DIVERSITY FOR FADING CHANNELS

B. Sainath
[email protected]

Department of Electrical and Electronics Engineering

Birla Institute of Technology and Science, Pilani

April 1, 2019

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 1 / 27

OUTLINE

1 INTRODUCTION

2 COHERENT DETECTION

3 SEP1 ANALYSIS

4 AWGN VS FADING

5 DIVERSITY

6 COMBINING

7 REFERENCES

1 symbol error probability

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 2 / 27
Introduction

Communication via fading channel

poor performance due to high probability of deep fade

Q. How to increase probability of decoding received symbol correctly at

Rx ?

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 3 / 27

Introduction

Communication via fading channel

poor performance due to high probability of deep fade

Q. How to increase probability of decoding received symbol correctly at

Rx ?
Ans. Diversity
Provide independently fading paths across time/frequency/space

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 3 / 27

COHERENT DETECTION:
BPSK, AWGN Channel

Modulation: M-ary phase shift keying (MPSK)

Model: BPSK Tx, AWGN channel, ML Rx

y [m] = x[m] + w[m].

√ √
x[m] : + Es (bit ‘1’) , x[m] : − Es (bit ‘0’)
w[m] ∼ N(0, N20 )
Received SNR: signal energy-to-noise PSD
γ=

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 4 / 27

COHERENT DETECTION:
BPSK, AWGN Channel

Modulation: M-ary phase shift keying (MPSK)

Model: BPSK Tx, AWGN channel, ML Rx

y [m] = x[m] + w[m].

√ √
x[m] : + Es (bit ‘1’) , x[m] : − Es (bit ‘0’)
w[m] ∼ N(0, N20 )
Received SNR: signal energy-to-noise PSD
Es
γ= N0

Symbol(or bit) error probability:

√
Q( 2γ) ≤ exp (−γ)
log2 M
Rate = T , where T is symbol period
e.g., BPSK: 1 bit/symbol period

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 4 / 27

COHERENT DETECTION:
BPSK, Rayleigh Fading plus AWGN

Model: BPSK Tx, frequency-flat Rayleigh fading channel, ML Rx

Assumption: perfect channel state information (CSI) is available at Rx

y [m] = h[m]x[m] + w[m], y , h, w ∈ C

√ √
x[m] : + Es (bit ‘1’), x[m] : − Es (bit ‘0’) ⇐ symbols are equally-likely
h[m] ∼ CN (0, 1)
Q: pdf of |h|2 ?

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 5 / 27

COHERENT DETECTION:
BPSK, Rayleigh Fading plus AWGN

Model: BPSK Tx, frequency-flat Rayleigh fading channel, ML Rx

Assumption: perfect channel state information (CSI) is available at Rx

y [m] = h[m]x[m] + w[m], y , h, w ∈ C

√ √
x[m] : + Es (bit ‘1’), x[m] : − Es (bit ‘0’) ⇐ symbols are equally-likely
h[m] ∼ CN (0, 1)
Q: pdf of |h|2 ? (Ans. |h|2 ∼ exp(1))
w[m] ∼ CN (0, σ 2 ) ⇐ AWGN
For AWGN, σ 2 = N0
Received SNR (instantaneous):
Es |h|2
Γ= N0

Q. Average received SNR? (in class)

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 5 / 27

COHERENT DETECTION: ML RULE

Q: Consider BPSK symbols ±a (i.e., x1 = a, x2 = −a). Assuming

equiprobable symbols, derive ML rule

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 6 / 27

COHERENT DETECTION: ML RULE

Q: Consider BPSK symbols ±a (i.e., x1 = a, x2 = −a). Assuming

equiprobable symbols, derive ML rule
Conditional pdfs

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 6 / 27

COHERENT DETECTION: ML RULE

Q: Consider BPSK symbols ±a (i.e., x1 = a, x2 = −a). Assuming

equiprobable symbols, derive ML rule
Conditional pdfs
!
1 |y − ah|2
p(y |x1 , h) = 2
exp −
πσ σ2
!
1 |y + ah|2
p(y |x2 , h) = 2
exp −
πσ σ2

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 6 / 27

ML RULE Contd.,

ML rule:

p(y|x1 , h) ≷xx12 p(y|x2 , h) ⇒ ln p(y |x1 , h) ≷xx12 ln p(y |x2 , h)

After simplification, we get (verify)

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 7 / 27

ML RULE Contd.,

ML rule:

p(y|x1 , h) ≷xx12 p(y|x2 , h) ⇒ ln p(y |x1 , h) ≷xx12 ln p(y |x2 , h)

After simplification, we get (verify)

