Diversity and MIMO System for

Fading Channels
Dr. Waleed Salos

1
Realization of channel

2
3
4
5
 Overview

• Diversity and its types

• MIMO systems

6
 Why Diversity?
• BER in fading channel is calculated by
averaging the AWGN BER over the fading pdf :
 
Pb   Pb  g  p  g  d g and Ps   Ps  g  p  g  d g
0 0

• In Rayleigh fading channel the pdf of the SNR, g

is given by the exponential function
exp   g g 
p g  
g
• This pdf causes large performance loss
• If we get a better pdf of the SNR, performance
will increase
• This is done by diversity techniques
7
 What is Diversity ?
• Diversity is achieved by creating several
independent paths between the transmitter and
receiver
• Each path fades independently, hence, there is
a low chance they fade together
• Receiver combines the received signal for the
several paths using some method
• Diversity is used in all mobile communication
systems

8
 Types of Diversity
• Space diversity: multiple transmit and multiple
receive antennas
– Multiple Tx: split power over several Tx antennas.
More antennas = more power split
– Multiple Rx: collect signal by several Rx antennas.
More antennas = more collected power
– Antennas separation about l/2 is required
– If directional antennas (typically) larger separation is
required

9
10
 Types of Diversity
• Polarization diversity:
– Transmit and/or receive with both vertical and
horizontal polarization
– Scattering is independent for each polarization, giving
independent paths
– Limited to 2 transmit and 2 receive diversity
– Tx polarization diversity: half power for each
polarization

11
 Types of Diversity
• Frequency diversity:
– Transmit same signal with several frequencies
– Frequencies separated by > coherence bandwidth
– Also wideband signals achieve frequency diversity,
like OFDM techniques over wideband (WLAN,
WiMax, LTE)
• Multipath diversity:
– In Direct-Sequence-Spread-Spectrum signals we can
receive from multipath separately using Rake receiver
– Used in all CDMA systems (IS-95, CDMA2000,
WCDMA)

12
 Types of Diversity
• Time diversity:
– Signal is re-transmitted (repeated) after > coherence
time
– Also achieved using coding and interleaving
– Reduces overall transmission data rates
– Coding and interleaving used in all mobile
communication systems
– Also combined with repeat-diversity in what is called
Hybrid Automatic Repeat Request (H-ARQ)

13
 Types of Diversity
• Beam-forming:
– Transmit with antenna array
– Each antenna is fed with different phase
– Forms a directional beam towards the receiver, or
group of receivers
– Antenna beam tracks the intended receiver
– Requires knowledge of the fading channel at Tx
– Optional is 4G mobile communication systems
(WiMax and LTE)

14
 Types of Diversity
• In all types of diversity we achieve several
independent paths between Tx and Rx
• If not independent there is some performance
degradation
• If correlation coefficient < 0.5 degradation is not
noticeable
• Space diversity will be studied, but approach
applies to all methods

15
 Diversity Order
• In many cases the average probability of error
looks like Ps   g  M

– M is the diversity order

• If the number of paths between Tx and Rx is N,
then diversity order  N
• For example: if Nt is the number of Tx antennas
and Nr is the number of Rx antennas
– Number of branches N= Nt Nr
– Diversity order M  N
• Full diversity order if M = number of branches

16
 Receiver Diversity
• Signal from multiple receiver antennas are
linearly combined:
– Weighted sum of each branch
 j
– Each branch weight is complex  i  ai e i
– The phase of each received path are aligned
– Signals from each branch are coherently combined

• Receiver must estimate the

phase (and some times gain) of
each branch
• Estimation through pilot signals

17
 Receiver Diversity
• Receiver diversity achieves two types of gain:
Array gain and Diversity gain
• Array gain: due to coherent combining of multiple
branches
– Defined: average combined SNR /average branch SNR
– Achieved even if there is no fading
• Diversity gain: due to the better (than exponential)
pdf p(g) of the combined SNR
– Better BER results of the integration
 
Pb   Pb  g  p  g  d g and Ps   Ps  g  p  g  d g
0 0

18
 Receiver Diversity
• Several methods to combine the receiver
branches:
– Selection combining
– Threshold combining
– Maximal ratio combining
– Equal gain combining
– Mix of the above
• Tradeoff between performance and complexity

