Diversity techniques

for
flat fading channels

BER vs. SNR in a flat fading channel

Different kinds of diversity techniques
Selection diversity performance
Maximum Ratio Combining performance
 BER vs. SNR in a flat fading channel

Proakis, 3rd Ed. 14-3

In a flat fading channel (or narrowband system), the CIR

(channel impulse response) reduces to a single impulse
scaled by a time-varying complex coefficient.
The received (equivalent lowpass) signal is of the form

r t a t e s t n t
j t

We assume that the phase changes slowly and can be

perfectly tracked
=> important for coherent detection
 BER vs. SNR (cont.)

We assume:
the time-variant complex channel coefficient changes
slowly (=> constant during a symbol interval)
the channel coefficient magnitude (= attenuation
factor) a is a Rayleigh distributed random variable
coherent detection of a binary PSK signal (assuming
ideal phase synchronization)

Let us define instantaneous SNR and average SNR:

a 2 Eb N 0 0 E a 2 Eb N 0
 BER vs. SNR (cont.)

Since
2a a2 Ea2
p a e a 0,
E a
2

using
p a Rayleigh distribution
p
d da
Exponential distribution
we get
1
p e 0
0.
0
 BER vs. SNR (cont.)
The average bit error probability is
Important formula
for obtaining
Pe Pe p d statistical average

where the bit error probability for a certain value of a is

Pe Q 2a 2 Eb N 0 Q 2 . 2-PSK

We thus get

1

0
Pe Q 2 1
e 0
d 1
2 1 0
.
0 0
 BER vs. SNR (cont.)

Approximation for large values of average SNR is obtained

in the following way. First, we write

1 0 1 1
Pe 1 1 1
2 1 0 2 1 0
Then, we use
1 x 1 x 2

which leads to
Pe 1 4 0 for large 0 .
 BER vs. SNR (cont.)

Frequency-selective channel
BER (equalization or Rake receiver)
( Pe )
Frequency-selective channel
(no equalization)
BER floor
AWGN
channel Flat fading channel
(no fading)

SNR ( 0)
Pe 1 4 0 means a straight line in log/log scale
 BER vs. SNR, summary

Modulation Pe Pe Pe ( for large 0 )

2-PSK Q 2 1

2
1
0
1 0



1 4 0

DPSK e 2 1 2 0 2 1 2 0

2-FSK
(coh.) Q
1

2
1
0
20



1 2 0

2-FSK
(non-c.) e 2
2 1 0 2 1 0
 Better performance through diversity

Diversity the receiver is provided with multiple copies

of the transmitted signal. The multiple signal copies
should experience uncorrelated fading in the channel.
In this case the probability that all signal copies fade
simultaneously is reduced dramatically with respect to
the probability that a single copy experiences a fade.
As a rough rule:

1 Diversity of
Pe is proportional to
0L L:th order

BER Average SNR

 Different kinds of diversity methods

Space diversity:
Several receiving antennas spaced sufficiently far apart
(spatial separation should be sufficently large to reduce
correlation between diversity branches, e.g. > 10l).

Time diversity:
Transmission of same signal sequence at different times
(time separation should be larger than the coherence
time of the channel).

Frequency diversity:
Transmission of same signal at different frequencies
(frequency separation should be larger than the
coherence bandwidth of the channel).
 Diversity methods (cont.)

Polarization diversity:
Only two diversity branches are available. Not widely
used.

Multipath diversity:
Signal replicas received at different delays
(RAKE receiver in CDMA)
Signal replicas received via different angles of
arrival (directional antennas at the receiver)
Equalization in a TDM/TDMA system provides
similar performance as multipath diversity.
Selection diversity vs. signal combining

Selection diversity: Signal with best quality is selected.

Equal Gain Combining (EGC)

Signal copies are combined coherently:
L L
Z EGC ai e e ai
j i j i

i 1 i 1

Maximum Ratio Combining (MRC, best SNR is achieved)

Signal copies are weighted and combined coherently:
L L
Z MRC ai e ai
j i j i 2
ai e
i 1 i 1
 Selection diversity performance

We assume:
(a) uncorrelated fading in diversity branches
(b) fading in i:th branch is Rayleigh distributed
(c) => SNR is exponentially distributed:
1
p i e i 0
, i 0. PDF
0
Probability that SNR in branch i is less than threshold y :
y

P i y p i d i 1 e y 0 . CDF
0
 Selection diversity (cont.)