R{h? y } ≷xx12 0
? 2
For x1 = a: ỹ = R{ h|h|y } = a |h| + w̃R , where w̃R ∼ N(0, σ2 )
Symbol error probability (SEP) event

SEP|(h, x1 = a) = P(ỹ < 0|h, x1 = a) = P(w̃R < −a |h|)

q  √ 
2a2 |h|2
Inst. SEP = Q σ2
= Q 2Γ
Γ denotes inst. fading SNR received
2
For a2 = Es , σ 2 = N0 , we get Γ = EsN|h|
0

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 7 / 27

ML DETECTION & SEP

Q: Derive fading-averaged SEP & verify it with Craig’s formula (in class)
q 
2
Inst. Symbol(or bit) error probability: Q 2 |h| γ , where γ = NEs0

Fading-averaged SEP
 q   r 
2 1 γ 1
SEP = E Q 2 |h| γ = 1− ≈
2 1+γ 4γ
√

For AWGN, average SEP = Q 2γ ≤ exp (−γ)
Compare with fading plus AWGN channel performance

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 8 / 27

AWGN VS FADING

Q. Why poor SEP performance with fading?

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 9 / 27

Typical Error Event

q 
2
Inst. SEP: SEP|h = Q 2 |h| γ
2
|h| γ >> 1: very small probability of error (why?)
2
|h| γ < 1: significant error probability (deep fade event)
Probability of deep fade event (DFE), pdfe
2
pdfe = P(|h| < γ1 ) ≈ 1
γ

(in class)
typical error event is due to channel being in deep fade

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 10 / 27

Summary So Far

Model: BPSK Tx, AWGN channel, ML Rx (Coherent)

Average SEP:
ASEP ∝ e−SNR

Wireless fading channel

Model: BPSK Tx, frequency-flat Rayleigh fading plus AWGN, ML Rx
Fading-Averaged SEP:
1
ASEP ∝ P[deep fade] ∝ SNR

Fading Mitigating Techniques: Diversity & combining

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 11 / 27

Diversity & Combining
Basic idea: transmit same data over independent fading paths
Combine independent paths to reduce fading of resultant signal
Microdiversity to mitigate effect of multipath fading, eg: Time, Frequency,
Space/antenna, Polarization
Macrodiversity to mitigate effects of shadow fading
Time diversity obtained by interleaving and coding over symbols across
different Tc

Figure: consecutive symbols are transmitted sufficiently far apart in time, so that
channel taps (h` ) are statistically independent.

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 12 / 27

Remarks on Interleaving

Interleaving: process to rearrange code symbols so as to spread bursts

of errors over multiple code-words that can be corrected by error
correcting codes
Useful technique for improving performance of error correcting codes by
combating bursts of errors

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 13 / 27

Repetition coding: SEP Performance

Error probability:
q 
2k h k2 γ , where k h k2 = L`=1 h`2 ← Chi-square distribution
P
Q
2
Let X = k h k , pdf of X

1
pX (x) = x L−1 exp (−x) , x ≥ 0
(L − 1)!
For small x, pX (x) = 1
(L−1)!
x L−1 , x ≥0
Average SEP as function of received SNR

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 14 / 27

Repetition coding: SEP Performance

Error probability:
q 
2k h k2 γ , where k h k2 = L`=1 h`2 ← Chi-square distribution
P
Q
2
Let X = k h k , pdf of X

1
pX (x) = x L−1 exp (−x) , x ≥ 0
(L − 1)!
For small x, pX (x) = 1
(L−1)!
x L−1 , x ≥0
Average SEP as function of received SNR
Z ∞ √ 
ASEP = Q 2xSNR pX (x) dx
0

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 14 / 27

Repetition coding: SEP curves

Average SEP decreases rapidly as L increases

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 15 / 27

Repetition coding: DFE Probability

2
Let X = k h k , pdf of X

1
pX (x) = x L−1 exp (−x) , x ≥ 0
(L − 1)!
For small x, pX (x) = 1
(L−1)!
x L−1 , x ≥0
Q. What is probability of DFE? (In class)
‘L’ is diversity gain of the system
More sophisticated codes can achieve coding gain (e.g., Rotation code)
Important: tradeoff between throughput vs. diversity

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 16 / 27

Time Diversity: Rotation Code

Rotate square constellation by angle θ

Consider L = 2 scenario
ASEP upper bound: corresponds to
optimum θ = 0.5 tan−1 (2) =

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 17 / 27

Time Diversity: Rotation Code

Rotate square constellation by angle θ

Consider L = 2 scenario
ASEP upper bound: corresponds to
optimum θ = 0.5 tan−1 (2) = 31.7 degrees
maximize the minimum Euclidean distance
15
ASEP ≤
SNR2
Coding gain over repetition code = 3.5 dB
Read Ch. 3. Section 3.2.2. Tse & Viswanath
B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 17 / 27
GSM Uses Time Diversity

