
A, Shashank
V
Motivation
Signal model
Block
processing
Filterbank
framework
Mutual
information
Simulations
Conclusion
Figure: MI for different LF ’s. SNR=15 dB
47. Conclusion
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective Filterbank equalization achieves significantly higher
channels
information rate when compared to block processing
Vijaya Krishna
A, Shashank
V
We have the flexibility of choosing the delay so as to
Motivation maximize MI
Signal model
Block Disadvantage of this scheme: Processing complexity,
processing
Filterbank
similar to BP
framework
Mutual Future: Mutual information using zero forcing equalizers
information
Simulations
Conclusion
48. Conclusion
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective Filterbank equalization achieves significantly higher
channels
information rate when compared to block processing
Vijaya Krishna
A, Shashank
V
We have the flexibility of choosing the delay so as to
Motivation maximize MI
Signal model
Block Disadvantage of this scheme: Processing complexity,
processing
Filterbank
similar to BP
framework
Mutual Future: Mutual information using zero forcing equalizers
information
Simulations
Conclusion
49. Conclusion
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective Filterbank equalization achieves significantly higher
channels
information rate when compared to block processing
Vijaya Krishna
A, Shashank
V
We have the flexibility of choosing the delay so as to
Motivation maximize MI
Signal model
Block Disadvantage of this scheme: Processing complexity,
processing
Filterbank
similar to BP
framework
Mutual Future: Mutual information using zero forcing equalizers
information
Simulations
Conclusion
50. Conclusion
Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective Filterbank equalization achieves significantly higher
channels
information rate when compared to block processing
Vijaya Krishna
A, Shashank
V
We have the flexibility of choosing the delay so as to
Motivation maximize MI
Signal model
Block Disadvantage of this scheme: Processing complexity,
processing
Filterbank
similar to BP
framework
Mutual Future: Mutual information using zero forcing equalizers
information
Simulations
Conclusion
51. References
Mutual
Information
with filterbank
equalization
for MIMO
frequency Vijaya Krishna. A, A filterbank precoding framework for
selective
channels MIMO frequency selective channels, PhD thesis, Indian
Vijaya Krishna Institute of Science, 2006.
A, Shashank
V
G. D. Forney Jr., “Shannon meets Wiener II: On MMSE
Motivation estimation in successive decoding schemes,” In Proc.
Signal model Allerton Conf., Sep. 2004.
Block
processing
(http://arxiv.org/abs/cs/0409011)
Filterbank
framework
X. Zhang and S.-Y. Kung, “Capacity analysis for parallel
Mutual and sequential MIMO equalizers,” IEEE Trans on Signal
information
Processing, vol. 51, pp. 2989- 3002, Nov. 2003.
Simulations
Conclusion
52. References
Mutual
Information
with filterbank
equalization
for MIMO
frequency P. P. Vaidyanathan, Multirate systems and filter banks,
selective
channels Englewood Cliffs, NJ: Prentice-Hall, 1993.
Vijaya Krishna
A, Shashank Vijaya Krishna. A, K. V. S. Hari, ”Filterbank precoding for
V
FIR equalization in high rate MIMO communications,”
Motivation IEEE Trans. Signal Processing, vol. 54, No. 5, pp.
Signal model 1645-1652, May 2006.
Block
processing
A. Scaglione, S. Barbarossa, and G. B, Giannakis,
Filterbank
framework “Filterbank transceivers optimizing information rate in
Mutual block transmissions over dispersive channels,” IEEE
information
Trans. Info. Theory, Vol. 45, pp. 1019-1032, Apr. 1999.
