ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

Revision of Lecture Nine

• AWGN channel: decision variable rk = r̄k + nk = g0sk + nk , where channel g0 is known
N e = Es
– AWGN nk with PSD 20 , average symbol energy Es, average channel SNR Λ N0
• Bit error ratio performance
“q ”of BPSK
“pand QPSK
” “q ” “p ”
– BPSK Pe = Q 2Es
=Q e , QPSK Pe = Q
2Λ Es =Q Λe
N0 N0
• Bit error ratio performance of 16QAM
– 4-ary modulation: two classes of bits, and Pe,2 ≈ 2Pe,1
C1 bit decision boundary C2 bit decision boundaries
2d

11 10 00 01 11 10 00 01
“q ” “q ” “q ” „q «
Es 9Es 25Es e
– 16QAM Pe = 43 Q 5N0 + 12 Q 5N0 − 14 Q 5N0 ≈ 43 Q Λ
5

• Bit error ratio performance of 64QAM

– 8-ary modulation: three“classes of bits,”and Pe,3“≈ 2Pe,2 and P”e,2 ≈ 2Pe,1
“ p ”
7
p 1
p 1
– 64QAM Pe = 12 Q Es /21N0 + 2 Q 3 Es/21N0 − 12 Q 5 Es/21N0 +
“ p ” “ p ” „q «
1 1 7 e
Λ
12 Q 9 E s /21N 0 − 12 Q 13 E s /21N 0 ≈ 12 Q 21

• This lecture we consider fading channel performance

128
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

Flat Rayleigh Fading Channels

• A narrowband channel is represented by c(t) = α(t) · ejφ(t). Assume that fading is

sufficiently slow, c(t) and φ(t) are symbol invariant → during one symbol period

c = α · ejφ

– We consider uncorrelated Rayleigh fading, i.e. Doppler spread→ ∞

• The Rayleigh fading envelope α has a probability density function

 2

α α
pα(α) = 2 exp − 2 , α≥0
α0 2α0
and the channel phase φ is uniformly distributed in [−π, π] with PDF
 1
pφ(φ) = 2π , −π ≤ φ ≤ π
0, otherwise

2
• Note that Re[c]
pπ and Im[c] are i.i.d. Gaussian with variance α 0 , and α has mean
E[α] = ᾱ = 2 α0, 2nd moment E[α2] = 2α02 and variance 4−π 2 α0
2

129
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

Flat Fading Performance

• Given the transmitted baseband signal m(t) with average energy Es, the received signal is
jφ
r(t) = α · e · m(t) + n(t)
N0
where n(t) is AWGN with PSD 2
• Define the instantaneous channel SNR
2 Es
λ=α
N0
Note that λ > 0 is a chi-square distribution with PDF

1 −λ/Λ
pλ(λ) = e
Λ
• The average channel SNR Λ is defined as

2 Es
Λ = E[λ] = λ̄ = 2α0
N0

• Let P (λ) be the instantaneous error probability, the average error probability is then defined as
Z ∞ Z ∞
1
Pe = P (λ)pλ (λ)dλ = P (λ)e−λ/Λ dλ
0 Λ 0

130
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

4QAM Flat Fading Performance

• Fading does not change the decision boundaries I, Q = 0, but I, Q = d becomes I, Q = αd
p
• Using non-fading result Pe = Q(d/ N0/2) → the instantaneous error probability
!
αd “√ ”
Pe (λ) = Q p =Q λ
N0/2

• Note λ ≥ 0 and Λ = E[λ] = 2α20 Es/N0 , the average error probability is

Z ∞ “√ ”
1 −λ/Λ
Pe = Q λ e dλ
Λ 0

• Note that λ = α2Es /N0 and Es = 2d2, the close-form solution for Pe is:
Z ! Z ! !
∞ ∞ 2
αd αd α α
Pe = Q p pα(α)dα = Q p exp − dα
0 N0/2 0 N0/2 α20 2α20
0 v 1
u 2
0 s 1
u 2α 2 d
1BB u 0 N0 C
C = 1 @1 − Λ
= @1 − t A
2 d 2 A 2 2+Λ
1 + 2α20 N0

131
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

4QAM Flat Fading Performance (Derivation)

• Note !
Z ∞ “√ ” 2 1 β
integration formula 2Q 2βx e−µx x dx = 1− p
0 2µ µ + β2
Z ! !
∞ 2
α αd α
average error probability Pe = Q p − 2 dα exp
0 N0/2 2α0 α20
Z ∞ ! ! „ «
√ dα0 α α2 α α
= 2Q 2p √ exp − 2 √ d √
0 N0/2 2α0 2α 0 2α0 2α0
dα
• Letting µ = 1, β = √ 0 and x = √α in the above integration formula leads to
N0 /2 2α0

