TSKS14

Multiple Antenna
Communications
Lecture 3, 2020

Emil Björnson
TSKS14 Multiple Antenna Communications 2020-04-09 2

Outline of this lecture

• Multipath propagation and Rayleigh fading
• Slow fading
• Outage probability
• Outage capacity
• Spatial diversity
• Fast fading
• Ergodic capacity
• Channel hardening
TSKS14 Multiple Antenna Communications 2020-04-09 3

Multipath propagation

Non-line-of-sight
channel:
• Scattering

• Channel with 𝐿 propagation paths:

(
/0 +/
+,-.
𝑔 = $ 𝛼% 𝑒 1
Reference distance
%&'

Channel gain Wavelength

TSKS14 Multiple Antenna Communications 2020-04-09 4

Multipath fading 1

• Example: 0.8
Approximation:
' 𝑔~𝐶𝑁 0,1
• 𝛼% = (
0.6
/ +/
• 𝜃% = 2𝜋 0 ~𝑈(0,2𝜋)
1 Pr{ 𝑔 < 𝑥}
0.4
• Channel magnitude:
( 0.2
1 +,<
|𝑔| = $ 𝑒 0
𝐿
%&' 0 𝑥
0 0.5 1 1.5 2 2.5 3
TSKS14 Multiple Antenna Communications 2020-04-09 5

Rich scattering: Rayleigh fading

Central limit theorem
Let 𝑋' , … , 𝑋( be a sequence of 𝐿 real-valued independent and identically
distributed random variables with zero mean and variance 𝜎 - . As 𝐿 → ∞,
(
1
$ 𝑋%
𝐿𝜎 - %&'
converges to a standard Gaussian distribution 𝑁 0,1 .

• Rich multipath propagation

• Very large number paths: Gaussian distribution
• Channel gain: 𝑔 ∼ 𝐶𝑁(0, 𝛽)
• Called Rayleigh fading since |𝑔| ∼ Rayleigh( 𝛽/2)
TSKS14 Multiple Antenna Communications 2020-04-09 6

Rayleigh fading, |𝑔| ∼ Rayleigh(1/ 2)

1
• Channel gain changes
over time
0.8

0.6

0.4

0.2
• In this case: 𝑔~𝐶𝑁(0,1)

0
0 1 2 3 4
 TSKS14 Multiple Antenna Communications 2020-04-09 7

Rayleigh fading, zooming in on tail

• Risk of very small
channel gain

Two issues:
• Variations in channel quality
• Unpredictable
TSKS14 Multiple Antenna Communications 2020-04-09 8

Capacity of fading channel

• AWGN channel with a random channel response 𝑔[𝑙]:
𝑦 𝑙 = 𝑔[𝑙] ⋅ 𝑥 𝑙 + 𝑛[𝑙]
• 𝑥 𝑙 ∼ 𝐶𝑁(0, 𝑞), energy per sample: 𝑞 = 𝑃/𝐵
• 𝑛 𝑙 ∼ 𝐶𝑁(0, 𝑁`)

• Two categories:
• Slow fading: 𝑔 𝑙 takes one realization during communication
• Fast fading: 𝑔 𝑙 takes “all” realizations during communication

Reality might be somewhere in between

TSKS14 Multiple Antenna Communications 2020-04-09 9

Slow fading
• Received signal
𝑦 𝑙 = 𝑔 ⋅ 𝑥 𝑙 + 𝑛[𝑙]

• Fixed channel 𝑔 𝑙 = 𝑔 for the entire transmission

• Assumption: Receiver knows 𝑔, but not the transmitter

𝑞/𝑁`
• Capacity for a realization 𝑔:
𝐶a = log -(1 + 𝑔 -SNR)

Transmitter does not know 𝐶a

Cannot encode the data to achieve it!
TSKS14 Multiple Antenna Communications 2020-04-09 10

Opportunistic transmission
• Suppose transmitter encode using the rate 𝑅 bit/s/Hz

• Two possible events:

• If 𝑅 ≤ 𝐶a : Successful transmission
• If 𝑅 > 𝐶a : Large error probability

System is outage if 𝑅 > 𝐶a

• Outage probability for rate 𝑅

𝑝ijk 𝑅 = Pr 𝐶a < 𝑅 = Pr log - 1 + 𝑔 -SNR < 𝑅
TSKS14 Multiple Antenna Communications 2020-04-09 11

Outage probability with 𝑔~𝐶𝑁 0,1 High SNR

no pq 2v − 1
+
• Outage probability for rate 𝑅: 𝑝ijk 𝑅 = Pr 𝐶a < 𝑅 = 1 − 𝑒 rst ≈
SNR

Outage probability
decays with
SNR = 𝑞/𝑁`
as SNR+'
TSKS14 Multiple Antenna Communications 2020-04-09 12

Outage capacity
• Difference from AWGN channel
• Only 𝑅 = 0 can guarantee zero error probability
• Capacity is zero

• 𝜖-Outage capacity Cy :
• Largest rate 𝑅 such that 𝑝ijk 𝑅 ≤ 𝜖

Interpretation:
With probability 1 − 𝜖, we can communicate at Cy with is zero error probability
TSKS14 Multiple Antenna Communications 2020-04-09 13

Outage capacity with 𝑔~𝐶𝑁 0,1

Cy = log - 1 + SNR ln 1 − 𝜖 +'

Difference from
AWGN channel

• Low 𝜖: Better with AWGN channel

• High 𝜖: Better with fading channel

SNR is 0 dB
TSKS14 Multiple Antenna Communications 2020-04-09 14

Outage capacity with small outage probability

• Fraction of AWGN capacity:
log - 1 + SNR ln 1 − 𝜖 +'
log - 1 + SNR

• Much lower capacity than

with AWGN channel

Can we improve the situation?