Probability that SNR in every branch (i.e. all L branches)

is less than threshold y :
L

y

P 1 , 2 , ... , L y p i d i 1 e y 0 L
.
0

Note: this is true only if the fading in different branches is

independent (and thus uncorrelated) and we can write

p 1 , 2 , , L p 1 p 2 p L .
 Selection diversity (cont.)

Differentiating the cdf (cumulative distribution function)

with respect to y gives the pdf
L 1 e y 0
p y L 1 e y 0
0
which can be inserted into the expression for average bit
error probability

Pe Pe y p y dy .
0

The mathematics is unfortunately quite tedious ...

 Selection diversity (cont.)

but as a general rule, for large 0 it can be shown that

1
Pe is proportional to
0 L

regardless of modulation scheme (2-PSK, DPSK, 2-FSK).

The largest diversity gain is obtained when moving from L

= 1 to L = 2. The relative increase in diversity gain
becomes smaller and smaller when L is further increased.
This behaviour is typical for all diversity techniques.
 BER vs. SNR (diversity effect)

For a quantitative picture (related

BER to Maximum Ratio Combining),
( Pe ) see Proakis, 3rd Ed., Fig. 14-4-2

Flat fading channel,

Rayleigh fading,
AWGN L=1
channel
(no fading)

SNR ( 0)
L=4 L=3 L=2
 MRC performance

Rayleigh fading => SNR in i:th diversity branch is

i
Eb 2 Eb 2
N0
ai
N0

xi yi
2

Gaussian distributed
Rayleigh distributed magnitude quadrature components

In case of L uncorrelated branches with same fading

statistics, the MRC output SNR is

Eb
N0

a1 a2
2 2
aL
2

N0

Eb 2
x1 y1
2
xL y L
2 2

 MRC performance (cont.)

The pdf of follows the chi-square distribution with 2L

degrees of freedom
Reduces to exponential pdf when L = 1

L1 L1
p e o
e o

0 L L 0 L L 1!
Gamma function Factorial

For 2-PSK, the average BER is Pe Pe p d

0

1
L L 1
L 1 k 1
k
Pe Q 2
Pe
2

k 0




k 2 0 1 0
 MRC performance (cont.)

For large values of average SNR this expression can be

approximated by
L
1 2 L 1 Proakis, 3rd Ed.
Pe 14-4-1
4
0 L

which again is according to the general rule

1
Pe is proportional to .
0 L
 MRC performance (cont.)

The second term in the BER expression does not increase

dramatically with L:

2 L 1 2 L 1 !
L L! L 1 ! 1 L 1

3 L2
10 L3
35 L4
 BER vs. SNR for MRC, summary
L
1 2 L 1 Proakis 3rd Ed.
For large 0 Pe
k
0 L 14-4-1

Modulation Pe Pe ( for large 0 )

2-PSK Q 2 k 4
DPSK k 2
2-FSK
(coh.)
Q k 2

2-FSK k 1
(non-c.)
 Why is MRC optimum peformance?

Let us investigate the performance of a signal combining

method in general using arbitrary weighting coefficients gi .
Signal magnitude and noise energy/bit at the output of the
combining circuit:
L L
Z gi ai Nt N 0 gi
2

i 1 i 1

SNR after combining:

Z Eb Eb gi ai
2
2

N 0 gi
2
Nt
Why is MRC optimum peformance? (cont.)

Applying the Schwarz inequality

gi ai gi ai
2 2 2

it can be easily shown that in case of equality we must

have gi ai which in fact is the definition of MRC.

Thus for MRC the following important rule applies (the

rule also applies to SIR = Signal-to-Interference Ratio):

L
i Output SNR or SIR = sum of
branch SNR or SIR values
i 1
 Matched filter = "full-scale" MRC

Let us consider a single symbol in a narrowband system

(without ISI). If the sampled symbol waveform before
matched filtering consists of L+1 samples
rk , k 0,1, 2, ,L
the impulse response of the matched filter also consists
of L+1 samples
hk rL k
*
Definition of matched filter

and the output from the matched filter is MRC !

L L L
Z hk rL k r
rL k rk
2
Lk
k 0 k 0 k 0
 Matched filter = MRC (cont.)

The discrete-time (sampled) matched filter can be

presented as a transversal FIR filter:
rL r0
T T T

h0 rL* h1 h L 1 hL

Z

L
Z rk
2
=> MRC including all L+1
values of rk k 0