19
 Selection Combining
• For M branch diversity, the CDF of the combined SNR is
given by
– Probability of best SNR < g = Probability all SNR < g
• In Rayleigh fading each SNR is exponential
exp   g i g i 
p g i  
gi
• CDF of M branches is given by
M M g
exp   g i g i   M
Pg  g    Prob  g i  g         1  e 
g gi
d gi
i 1 i 1  0 gi  i 1
• If the target SNR is go, outage is given by
M
Outage  Pg  g o    1  e g o gi

i 1

• If the average SNR per branch are equal:

gi  g 
Outage  Pg  g o   1  e g o g

M

20
 Selection Combining
• Probability of error is calculated by averaging the AWGN
formula over the PDF of the SNR
 
Pb   Pb  g  p  g  d g and Ps   Ps  g  p  g  d g
0 0
• Now we use
M
1  e 
M 1
p g   g g
eg g

g
• For BPSK there is no closed form, needs numerical
integration (or simulation)
• For DPSK we get

 1
m

1 g M M 1
 M  1
Pb   e p g   g  d g 
2 2
 1  m  g 
m

0 m 0  

21
 Selection Combining
• From basic probability, the PDF is the differential of CDF
• If the average SNR per branch are equal the CDF is
given by M
Pg  g    1  e g g i
  1  e 
g g M

i 1
• The PDF of the combined SNR is given by differentiating
w.r.t g M
p g  g  
g
1e 
 g g M 1  g g
e

• The average SNR after combining is

 
gM
 
M 1
g    g pg  g  d g   1  e g g
e g g
dg
0

0
g
M
g 1 i
i 1

• Increases with M, but not linear

22
 Selection Combining
• Most benefit from M=1 (no diversity) to M=2
• Benefit reduces as M increases

10 log10  g g o 

23
 Selection Combining
• Fading channel path gain to each branch is
ri exp  j  
• Noise PSD at each branch is Ni
• Receiver selects the branch with largest
instantaneous SNR ri 2 N i
• If the noise PSD No is the same for all branches,
this is equivalent to selecting largest ri 2  N o
– Easier to implement
• Combiner output SNR = best branch SNR

24
 Selection Combining
• Similarly, more gain
when moving from
M=1 to M=2
• Gains reduces as M
increases

25
 Selection Combining
• Same
observation for
D-BPSK
(equation 7.11)
• Gains reduces
as M increases

26
 Threshold Combining
• A simpler approach is to stay with a branch till its
SNR falls below a threshold
• If the SNR falls below threshold, switch to
another branch according to:
– Random selection, or
– Best branch
• Example: 2 branches

Bad switch

27
 Threshold Combining
• Assume M=2 and switching happens when the SNR on
one antenna falls below gT (even if to lower SNR)
• The CDF of selected SNR is the probability that SNR < g
• Two possibilities for overall SNR, SNR < gT , or > gT
• If SNR < gT
 Prob(SNR<g)=Prob(SNR1<gT) x Prob(SNR2<g)
• If SNR > gT
 Prob(SNR<g)=Prob(gT <SNR1<g)
+ Prob(SNR1<gT) x Prob(SNR2<g)
• For Rayleigh fading and equal average SNR:

 1  e  gT g
 1  e g g
 g  gT
Pg   g   
e T   1  e  1  e 
g g  gT g
 e g g g g
g  gT
28
 Threshold Combining
• Hence, the CDF is given by:
1  e gT g  e g g  e g gT  g g  gT
Pg   g   
e  T 
g g  g g g
 1  2e g  gT
• The outage probability is given by
1  e gT g  e g o g  e g o gT  g g o  gT
Pout  g 0   Pg   g 0    g g  g
g g
 1  2e o  e o T g o  gT

• If the threshold gT = go, the outage probability is similar to

selection diversity with M=2
Pout  g 0   1  2e g o g
 e 2g o g

 
2
 1  e g o g

29
 Threshold Combining
• The PDF is found by differentiating the CDF
1
 g 1  e gT g
e g g
g  gT
Pg   g   
 1  2  e gT g
e g g
g  gT
 g
• Again, for BPSK there is no closed form for
BER. Requires numerical integration or
simulation
• For DPSK the BER is given by