0 1
v 0 1
B C u 0 s 1
√d α0 u d 2
1B C 2
2α0 N
B N0 /2 C 1BB u 0
C
C 1 Λ
Pe = B1 − s = @1 − t = @ 1− A
2B „ «2 C
C 2 2
1 + 2α0 N d 2A
2 2+Λ
@ √d α0 A 0
1+
N0 /2

4α2
0d
2
where Λ = 2α20 N
Es
, 2
Es = 2d ⇒ Λ = N0
0

132
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

4QAM Fading / Non-Fading BER Comparison

Es
• In fading, average SNR Λ = 2α02 N 1
0 Rayleigh fading
AWGN
– Average BER 0.1

r ! 0.01
1 Λ
Pe = 1− 0.001

4QAM BER
2 2+Λ
0.0001

• Compare with AWGN, average SNR 1e-05

e = Es/N0
is defined as Λ
1e-06
– Average error probability
1e-07
p 
e 1e-08
Pe = Q Λ 0 5 10 15 20 25 30
channel SNR (dB)
• Clearly, fading is “a big killer” limiting communication system’s performance
– Many mobile communication technologies developed in past three decades are
“counter-fading measures”

133
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

16QAM Flat Fading Performance: C1 BER

• Motivate by success of applying non-fading 4QAM to fading 4QAM: try the same to 16QAM?
• For C1 bits, the decision boundaries are not changed:

I, Q > 0 → i1, q1 = 0, I, Q ≤ 0 → i1 , q1 = 1

But ±d, ±3d become ±αd, ±3αd. Also Es = 10d2 and Λ = 20α20 d2/N0
• Thus, the instantaneous C1 error probability is
„ « „ « r ! r !
1 αd 1 3αd 1 λ 1 λ
Pe,1 (λ) = Q + Q = Q + Q 3
2 σn 2 σn 2 5 2 5

• The average C1 error probability is therefore (success again!)

0 v 1 0 v 1
Z ∞ u u
B u 2α2 d2 C B u 18α2 d2 C
1 −λ/Λ 1 u 0 N0
C + 1 B1 − u 0 N0
Pe,1 = Pe,1 (λ)e dλ = B @ 1 − t 2 A @ t 2
C
A
Λ 0 4 1 + 2α20 N d 4 1 + 18α20 Nd
0 0

0 s 1 0 s 1
1 Λ 1 9Λ
= @1 − A + @1 − A
4 10 + Λ 4 10 + 9Λ

134
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

16QAM Fading Performance: C2 BER

• For C2 bits, the decision boundaries are changed to:

I, Q > 2ᾱd or I, Q ≤ −2ᾱd → i2 , q2 = 1

−2ᾱd < I, Q ≤ 2ᾱd → i2 , q2 = 0

Note that the average value of α, i.e. ᾱ, has to be used for decision threshold
• Two cases of i2 , q2 = 0 error need consideration
1. i2 , q2 = 0 error in case of α < 2ᾱ α< 2 α : PDF 2αd
– Instantaneous symbols −αd and -2αd
αd are within region defined by
two decision boundaries I,Q
-3αd -αd αd 3αd
– Error occurs when noise makes 11 10 00 01
received signal outside the region i2,q2 =0
and instantaneous C2 bit = 0
error probability for α < 2ᾱ is d1 d2

„ q « „ q « ! !
(2ᾱ − α)d (2ᾱ + α)d
P2,0,<2 (α) = Q d1 / N0/2 +Q d2 / N0/2 = Q p +Q p
N0/2 N0/2

135
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

16QAM: C2 BER (continue)

2. i2 , q2 = 0 error in the case of α > 2ᾱ: Instantaneous symbols −αd and αd are outside region
defined by two decision boundaries
– Correct decision occurs only α> 2 α: -2αd PDF
2αd
when noise moves received signal
inside the region
– Thus instantaneous C2 bit = 0 I,Q
-3αd -αd αd 3αd
error probability for α > 2ᾱ is
11 10 00 01
i2,q2 =0
d1
P2,0,>2(α) = d2
„ q « „ q « ! !
(−2ᾱ + α)d (2ᾱ + α)d
1 − Q d1/ N0/2 + Q d2/ N0/2 = 1 − Q p +Q p
N0/2 N0/2

• Note α > 0, the average error for i2 , q2 = 0 is therefore

Z 2ᾱ Z ∞
P2,0 = P2,0,<2 (α)pα (α)dα + P2,0,>2(α)pα (α)dα
0 2ᾱ

– There exists closed form solution for this integration but it is very complicated
– So far not too bad