Coded bits are interleaved across 8 consecutive time slots

Maximum possible time-diversity: 8
Actual gain depends on mobile speed
Exercise: Tc = 5 ms, fc = 900 MHz, Ds = ? & mobile speed = ?
Express Tc in terms of Ds

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 18 / 27

Frequency Diversity

Wideband channels: transmission BW > coherence BW (of channel)

Sending symbols more frequently ⇒ ISI issue
How to mitigate ISI while exploiting channel’s diversity ⇐ challenge
Modulate the transmitting signal through ‘L’ different carriers
Approaches: single-carrier systems with equalization, DSSS, Multi-carrier
systems
System Example
Single-carrier system GSM
Direct-sequence spread-spectrum IEEE 802.11b, IS-95
Multi-carrier system IEEE 802.11a2

2 https://en.wikipedia.org/wiki/IEEE_802.11

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 19 / 27

Direct-Sequence Spread-Spectrum
Principle: spread data via pseudo-random noise(PN) sequence across
bandwidth much larger than data rate
The signal received along ‘L’ nearly orthogonal is maximal-ratio combined
using RAKE Rx

Average probability of error when anti-podal modulation (XA = −XB = u)

used
h √ i
E Q 2Γ
kuk2
PL 2
l=0 |hl |
Γ= N0

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 20 / 27

DSSS: Asymptotic Average SEP

Channel gains are i.i.d. & hl ∼ CN (0, 1L )

2
Eb , k u k is average total energy received per bit of information
SNR per branch ?

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 21 / 27

DSSS: Asymptotic Average SEP

Channel gains are i.i.d. & hl ∼ CN (0, 1L )

2
Eb , k u k is average total energy received per bit of information
SNR per branch ?
 
1 Eb 1
L N0
, factor L
accounts for splitting of energy due to spreading
q 
2Eb
As L → ∞, average SEP → Q N0 ⇐ performance of AWGN

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 21 / 27

Space Diversity

Antenna diversity: receive diversity, transmit diversity, & transmit and

receive diversity

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 22 / 27

Combining

Combining techniques to achieve array gain

Eg. Linear combining at coherent Rx (assuming AWGN channel and
BPSK transmitter model)
Combining ’M’ branches each with SNR of γ = NEs
0
Total SNR = M × γ in each branch (details & proof discussed in class)
Array gain is M
What is the symbol/bit error probability?
Combining techniques for fading channels (to be discussed in class)
Maximal ratio combining
Selection combining

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 23 / 27

MRC: AWGN Scenario (No fading)

AWGN w ∼ CN(0, N0 )
L− branch Linear combining: Received SNR (derivation in class)

LEs
ΓE =
N0

L− fold increase in SNR due to coherent combining

SNR increase in the absence of fading ⇒ array gain
Q:Write down expressions for outage, average SEP, spectral efficiency

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 24 / 27

MRC: Fading Scenario

Consider L− branch MRC, frequency-flat Rayleigh fading

Principle Maximally combine signal by choosing optimal weights ⇒
maximize SNR
Received signal:

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 25 / 27

MRC: Fading Scenario

Consider L− branch MRC, frequency-flat Rayleigh fading

Principle Maximally combine signal by choosing optimal weights ⇒
maximize SNR
Received signal: yML = w1 y1 + . . . + wL yL
Received instantaneous fading SNR
2 2
Es |h1 | Es |hL |
ΓE = + ... +
N0 N0
Q: What is the average SNR?
Diversity order indicates
how the slope of the average symbol error probability (SEP) as a function of
average SNR changes with diversity
Exercise: Derive expressions for
Outage probability
Fading-averaged SEP (Hint: Use MGF approach) & diversity order
Fading-averaged spectral efficiency

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 25 / 27

Selection Combining

Combiner outputs the signal on the branch with the highest SNR
Average SNR of the combiner output in i.i.d. Rayleigh fading is
L
X 1
ΓE = Γ
n
n=1

Exercise: Derive expressions for

Outage probability
Fading-averaged SEP (Hint: Use MGF approach) & diversity order
Fading-averaged spectral efficiency
Compare MRC vs SC performance

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 26 / 27

References

Fundamentals of wireless communication by Tse & Viswanath

Wireless communication by Andrea Goldsmith
www.wirelesscommunication.nl/reference/about.htm

B. Sainath (BITS, PILANI) MOBILE TELECOM: Diversity April 1, 2019 27 / 27