16.

548 Coding and Information

Theory
Lecture 15: Space Time Coding and
MIMO:

1
Credits

2
Wireless Channels

3
Signal Level in Wireless Transmission

4
Classification of Wireless Channels

5
Space time Fading, narrow beam

6
Space Time Fading: Wide Beam

7
Introduction to the MIMO Channel

8
Capacity of MIMO Channels

9
10
 Single Input- Single Output
systems (SISO)
x(t): transmitted signal
g y(t): received signal
g(t): channel transfer function
n(t): noise (AWGN, 2)
y(t)
x(t)

y(t) = g • x(t) + n(t)

Ex 2
Signal to noise ratio : ρ  g 2
σ
Capacity : C = log2(1+) 11
Single Input- Multiple Output (SIMO
Multiple Input- Single Output (MISO
• Principle of diversity systems (transmitter/ receiver)
• +: Higher average signal to noise ratio
Robustness
• - : Process of diminishing return
Benefit reduces in the presence of correlation
• Maximal ratio combining >
Equal gain combining >
Selection combining
12
 Idea behind diversity systems
• Use more than one copy of the same signal
• If one copy is in a fade, it is unlikely that all the others
will be too.

1
C1xN  log 2 (1  N )
N
• C1xN>C1x1
• C1xN more robust than C1x1
13
Background of Diversity Techniques
• Variety of Diversity techniques are proposed to combat Time-Varying
Multipath fading channel in wireless communication
– Time Diversity
– Frequency Diversity
– Space Diversity (mostly multiple receive antennas)
• Main intuitions of Diversity:
– Probability of all the signals suffer fading is less then probability of single
signal suffer fading
– Provide the receiver a multiple versions of the same Tx signals over
independent channels
• Time Diversity
– Use different time slots separated by an interval longer than the coherence
time of the channel.
– Example: Channel coding + interleaving
– Short Coming: Introduce large delays when the channel is in slow fading

14
 Diversity Techniques
• Improve the performance in a fading environment
– Space Diversity
• Spacing is important! (coherent distance)
– Polarization Diversity
• Using antennas with different polarizations for
reception/transmission.
– Frequency Diversity
• RAKE receiver, OFDM, equalization, and etc.
• Not effective over frequency-flat channel.
– Time Diversity
• Using channel coding and interleaving.
• Not effective over slow fading channels.

15
RX Diversity in Wireless

16
Receive Diversity

17
Selection and Switch Diversity

18
Linear Diversity

19
Receive Diversity Performance

20
Transmit Diversity

21
Transmit Diversity with Feedback

22
TX diversity with frequency weighting

23
TX Diversity with antenna hopping

24
TX Diversity with channel coding

25
Transmit diversity via delay diversity

26
Transmit Diversity Options

27
 MIMO Wireless Communications:
Combining TX and RX Diversity
• Transmission over Multiple Input Multiple Output (MIMO)
radio channels

Data
symbols
d_hat
Data Space-Time
d symbols Decoder
N Data Space-Time Wireless Channel
Symbols Encoder
(What a Big Cloud!)

L_r

symbols
Receive

Pilot
L_t
symbols

Transmit antennas
Pilot

antennas
P
P

• Advantages: Improved Space Diversity and Channel Capacity

• Disadvantages: More complex, more radio stations and required
channel estimation
28
 MIMO Model

T: Time index
W: Noise
YN T  H N M  X M T  WN T
• Matrix Representation

– For a fixed T
29
Part II: Space Time Coding

30
 Multiple Input- Multiple Output
H11
systems (MIMO)
1 1
HN1 y Nx1  H NxM x Mx1  n Nx1
H
H1M
M N
HNM

• Average gain   E H ij
2
  2 1
,H  H

Ptotal 2
• Average signal to noise ratio  2 
 31
 Shannon capacity
  Ex H    PT 2 H 
C  log2 det  I  2 HH   log2 det  I  2
g HH  
  σ    Mσ 
  ρ H 
log2 det  I  HH 
  M 

K= rank(H): what is its range of values?