Simulations
Conclusion
53. Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Vijaya Krishna
A, Shashank
V
Motivation THANK YOU
Signal model
Block
processing
Filterbank
framework
Mutual
information
Simulations
Conclusion
54. Mutual information
Mutual
Information
with filterbank
equalization
for MIMO
frequency I(X ; Y ) = h(X ) − h(X |Y )
selective
channels
Vijaya Krishna For the MMSE equalizer, h(X |Y ) = h(E), the entropy of
A, Shashank
V the error vector
Motivation
Signal model
1 |Rxx |
IF (H) = h(X ) − h(E) = log2
Block N |Ree |
processing
Ree = E{xx ∗ } − E{x y ∗ }E{y y ∗ }E{y x ∗ }
Filterbank
framework ˜ ˜˜ ˜
Mutual
information
Simulations Ree = Rxx − Rxx Jd H∗ (HRx x H∗ + Rv v )−1 HJd Rxx
¯¯ ¯¯
Conclusion
55. Mutual information
Mutual
Information
with filterbank
equalization
for MIMO
frequency I(X ; Y ) = h(X ) − h(X |Y )
selective
channels
Vijaya Krishna For the MMSE equalizer, h(X |Y ) = h(E), the entropy of
A, Shashank
V the error vector
Motivation
Signal model
1 |Rxx |
IF (H) = h(X ) − h(E) = log2
Block N |Ree |
processing
Ree = E{xx ∗ } − E{x y ∗ }E{y y ∗ }E{y x ∗ }
Filterbank
framework ˜ ˜˜ ˜
Mutual
information
Simulations Ree = Rxx − Rxx Jd H∗ (HRx x H∗ + Rv v )−1 HJd Rxx
¯¯ ¯¯
Conclusion
56. Mutual information
Mutual
Information
with filterbank
equalization
for MIMO
frequency I(X ; Y ) = h(X ) − h(X |Y )
selective
channels
Vijaya Krishna For the MMSE equalizer, h(X |Y ) = h(E), the entropy of
A, Shashank
V the error vector
Motivation
Signal model
1 |Rxx |
IF (H) = h(X ) − h(E) = log2
Block N |Ree |
processing
Ree = E{xx ∗ } − E{x y ∗ }E{y y ∗ }E{y x ∗ }
Filterbank
framework ˜ ˜˜ ˜
Mutual
information
Simulations Ree = Rxx − Rxx Jd H∗ (HRx x H∗ + Rv v )−1 HJd Rxx
¯¯ ¯¯
Conclusion
57. Mutual information
Mutual
Information
with filterbank
equalization
for MIMO
frequency I(X ; Y ) = h(X ) − h(X |Y )
selective
channels
Vijaya Krishna For the MMSE equalizer, h(X |Y ) = h(E), the entropy of
A, Shashank
V the error vector
Motivation
Signal model
1 |Rxx |
IF (H) = h(X ) − h(E) = log2
Block N |Ree |
processing
Ree = E{xx ∗ } − E{x y ∗ }E{y y ∗ }E{y x ∗ }
Filterbank
framework ˜ ˜˜ ˜
Mutual
information
Simulations Ree = Rxx − Rxx Jd H∗ (HRx x H∗ + Rv v )−1 HJd Rxx
¯¯ ¯¯
Conclusion
58. Mutual information
Mutual
Information
with filterbank 2
If Rxx = p0 I and Ree = σv I then
equalization
for MIMO
frequency
selective
Ree = p0 I − p0 Jd H∗ (p0 HH∗ + σv I)−1 HJd p0
2
channels
Vijaya Krishna
A, Shashank
Using matrix inversion lemma,
p
V Ree = p0 Jd (I + σ0 H∗ H)−1 J∗
2 d
v
Motivation
1 1
Signal model IF (H) = log2 p0 ∗ −1 J∗ |
N |Jd (I + 2 H H) d
Block σv
processing
Filterbank
framework For the case of symbol by symbol detection,
Mutual N−1
information 1 1
Simulations
IF (H) = log2
N p0 ∗ −1 J∗
Conclusion k =0 Jd (I + 2 H H)
σv d k ,k
59. Mutual information
Mutual
Information
with filterbank 2
If Rxx = p0 I and Ree = σv I then
equalization
for MIMO
frequency
selective
Ree = p0 I − p0 Jd H∗ (p0 HH∗ + σv I)−1 HJd p0
2
channels
Vijaya Krishna
A, Shashank
Using matrix inversion lemma,
p
V Ree = p0 Jd (I + σ0 H∗ H)−1 J∗
2 d
v
Motivation
1 1
Signal model IF (H) = log2 p0 ∗ −1 J∗ |
N |Jd (I + 2 H H) d
Block σv
processing
Filterbank
framework For the case of symbol by symbol detection,
Mutual N−1
information 1 1
Simulations
IF (H) = log2
N p0 ∗ −1 J∗
Conclusion k =0 Jd (I + 2 H H)
σv d k ,k
60. Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Observation
Vijaya Krishna
A, Shashank 1 1
V IF (H) = log2 p0 ∗ −1 J∗ |
N |Jd (I + 2 H H) d
Motivation σv
Signal model 1 1
Block IB (H) = log2
N(P + LH − 1) p0 ∗
I+ 2H H
processing
σv P P
Filterbank
framework
Mutual
information
Simulations
Conclusion
61. Mutual
Information
with filterbank
equalization
for MIMO
frequency
selective
channels
Observation
Vijaya Krishna
A, Shashank 1 1
V IF (H) = log2 p0 ∗ −1 J∗ |
N |Jd (I + 2 H H) d
Motivation σv
Signal model 1 1
Block IB (H) = log2
N(P + LH − 1) p0 ∗
I+ 2H H
processing
σv P P
Filterbank
framework
Mutual
information
Simulations
Conclusion