136
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

16QAM: C2 BER (continue)

• Two cases of i2 , q2 = 1 error need consideration

1. i2 , q2 = 1 error in case of α > 2ᾱ/3

– Instantaneous symbols −3αd α> 2 α /3 : -2αd PDF 2αd
and 3αd are within correct
regions corresponding to
respective decision boundaries I,Q
-3αd -αd αd 3αd
– Error occurs when noise makes i2,q2 =1 11 10 00 01 i2,q2 =1
received signal outside the d2
d1
corresponding region

– Thus instantaneous C2 bit = 1 error probability for α > 2ᾱ/3 is

q « „ „ q «
P2,1,>2/3 (α) = Q d1/ N0/2 − Q d2/ N0/2
! !
(3α − 2ᾱ)d (3α + 2ᾱ)d
=Q p −Q p
N0/2 N0/2
– Oh, not, nightmare now

137
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

16QAM: C2 BER (continue)

2. i2 , q2 = 1 error in case of α < 2ᾱ/3
α< 2 α /3 :
-2αd 2αd PDF
– Instantaneous symbols −3αd
and 3αd are outside correct
regions defined by respective
I,Q
decision boundaries -3αd -αd αd 3αd
– Correct decision occurs only i2,q2 =1 11 10 00 01 i2,q2 =1
when noise moves received d2
signal to the correct region d1
– The instantaneous C2 bit = 1 error probability for α < 2ᾱ/3 is
„ q « „ q «
P2,1,<2/3(α) = 1 − Q d1/ N0/2 − Q d2/ N0/2

! !
(2ᾱ − 3α)d (2ᾱ + 3α)d
=1−Q
p −Q p
N0/2 N0/2
• The average error for i2, q2 = 1 is therefore
Z 2ᾱ/3 Z ∞
P2,1 = P2,1,<2/3(α)pα (α)dα + P2,1,>2/3 (α)pα (α)dα
0 2ᾱ/3

– Really nightmare, too much for me

138
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

16QAM Fading /Non-Fading BER Comparison

• OK, just to continue, we known 1
– average C2 error probability is
0.1
1
Pe,2 = (P2,0 + P2,1)
2 0.01

– Therefore average error probability

16QAM BER
0.001
for 16QAM is
0.0001
1
Pe = (Pe,1 + Pe,2 )
2 1e-05 fading C2
fading C1
• Although no closed-form solution for 1e-06 AWGN C2
AWGN C1
Pe,2 and Pe , we can still make some
sense out of them 1e-07
– 16QAM fading / non-fading BER:
1e-08
– Fading degrades BER performance 0 5 10 15 20 25 30
seriously ⇒ counter fading measures channel SNR (dB)

• Alternatively, Monte Carlo simulation is often used to evaluate fading BER

– Recall slide 48 for flat Rayleigh fading channel simulation

139
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

Fading Channel Simulation Considerations

• For uncorrelated fading , Re{c} and Im{c} are i.i.d. Gaussians with zero mean and unit variance
– Channel is simulated by generating such a sequence of {ck }
• For correlated fading, {ck } can be generated by flat Rayleigh fading channel simulator of slide 48
• For AWGN channel BER simulation, at least a few hundred of error counts should be obtained
– For fading channel BER simulation, this is insufficient
simulation time
magnitude of channel tap

simulation time
– If you simulate at pink simulation zone, channel is in a deep fade, you get too bad BER
– If you simulate at green simulation zone, channel gain is high, you get too good BER
• Basically, all magnitude possibilities of channel tap c should be simulated or seen, such a sequence
{ck } can be very very long, and simulation is really time-consuming
– Roughly tens of “cycles” in channel fading envelope should be simulated to get accurate average
fading BER → for slow fading channel, this is very very long

140
ELEC6214 Advanced Wireless Communications Networks and Systems S Chen

Summary

• Narrowband Rayleigh fading channel:

– Fading envelope and phase PDFs, instantaneous and average channel SNRs,
instantaneous and average error probabilities

• Fading channel BER performance analysis

– For 4QAM, by applying non-fading BER result, closed-form BER solution of
fading channel is obtained
– Apply same approach to 16QAM is less fruitful, but nevertheless useful insight
can be obtained
– Fading degrades BER performance seriously ⇒ counter fading measures are
necessary

• Practical considerations in Monte Carlo simulation of fading channel performance

– Basically, all magnitude possibilities of channel tap c should be simulated or seen
– This corresponds to roughly tens of “cycles” in channel fading envelope

141