Parameters that affect the system capacity
• Signal to noise ratio 
• Distribution of eigenvalues (u) of H

32
 Interpretation I:
The parallel channels approach

• “Proof” of capacity formula

• Singular value decomposition of H: H = S·U·VH

• S, V: unitary matrices (VHV=I, SSH =I)
U : = diag(uk), uk singular values of H

• V/ S: input/output eigenvectors of H
• Any input along vi will be multiplied by ui and will appear as an output along si

33
 Vector analysis of the signals
1. The input vector x gets projected onto the vi’s
2. Each projection gets multiplied by a different
gain ui.
3. Each appears along a different si.
u1
<x,v1> · v1 <x,v1> u1 s1
u2
<x,v2> · v2 <x,v2> u2 s2

uK
<x,vK> · vK <x,vK> uK sK 34
 Capacity = sum of capacities
• The channel has been decomposed into K
parallel subchannels
• Total capacity = sum of the subchannel
capacities
• All transmitters send the same power:
Ex=Ek
K K
C   Ck   log2 1  ρ k  K
 Ek 2
C   log 2 1  2 uk 
i 1 i 1
  
u E x, v  u E
i 1
2 2 2

E  n, s 
k k
ρk   2
k
2
k

k
σ 35
 Interpretation II:
The directional approach
• Singular value decomposition of H: H = S·U·VH
• Eigenvectors correspond to spatial directions
(beamforming)
1 1 (si)1

(si)N
M N
36
Example of directional interpretation

37
38
 Space-Time Coding
• What is Space-Time
Coding?
– Space diversity at antenna
– Time diversity to introduce
redundant data
• Alamouti-Scheme
– Simple yet very effective
– Space diversity at
transmitter end
– Orthogonal block code
design

39
Space Time Coded Modulation

40
Space Time Channel Model

41
42
STC Error Analysis

43
STC Error Analysis

44
45
46
STC Design Criteria

47
48
STC 4-PSK Example

49
STC 8-PSK Example

50
STC 16-QAM Example

51
STC Maximum Likelihood Decoder

52
STC Performance with perfect CSI

53
54
55
Delay Diversity

56
Delay Diversity ST code

57
58
Space Time Block Codes (STBC)

59
Decoding STBC

60
61
62
 Block and Data Model
• 1X(N+P) block of information symbols broadcast from
transmit antenna: i
Function (mapping) S at antenna
Si(d, t) i defines the Space-Time
encoding process

• 1X(N+P) block of received information symbols taken from

antenna: j
Rj = hjiSi(d, t) + nj Assuming a single user and
quasi-static and independent
• Matrix representation: fading, n is AWGN

r  h11
h12
. . . h 1 Lt  S (d , t ) 
h 
1 1

. 
h . . . h

S (d , t ) 
21 22 2 Lt
 
 
2

 . . . .   . 
R   .   HS  N H    S   
 . . . . .
.   
   . . . .   . 
   
r 
Lt hLr 1
hLr 2
. . . h Lr Lt  S Lt
(d , t )
63
 Related Issues
Data Data
d symbols d_hat
Space-Time symbols
Encoder Space-Time
N Data
S_i(d,t) Decoder
Symbols
i=1,2,…,K Wireless Channel
(What a Big Cloud!)

L_t L_r
symbols

symbols
Transmit Receive
Pilot

Pilot
antennas antennas

P P

• How to define Space-Time mapping Si(d,t) for diversity/channel

capacity trade-off?
• What is the optimum sequence for pilot symbols?
• How to get “best estimated” Channel State Information (CSI) from the
pilot symbols P?
• How to design frame structure for Data symbols (Payload) and Pilot
symbols such that most optimum for FER and BER?
64
 Specific Example of STBC:
Alamouti’s Orthogonal Code
• Let’s consider two antenna i and i+1 at the transmitter side, at two
consecutive time instants t and t+T:
Space

Ant. i Ant i+1

Time t
d_0 d_1
Time

Time t+T

-d_1* d_0*

• The above Space-Time mapping defines Alamouti’s Code[1].