1 g 1
Pb   e pg   g  d g 
2 2 1  g 
1  e gT g
 e  g T e  gT g

0

30
 Threshold Combining
• M=2
• DBPSK with
threshold
combining
• Slightly worse
than selection
combining

31
 Maximal Ratio Combining
• Instead of selecting one branch, all branches are added
with weight ai exp(-ji)
• If a transmitted symbol is s with unity power |s|2=1
• The received symbol at branch i is: s ri exp(ji) + ni
• The combined symbol is:
M M
S comb  s  ri ai   n i ai
i 1 i 1
    
Signal Part Noise Part

• Assume all ni have equal variance No, the SNR is:

2
M 
  ri ai 
g    i 1 M 
N o  ai2
i 1
• We need to select ai to maximize SNR
32
 Maximal Ratio Combining
• The outage probability is found from PDF
go
Pout  g o   Prob  g   g o    pg g  d g

0

g o g 
k 1
M
 1e g o g

k 1  k  1 !

• The average BER is calculated by averaging the AWGN

BER over the random SNR
• For BPSK: 
P  Q  2g  p  g  d g
b 
0
g

M M 1 m
 1    M 1  m   1   

 2 
   m   2 
m 0  

 g 1  g 

33
 Maximal Ratio Combining
• BER performance
for BPSK
• Best diversity
method

34
 Maximal Ratio Combining
• Using Swartz inequality we find the optimum weights
ai  ri N o
• The resulting SNR is
M M
1
g 
No
r
i 1
i
2
 gi
i 1

• Combined SNR is the sum of all branches SNR

• Each gi is exponential with mean g i
• Assuming all branch SNR have equal mean, g is chi-
square with 2M degrees of freedom (see Proakis,
chapter 2) and mean g  M g

g M 1 e g g

• The PDF is given by pg   g  

g M  M  1 !
35
 Comparison

BPSK with Maximal Ratio BPSK with Selection

Combining Combining

Maximal Ratio Combining provides better performance (lower BER)

36
 Equal Gain Combining
• MRC required knowledge of the SNR on each branch
• Simpler approach is equal weight for all branches (all ai=1)
2
• The combined SNR is g  1  r  M

 i
M N o  i 1 

• There is no closed form solution for the CDF or PDF

except for M=2
CDF is Pg   g   1  e  2g g 
 g g
g
e g
1  2Q  2g g 
 1 1 g 
PDF is p g   g  
1
g
e  2g g
 e g g
 
 4g g g g 
 1  2Q   2g g 

37
 Equal Gain Combining
• From the CDF we find the outage probability
Pout  g o   1  e  2g o g

 g o g
g
e o g
1  2Q  2g o g 
• Also the BER for BPSK is

Pb   Q
0
 
2g p g   g  d g

  1  
2
1
 1 1  
2  1 g  
 

38
 Equal Gain Combining
• BER for BPSK

BPSK with Maximal Ratio BPSK with Equal Gain

Combining Combining

39
40
 MIMO
• Traditional diversity is based on multiple receiver
antennas
• Multiple-In Multiple-Out (MIMO) is based on both
transmit and receive diversity
• Also known as Space Time Coding (STC)
• With Mt transmission antennas and Mr receiver antennas
we have Mt Mr branches
• Tx and Rx processing is performed over space
(antennas) and time (successive symbols)

41
 MIMO or STC
• In Mobile communication systems it may be difficult to
put many antennas in the mobile unit
• Diversity in the downlink (from base station to mobile
station) can be achieved by Multiple-In Single-Out
(MISO) (i.e., Mr=1)
• In the uplink (from mobile station to base station)
diversity is achieved my conventional diversity (SIMO)
• Hence, all diversity cost is moved to the base station
• All 3G and 4G mobile communication system employ
MIMO in their standard

42
 Type of MIMO
• Two major types of space time coding
– Space time block coding (STBC)
– Space time trellis coding (STTC)
• STBC is simpler by STTC can provide better
performance
• STBC is used in mobile communications. STTC
is not used in any systems yet
• We will talk only about STBC

43
 Space Time Block Codes
• There are few major types
– Transmit diversity: main goal is diversity gain
– Spatial multiplexing: main goal is increase data rate
– Eigen steering: main goal is both. Requires
knowledge of the channel at the transmitter side
– Mix of the above: Lots of research
• Transmit diversity, spatial multiplexing and
simplified version of Eigen steering are used in
3G and 4G standards
• While in 3G standards MIMO was an
enhancement, in 4G MIMO is a main part
44
 Transmit Diversity
• Take Mt=2 and Mr=1
• Two symbols so and s1 are transmitted over two
transmission periods
• No change in data rate (denoted as rate 1 STBC)
• Channel is known at receiver only