TSKS14 Multiple Antenna Communications 2020-04-09 15

Fading multiple antenna channels

• Independent and identically distributed Rayleigh fading

• Channel gain: 𝒈 ∼ 𝐶𝑁(𝟎, 𝛽𝑰~ ) Independent:
p
ƒ
Uniform linear array
€ •pq ‚ „
• Distribution of 𝒈 -: 𝑓 𝒈 n 𝑥 = ~+' !†•
with Δ = 𝜆/2
TSKS14 Multiple Antenna Communications 2020-04-09 16

𝑀 receive antennas and i.i.d. Rayleigh fading

High SNR
• Outage probability
-o +' ~
- SNR
‹Œ• 2v − 1 1
𝑝ijk 𝑅 = Pr log - 1 + 𝒈 <𝑅 =Š 𝑓 𝒈 n 𝑥 𝑑𝑥 ≈
SNR 𝑀!
`

Spatial diversity gain

𝑝ijk 𝑅 proportional to SNR+~
𝑀 is the diversity order
TSKS14 Multiple Antenna Communications 2020-04-09 17

Outage probability with 𝑀 receive antennas

• Outage probability decays as
SNR+~

• Makes a huge difference!

Multiple receive antennas gives:

• Beamforming gain
• Diversity gain
TSKS14 Multiple Antenna Communications 2020-04-09 18

Fast fading
• Received signal
𝑦 𝑙 = 𝑔[𝑙] ⋅ 𝑥 𝑙 + 𝑛[𝑙]

• Block fading
• One realization of channel 𝑔 𝑙 per 𝑙 (or a finite-sized block of symbols)
• New independent realization every time (ergodic process)
TSKS14 Multiple Antenna Communications 2020-04-09 19

Opportunistic transmission
• Suppose transmitter encode using the rate 𝑅 bit/s/Hz
• There are L fading realization: 𝑔 1 , … , 𝑔 𝐿

• Reliable communication if
(
$ log - 1 + SNR 𝑔 𝑙 - > 𝐿𝑅
•&'
• Many fading realizations:
(
1 -
𝑅 < $ log - 1 + SNR 𝑔 𝑙 → 𝔼 log -(1 + 𝑔 -SNR)
𝐿
•&'
Mean value with respect
As 𝐿 → ∞ to channel fading
TSKS14 Multiple Antenna Communications 2020-04-09 20

Ergodic capacity
• This is called ergodic capacity:
𝔼 log -(1 + 𝑔 -SNR)

• Deterministic: Transmitter knows it even if 𝑔 is unknown

• There are no outage issues!

• Extension to SIMO case with channel:

𝔼 log -(1 + 𝒈 -SNR)
 TSKS14 Multiple Antenna Communications 2020-04-09 21

Comparison with AWGN channel

• AWGN channel:
log -(1 + SNR)

• Rayleigh fading:
𝔼 log -(1 + 𝑔 -SNR)

Low SNR: Little difference

High SNR: Ergodic capacity is lower
TSKS14 Multiple Antenna Communications 2020-04-09 22

Jensen’s inequality and concave functions

• For any random variable 𝑧 and concave function 𝑓(⋅),
𝔼 𝑓(𝑧) ≤ 𝑓 𝔼 𝑧

𝑓 𝑧 = log - 1 + 𝑧
A function is concave if
• Any line between two
points on the curve
is below the curve 𝑓 𝑧 = −log - 1 + 1/𝑧
• Second derivative
is negative
TSKS14 Multiple Antenna Communications 2020-04-09 23

Ergodic capacity with SIMO channel

• Can be used to prove
SNR
log - 1+ ≤ 𝔼 log - 1 + 𝒈 - SNR ≤ log - 1 + 𝔼 𝒈 - SNR
𝔼 𝒈 +-

Jensen’s inequality with Jensen’s inequality with

𝑓 𝑧 = −log - 1 + 1/𝑧 𝑓 𝑧 = log - 1 + 𝑧

• Rayleigh fading with 𝒈 having i.i.d. 𝐶𝑁(0,1) elements:

log - 1 + (𝑀 − 1)SNR ≤ 𝔼 log - 1 + 𝒈 - SNR ≤ log - 1 + 𝑀 SNR

Non-fading channel with 𝒈 - =𝑀−1 Non-fading channel with 𝒈 - =𝑀

TSKS14 Multiple Antenna Communications 2020-04-09 24

Comparison
• Small 𝑀
• Large loss from
channel fading

• Larger 𝑀
• Small loss

Channel hardening:
When 𝑀 is large, no penalty
from channel fading
TSKS14 Multiple Antenna Communications 2020-04-09 25

Summary
• Slow fading: One realization during transmission
• Outage probability, outage capacity
• Reliable communication → Large performance loss
• Multiple antennas give more reliability

• Fast fading: Many realizations during transmission

• Ergodic capacity with averaging over fading
• No reliability issue, but performance loss
• Multiple antennas give similar capacity as with non-fading channels
 End of Lecture 3

TSKS14 Multiple Antenna

Communications