• A general frame design requires concatenation of blocks (each 2X2)
of Alamouti code,  do d1 | d 2 d 3 | . . .
D  d * | . . .
* * *
 1 do |  d3 d2
65
Estimated Channel State Information
(CSI)
• Pilot Symbol Assisted Modulation (PSAM)
[3] is used to obtain estimated Channel
State Information (CSI)
• PSAM simply samples the channel at a rate
greater than Nyquist rate,so that
reconstruction is possible
• Here is how it works…

66
 Channel State Estimation
Frame size = 300

L_t = i P 5P P 5D P 2P 3D P 5D P 2P 3D P 2P 3D P 5D …... 5D P 3D 2P P 3D 2P P …... P 5P

1 7 13 19 25 31 37 265 271 277 295
uniform
edge part edge part
part

A typical slow fading

channel

67
 Estimated CSI (cont.d)
Block diagram of the receiver
Lt
 h( l ) ( t ) s ( l ) ( t ) r_k+1,r_k+2,…,r_k+K
l 1 r_k
r(t) Matched Filter A/D ML D_hat
+ Delay
u*(-t) Converter Decoder

Pilot Symbol
Extractor h_hat
n(t)

r_p_1, r_p_2, …, r_p_N

Channel State
Estimator

68
 Channel State Estimation (cont.d)

Frame size = 300

L_t = i P 5P P 5D P 2P 3D P 5D P 2P 3D P 2P 3D P 5D …... 5D P 3D 2P P 3D 2P P …... P 5P

1 7 13 19 25 31 37 265 271 277 295
uniform
edge part edge part
part

• Pilot symbol insertion length, Pins=6.

• The receiver uses N=12, nearest pilots to obtain estimated
CSI
69
 Channel State Estimation Cont.d
• Pilot Symbols could be think of as redundant data symbols

• Pilot symbol insertion length will not change the performance much, as long
as we sample faster than fading rate of the channel

• If the channel is in higher fading rate, more pilots are expected to be inserted

70
 Estimated CSI, Space-time PSAM
frame design Frame size = 300

L_t = i P 5P P 5D P 2P 3D P 5D P 2P 3D P 2P 3D P 5D …... 5D P 3D 2P P 3D 2P P …... P 5P

1 7 13 19 25 31 37 265 271 277 295
uniform
edge part edge part
part

Frame size = 300

L_t = i+1 P 5P P 5D P 2P 3D P 5D P 2P 3D P 2P 3D P 5D …... 5D P 3D 2P P 3D 2P P …... P 5P

1 7 13 19 25 31 37 265 271 277 295
uniform
edge part edge part
part
• The orthogonal pilot symbol (pilots chosen from QPSK constellation)
matrix is, [4] 1 1  
P 
1  1
• Pilot symbol insertion length, Pins=6.
• The receiver uses N=12, nearest pilots to obtain estimated CSI
• Data = 228, Pilots = 72
71
 Channel State Estimation (cont.d)
MMSE estimation
• Use Wiener filtering, since it is a Minimum Mean Square
Error (MMSE) estimator hˆ  E[ h | r ] p

• All random variables involved are jointly Gaussian,

MMSE estimator becomes a linear minimum mean square
estimator [2]:
hˆ  E[hr ]Cov (r ) r  Wr
H 1

p p p p

• Wiener filter is defined as, W  E[hr . ]Cov(r )

H 1
p p

• Note, and Cov ( h )  E[ h  h ]

