HIAST-Ayman Alsawah Lecture on Multiple-Antenna Techniques in Advanced Mobile Systems v10

Multiple-Antenna Techniques

Communication Systems Master PROGRAM

Advanced Mobile Communication

2013-2014

Ayman [email protected]

Lecture 03-v10

Higher Institute of Applied Sciences & Technology

HIAST – Advanced Mobile Communication Ayman Alsawah, 2013/2014 2

Multiple-Antenna (MIMO) in the “Big Picture”

Narrow-Band FDMA-TDMA (200 KHz), GMSK

Multiple time-slots/user, packet-switching

8PSK, Adaptive Modulation & Coding

WCDMA (5 MHz), QPSK/BPSK, Freq. full-reuse, fast power control

Carrier Aggregation (to 40 MHz), 16/64 QAM, HARQ, MIMO

OFDM, 4x4 MIMO, All-IP

Uo to 5 x 20 MHz, 8x4 MIMO

Why Multiple Antennas?Exploit spatial dimension to: Enhance received S(I)NR (Spatial diversity/Diversity Gain or Beamforming/Array Gain) Enhance bit rate (Spatial Multiplexing/Multiplexing Gain)


WiFi APIEEE 802.11n (2007)IEEE 802.11ac (2012)

Multiple-Antenna Configurations


S = SingleM = Multiple

I = InputO = Output

SU-MIMO versus MU-MIMO


Single-UserMIMO

Multi-UserMIMO

LTE Rel’8 /DL (2008)Wifi 802.11ac /DL (2013)

Fading problem (Flat fading)

Fading of Rx power causes: - degradation in BER if the Bit Rate is fixed- limitation in Bit Rate if the BER is fixed

Time

Average Rx pwr

Min required pwr(Rx sensitivity)

Rx

Pow

er (

dB

m) Fading margin

Deep fade (target BER violation)


(dB)

BPSKUncodedFlat Rayleigh fading

Coherence time ≥ Tb

Coherent detection

(AWGN)

Erro

r p

rob

abili

tyExample of performance over flat fading

Solutions?


1-Time diversity via Coding & Interleaving

code word bit Error burst

Block fading model(Approximation)

A B C D E F G H INon-Interleaved:

A D G B E H C F IInterleaved:

Interleaving depth

tR

x P

wr Tc

Deep fade

Tc’

After deinterleaving, isolated errors have “more chance” to be corrected

Example: GSM

• Coded speech packet interleaved over 8 bursts• 1 user-assigned burst every frame of ~5 ms Packet interleaved

on 40 ms• @900 MHz, 120 km/h:

fd = 100 Hz

Tc = 10 ms


A B C D E F G H IDeinterleaved:

2- Frequency diversity via Freq. Hopping

Example: GSM

• Slow-FH ~200 hop/s

(Optional feature)

• Frame ≈ 4.6 ms(8 user bursts)• Typical Urban:

τRMS ≈ 1 µs

Bc = 1/(5τRMS)

= 200 KHzTime (ms)

Pow

er g

ain

(d

B)

900 MHz, 3 Km/h, Rayleigh flat fading

Frequencies much be spaced by more than the coherence

bandwidth Bc


2- Frequency diversity via OFDM

|H(f)|


3- Spatial diversity via Multiple Antennas

Time

Pow

er g

ain

(d

B)

Ant 1

Ant 2

Tx Rx

• For uniform surrounding scatterers:uncorrelated power gainsif antenna spacing = λ/2

• In practice: spacing λ

Example: GSM900

• 2 Rx antenna @BTS

• λ = 30 cm• Separation = 2-3 m


4- Polarization diversity

+45°Rx

-45°Rx

Methods for exploiting “Rx Diversity” intime, frequency, space, polarization, …?

DECT Handset(1900 MHz)


H

V

4- Polarization diversity in GSM


DuplexingFilter

+45° -45°

Main Diversity

To CombinerFrom Tx PA

Rx Diversity: mathematical model

s = transmitted symbol (M-QAM in general) with normalized average power E[|s|2] = 1 & symbol period Ts = 1

L = number of diversity brancheshk = baseband complex gain on antenna k, invariant during 1 symbol{hk} are iid zero-mean complex circular Gaussian random processes

|hk| is Rayleigh distributed with E[|hk|2] = 1 (normalized power gain)

|hk|2 (power gain) is exponentially distributed

{nk} are iid zero-mean complex circular WGN processes with E[|nk|2]=N0

rk = hk . s + nk , k = 1,…,L.

Flat fading complex baseband model (Narrow-band or per-OFDM-subcarrier)


Rx Diversity: SNR & CSI

rk = hk . s + nk , k = 1,…,L.