45
 Transmit Diversity
• Transmission matrix: Ant
oAnt1

 s o s1   Timeo
S  * *

 1s s o   Time1

• Transmission matrix columns are orthogonal to

guarantee simple linear processing at the receiver
• Other transmission matrices are defined in literature
• Received signal is:
 ro   s o s1   g o   no 
R    * *    
 r1   s1 s o   g 1   n1 
• Performance is same as MRC with M=2
• However, if Tx Power is the same, then transmit diversity
(2x1) is 3 dB worse than (1x2)
46
 Transmit Diversity
• Take Mt=2 and Mr=2
• Performance is the same as MRC with M=4
• However, if Tx Power is the same, then transmit diversity
(2x2) is 3 dB worse than (1x4)

47
 Performance
• MRRC=Maximal Ratio Receiver Combining
• Note 3 dB difference in favor of Rx MRC diversity
Reference: S. Alamouti, a simple transmit diversity
technique for wireless communications,
IEEE JSAC, October 98

No diversity

Order 2

Order
4

48
 Spatial Multiplexing
ro  s o g o  s 1 g 1
r1  s o g 2  s 1 g 3

• Purpose is to increase data rate (2x2 gives twice data rate)

• The 4 gains must be known at receiver
• Simplest way zero forcing algorithm:

 ro   g o g 1  s o  sˆo  1 H  o
r
 r   g  sˆ   G G  G  r 
H
g  s 
 1   2
3  1   1  1
G

49
 Spatial Multiplexing
ro  s o g o  s 1 g 1
r1  s o g 2  s 1 g 3

• Optimum method: Maximum Likelihood

– Try all combinations of s1 and s2
– Find the combination that minimizes the squared error:
2 2
e  e  ro  sˆo g o  sˆ1 g 1  r1  sˆo g 2  sˆ1 g 3
2
o
2
1

– Complexity increases with high order modulation

50
 Performance
• Equal rate comparison
• Reference: David
Gesbert, Mansoor Shafi,
Da-shan Shiu, Peter J. Zero forcing
Smith, and Ayman
Naguib, From theory to
practice: an overview of
MIMO space–time coded ML
wireless systems, IEEE
JSAC, April 2003
Alamouti

51
 Eigenvalue Steering
• Assume a MIMO system

52
 Eigenvalue Steering
• Example with Mt = 2 and Mr=4

 y 1   h11 h12   n1   y   H x   n 
 y  h  x n 
 2    21 h 22   1   2
  
 y 3   h31 h32   x 2   n 3 
     
 y 4  
 h41 h42 
  n4 
H
• Any matrix H can be represented using
Singular Value Decomposition as
H U V H
• U is Mr by Mr and V is Mt by Mt unitary
matrices
•  is Mr by Mt diagonal matrix, elements σi
53
 Eigenvalue Steering
• Using transmit pre-coding and receiver shaping

y  U H  H x  n 
 U H U V H
x n
 U H U V H V x  n 
 U H U V H V x U H n
  x  n

54
 Eigenvalue Steering
• This way we created r paths between the Tx and specific
Rx without any cross interference
• The channel (i.e., Channel State Information) must be
known to both transmitter and receiver
• The value of r = rank of matrix H, r min(Mt, Mr)
• Not all r paths have good SNR
• Data rate can increase by factor r
• See Appendix C for Singular Value Decomposition
• See Matlab function [U,S,V] = svd(X)

55
 SVD Matlab Example

X is 4 by 2 matrix V=
-0.6414 0.7672
X=[1 2 ; 3 4 ; 5 6 ; 7 8] -0.7672 -0.6414
[U,S,V] = svd(X)
S=
U= 14.2691 0
-0.1525 -0.8226 -0.3945 -0.3800 0 0.6268
-0.3499 -0.4214 0.2428 0.8007 0 0
-0.5474 -0.0201 0.6979 -0.4614 0 0
-0.7448 0.3812 -0.5462 0.0407
Then σ1 = 14.2691 and σ2 = 0.6268

56