Cov(r )   Cov( h )  N I
H H

p p p p o p p p

72
 Block diagram for MRRC scheme
with two Tx and one Rx
d_0 d_1
-d_1* d_0*
tx antenna 0 tx antenna 1

h_0(t) h_1(t)
h_0(t+T) h_1(t+T) Space

Ant. i Ant i+1

rx antenna

Time t
d_0 d_1
n_0 Interference
+
n_1 & noise

Time

Time t+T
-d_1* d_0*
h_hat_0
channel
combiner
estimator h_hat_1

h_hat_0 h_hat_1 d_hat_0 d_hat_1

maximum likelihood detector

73
 Block diagram for MRRC scheme
with two Tx and one Rx
• The received signals can then be expressed as,
ro  r (t )  ho (t )d o  h1 (t )d1  no

r1  r (t  T )  ho (t  T )d1  h1 (t  T )d 0  n1
* *

• The combiner shown in the above graph builds the following two
estimated signal
dˆ0  [h0 (t  T )r0  h1 (t )r1 ]*
* *

dˆ1  [h1 (t  T )r0  h0 (t )r1 ]*

* *

74
 Maximum Likelihood Decoding
Under QPSK Constellation

• Output of the combiner could be further simplified and could be expressed as

follows:
dˆ0  [h0 (t )h0 (t  T )  h1 (t )h1 (t  T )]d 0  h0 (t  T )n0  h1 (t )n1
* * * *

dˆ1  [h0 (t )h0 (t  T )  h1 (t )h1 (t  T )]d1  h1 (t  T )n0  h0 (t )n1

* * * *

• For example, under QPSK constellation decision are made according to the
axis.

75
Space-Time Alamouti Codes with
Perfect CSI,BPSK Constellation

76
Space-Time Alamouti Codes with
PSAM under QPSK Constellation

77
Space-Time Alamouti Codes with
PSAM under QPSK Constellation

78
 Performance metrics
• Measures of comparison • Eigenvalue distribution
– Gaussian iid channel
– “ideal” channel • Shannon capacity
– for constant SNR or
– for constant transmitted power

• Effective degrees of
freedom(EDOF)

• Condition number

79
 Measures of comparison
Gaussian Channel “Ideal” channel (max C)
Hij =xij+jyij : x,y i.i.d. rank(H) = min(M, N)
Gaussian random variables |u1 | = |u2 | = … = |uK |
Problem: poutage

80
 Eigenvalue
1 distribution
1

H
M N

Ideally: Limits
As high gain as possible Power constraints
As many eigenvectors as possible System size
As orthogonal as possible Correlation
81
 Example:
Uncorrelated & correlated channels

82
 Shannon capacity
• Capacity for a reference SNR (only channel
  ρ ref
info) H 
C  log2 det  I 
 HH  
  M 

• Capacity
  for constant transmitted power
Ex H 
C  (channel
log 2 det  I+power
2
HHroll-off
 info)
  σ 
83
 Building layout 4m
RCVR
(lab)
6m

RCVR XMTR
2m
90o (hall)

180o 0o 3.3m

270o 3.3m 84
 LOS conditions: Higher average SNR, High correlation
Non-LOS conditions: Lower average SNR,More scattering

4m
RCVR
(lab)
6m

XMTR
2m
90o

180o 0o 3.3m
85
270 o
3.3m
Example: C for reference SNR

86
Example: C for constant transmit pwr

87
 EDOFOther metrics Condition
(Effective degrees of number
freedom)

C 2δ ρ ref  p outage
Definition lim δ UMAX/ umin
δ0 2 ρ ref
+ Intuition Simplicity

- Dependence on No information
reference SNR on intermediate
eigenvalue
distribution 88
 From narrowband to wideband
• Wideband: delay spread >> symbol time

• -: Intersymbol interference
+: Frequency diversity
L
g(t)   g lδ t  τ l 
• SISO channel impulse response: l 1

 g
2
E  L

SISO capacity:C  log2 1  2 , g   gl

 x  2 2

 σ  l 1
89
Matrix formulation of wideband case
L
h ij  t    h ij,lδ t  τ l 
l 1

y  t   H  x  t   n (t )
 y1 (t )   H11  H1M   x1 (t ) 
      
       n(t )
 
 y N (t ) H N 1  H NM   x M (t )

90
 Equivalent treatment in the
frequency domain
• Wideband channel = Many narrowband
channels
H(t)  H(f)
Noise level

f
  Ex H 
C NB  log2 det I  2 HH , σ 2  N 0 (BW)
  σ 
  Ex ( f ) H

CWB  C NB   log2 det  I 
N0
H ( f )H ( f ) 
bandwidth bandwidth   
91
 Extensions
• Optimal power allocation
• Optimal rate allocation
• Space-time codes
• Distributed antenna systems

• Many, many, many more!