γk = |hk|2 . E[|s|2]/E[|nk|

2] =|hk|2/N0Instantaneous SNR:

Average SNR: γk, ave = E[|hk|2]/N0 = 1/N0 γk, AWGN

Instantaneous Ebno: (Eb /N0)k =|hk|2/(N0 log2 M),

Average Ebno:

M = Modulation order

(Eb /N0)k, ave = 1/(N0 log2 M) (per branch)

Received signal:

CSI Rx = subset of {{mag(h1), …, mag(hL)}, {arg(h1), …, arg(hL)}}

CSI Tx = none!

Channel State Information:


Rx Diversity: 1-Antenna Selection

Select the highest power gain branch (max |hk|

2)

Suitable for non-coherent detection where fading phases are not needed

CSI Rx = {mag(hk)}Used on LTE UL with 2 antennas


Rx Diversity: 2-Antenna Switching

Switch to the max power gain antenna when the current one falls below a given threshold

threshold

Ant. Sw.

Better solutions?


CSI Rx = {mag(hk)}

Simplified hardware at the price of degraded error performance compared to “Antenna Selection”

Rx Diversity: 3-Equal-Gain Combining

z = r1 exp(-j arg(h1)) + r2 exp(-j arg(h2))

= (|h1| + |h2|) . s + [ n1 exp(-j arg(h1)) + n2 exp(-j arg(h2)) ]

rk = hk . s + nk , k = 1,2.

Instant. SNR (= Instant. EbNo for BPSK):

γEGC = (|h1|+|h2|)2 E[|s|2]/E[|n1 exp(-j arg(h1))+n2 exp(-j arg(h2))|2]= (|h1|+|h2|)2 / (2N0)

BPSK error proba.: Pe, EGC = E[Q((2γEGC)1/2)] (1/N0) -L

Expectation w.r.t. (h1, h2) joint pdf

CSI Rx = {arg(hk)}


EGC is optimum when branches’ SINR’s have similar values

Rx Diversity: 4-Max. Ratio Combining

z = r1 h1* + r2 h2

*

= (|h1|2 + |h2|2) . s + [ n1 h1* + n2 h2

* ]

rk = hk . s + nk , k = 1,2.

Instant. SNR (= Instant. EbNo for BPSK):

γMRC = (|h1|2+|h2|2)2 E[|s|2]/E[|n1 h1* + n2 h2

*|2]= (|h1|2+|h2|2)2 / [(|h1|2+|h2|2)N0] = (|h1|2+|h2|2) /N0

= |h1|2/N0 +|h2|2 /N0 (sum of branches’ SNRs)

BPSK error proba.:

CSI Rx = {hk}

Pe, MRC = E[Q((2γMRC)1/2)] (1/N0) -L


MRC is THE optimum combining method, equivalent to a spatial matched filter

Bit

err

or

pro

bab

ility

BPSK performance over Rayleigh Flat Fading

Other methods for exploiting multiple antennas?

Eb/N0 (dB)Average SNR per branch

Diversity Gain

Slope increase


Reminder: Antenna Radiation Pattern

Tx power = PT

Uniform radiation intensity = PT /4 (W/strad)

Isotropic Antenna Directive Antenna

Radiation intensity = R(, )

Tx power = PT

Gain(, ) = 10 log10 D(, ) is measured in dBi (dB relative to isotropic antenna)


Vertical Cut

Maximum gain = 2.15 dBi

Example: Half-wave dipole antenna

Horizontal Cut

How to synthesize more complex directive patterns?

Antenna pattern is Tx/Rx reciprocal


Array antennas

Array antennas allows to control the radiation pattern by suitably arrangingantenna elements and adjusting the amplitude and phase of the signalreceived from/fed to each element, we talk about “Beamforming”

WiFi AP.


Adaptive beamforming

Var

. Gai

ns

Var

. ph

ases

Applications?

RF beamforming

Tx case

Main beam steering Remote electrical tilting Interference reduction …


App. 1: Radar Beam Steering


App. 2: BS electrical down-tilting

Max Gain = 15 - 20 dBi


App. 3: Switched beam

CSI @ BS:DL: CSI Tx = direction of userUL: CSI Rx = direction of user

1

2

3

4

Direction of Arrival(DoA)

Active beam selection

Index

Codebook

Weights

Array Gain

= Average SNR enhancement due to radiation focusing in the direction of user, w.r.t.

average SNR of single antenna

Multi-beam is also possible

No diversity gain


App. 4: SIR maximization

Desired user

Interfering user

Directions of Arrivals

Beam synthesis

Weights

What if baseband complex CSI was available instead of DoA?

Spatial filteringor

Zero-Forcing Beamforming

CSI @ BS:DL: CSI Tx = direction of usersUL: CSI Rx = direction of users

Array Gain


Baseband model:

Baseband Beamforming – no interference

On UL:CSI Tx = noneCSI Rx = full

Without noise: Maximize signal power <=> EGC

With noise: Maximize SNR <=> MRC

Prove that:

Exercise:


User 1

User 2

Baseband model:

Exercise: • Find weights that yield: y = S1 (interference from user 2 is cancelled) under perfect CSI Rx and without noise (Interference Rejection Combining (IRC)). • What is the feasibility condition of this IRC?• In the noisy case, give the expression of both y and the SINR.


Baseband Beamforming – with interference


Space-Division Multiple Access (SDMA)

S1

S2


“Baseband Multi-beam”

Multiplexing Gain

UL data rate is doubled: 2 users transmit simultaneously


Array Gain & Diversity Gain• Array Gain = Average SNR / Single-Branch average SNR

• Diversity Gain = - log10(Δ Average error proba.) / log10(Δ Single-Branch average SNR)@ Hi SNR


Multiple Tx Antennas (MISO)

Closed-Loop Techniques: Channel-State Information (CSI) known @ Tx side through Feedback

Tx Diversity: Antenna selection/switching (Feedback = antenna index only)

Precoding (pre-weighting) for in-phase combining @ Rx antenna

Tx RF Beamforming: beam steering, beam switching, multi-beam.

STBC (Space-Time Block Coding)

SFBC (Space-Frequency Block Coding)

Open-Loop Techniques: No CSI @ Tx


STBC Example: 2x1 Alamouti

CSI Tx = noneCSI Rx = full

S. M. Alamouti, “A simple transmit diversity technique for wireless communications,” IEEE J. Sel. Areas Comm., vol. 16, pp. 1451–1458, Oct. 1998.

Space-Time Precoding matrix:

No array gainDiversity gain = 2 (full)No multiplexing gain (full)


Used in WiFi 802.11n (2008)

2x1 Alamouti Precoding & Decoding

Tx Side:

Rx Side:

(power constraint ignored here)Decoding:

Exercise: Find instantaneous & average decoded SNR

H (orthogonal matrix)

(inverse ~ transpose & conjugate)


SFBC Example: 2x1 Alamouti

STBC & SFBC can be generalized to more than 2 Tx antennasSee also Alamouti 2x2

Used in UMTS & LTE


In OFDM-MIMO systems like LTE, time slots can be replaced with subcarriers when implementing Alamouti

MIMO Flat Fading Baseband Model

Pre

cod

er

Dec

od

er

GenerateCSI Tx

GeneratePrecoder

Data in out

T Antennas R Antennas

hRT

n1

n2

nR

x1

x2

xT

y1

y2

yR

hR1

h1T

h11


MIMO Signal Model

Rayleigh iid model:

AWGN Noise vector: n

Channel matrix:

(R x T)

nxHy Received vector:

RX1 RXT TX1 RX1

(full rank)


MIMO Spatial MultiplexingSpatial Multiplexing is a Closed-Loop (Full CSI Tx & Rx) MIMOtechnique for increasing data rate (i.e. obtaining a multiplexing gain)

where:

RXT RXR RXT TXT

• U & V are unitary (orthogonal) square matrices (i.e. UUH = IR, VVH = IT)• is a diagonal matrix whose main diagonal is formed of min{T, R} strictly positive real values:

R>T R=T R<T


- Channel matrix is considered deterministic (known to the receiver) and can be decomposed using SVD = Singular-Value Decomposition:

MIMO: simple & great idea!

Received vector: nxy H VU

Let’s send instead of (data pre-processing):xx V~ x

nxnxy H UV VU~

Now multiply the received vector by (post-processing):HU

nxnxy HHH ~UUU~Uy

y~

has the same distribution as since is unitary

nx ~y

nn HU~ n U

min{T, R} parallel Gaussian channels


• Case T<R: T data symbols are sent on parallelT received values + (R-T) zeros to be ignored

• Case T≥R: R data symbols + (T-R) dummy zeros are sent on parallelR received values

In all cases: data rate is increased by min{T, R} “Multiplexing Gain”(For the same error performance than SISO and without extra spectrum)

V

n1

n2

nR

UHH=UVHx

xx V~ yH ~Uy

y

x1

x2

xT

~

~

~

y1

y2

yR

~

~

~

T R

Feedback = V

LTE & WiMax (up to 8x4 on DL)WiFi .11n (4 streams)WiFi .11ac (8 streams)


42

Engineering

HIAST-Ayman Alsawah Lecture on Multiple-Antenna Techniques in Advanced Mobile Systems v10