92
 Optimal power allocation
• IF the transmitter knows the channel, it can
allocate
K
power
E so as
2 toK
maximize capacity
C   log2 1  2 u k ,  E k  Ptotal
k

i 1  σ  i 1

1 
Ek  (ν  )
• Solution: Waterfilling
λk
2
uk
λk  2
σ 93
 Illustration of waterfilling algorithm


1 
Ek  (ν  )
λk
2
uk
λk  2
σ

Stronger subchannels get the most power

94
• Criterion: Shannon capacity maximization
Discussion on waterfilling
(All the SISO discussion on coding, constellation
limitations etc is pertinent)

• Benefit depends on the channel, available power etc.

Correlation, available power  Benefit 

• Limitations:
– Waterfilling requires feedback link
– FDD/ TDD
– Channel state changes

95
 Optimal rate allocation
• Similar to optimal power allocation
• Criterion:
K throughput (T) maximization
C   log2 1  ρ k  (bps/Hz)
k 1
K
T   Bk (b/Hz)
k 1
• Bk : bits per symbol (depends on constellation size)
• Idea: for a given k, find maximum Bk for a target
probability of error Pe
96
Discussion on optimal rate allocation

• Possible limits on constellation sizes!

• Constellation sizes are quantized!!!
• The answer is different for different target
probabilities of error
• Optimal power AND rate allocation
schemes possible, but complex

97
 Distributed antenna systems
• Idea: put your antennas in different places
• +: lower correlation
- : power imbalance, synchronization,
coordination

98
 Practical considerations
• Coding
• Detection algorithms
• Channel estimation
• Interference

99
 Detection
• Maximum likelihood linear detector
algorithms
y = H x + n  xest = H+y
H+ = (HH H)-1 HH : Pseudo inverse of H

• Problem: find nearest neighbor among QM points

(Q: constellation size, M: number of transmitters)

• VERY high complexity!!!

100
 Solution: BLAST algorithm
• BLAST: Bell Labs lAyered Space Time

• Idea: NON-LINEAR DETECTOR

– Step 1: H+ = (HH H)-1 HH
– Step 2: Find the strongest signal
(Strongest = the one with the highest post detection SNR)
– Step 3: Detect it (Nearest neighbor among Q)
– Step 4: Subtract it
– Step 5: if not all yet detected, go to step 2

101
 Discussion on the BLAST algorithm
• It’s a non-linear detector!!!

• Two flavors
– V-BLAST (easier)
– D-BLAST (introduces space-time coding)

• Achieves 50-60% of Shannon capacity

• Error propagation possible

• Very complicated for wideband case

102
 Codingachievable
• Capacity = Maximum limitations
data rate that can
be achieved over the channel with arbitrarily low
probability of error
• SISO case:
– Constellation limitations
– Turbo- coding can get you close to Shannon!!!
• MIMO case:
– Constellation limitations as well
– Higher complexity
– Space-time codes: very few!!!!
103
 Channel estimation
• The channel is not perfectly estimated because
– it is changing (environment, user movement)
– there is noise DURING the estimation
• An error in the channel transfer characteristics can
hurt you
– in the decoding
– in the water-filling
• Trade-off: Throughput vs. Estimation accuracy
• What if interference (as noise) is not white????

104
 Interference
• Generalization of other/ same cell interference for
SISO case
• Example: cellular deployment of MIMO systems
• Interference level depends on
– frequency/ code re-use scheme
– cell size
– uplink/ downlink perspective
– deployment geometry
– propagation conditions
– antenna types
105
 Summary and conclusions
• MIMO systems are a promising technique for high
data rates
• Their efficiency depends on the channel between the
transmitters and the receivers (power and correlation)

• Practical issues need to be resolved

• Open research questions need to be answered

106