Cooperative Comm v3-1

49
Indian Institute of Science (IISc), Bangalore, India Cooperative Communications Neelesh B. Mehta ECE Department IISc, Bangalore Collaborators : Andreas Molisch (MERL), Ritesh Madan (Flarion), Raymond Yim (Olin College), Hongyuan Zhang (Marvell), Natasha Devroye (Harvard), Jin Zhang (MERL), Jonathan Yedidia (MERL), Vinod Sharma (IISc), Gaurav Bansal (IISc)

Transcript of Cooperative Comm v3-1

Page 1: Cooperative Comm v3-1

Indian Institute of Science (IISc), Bangalore, India

Cooperative Communications

Neelesh B. MehtaECE DepartmentIISc, Bangalore

Collaborators:

Andreas Molisch (MERL), Ritesh Madan (Flarion), Raymond Yim (Olin College),

Hongyuan Zhang (Marvell), Natasha Devroye (Harvard), Jin Zhang (MERL),

Jonathan Yedidia (MERL), Vinod Sharma (IISc), Gaurav Bansal (IISc)

Page 2: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Motivation Behind Cooperative Communications

• Multiple antenna spatial diversity

using only single antenna nodes

• Exploit two fundamental aspects

of wireless channels:

– Broadcast

– Multiple access

s

r1

dr2

r3

r4

Cooperative relaysd

s2

Tw

o co operativ e sourc es

s1

h1d

h2d

h12

h1d

h4d

h2d

h3dhsd

Page 3: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

What’s Different Between MIMO and Cooperation?

• Distributed nature of relays/nodes

– Different channel gain amplitudes and phases

– Each relay runs on its own timer and VCO

• Relay capabilities

– Single antenna

– Full duplex or half duplex

• Channel state information (CSI)

– Relay might not know states of other relay links

Page 4: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Outline

• Various cooperation schemes

• Cooperation in ad hoc networks

• Cooperation in infrastructure-based networks

• Cross-layer issues

• Other interesting topics

Page 5: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Cooperative Communication Schemes

• Amplify and forward

• Decode and forward

• Estimate and forward

Possibilities:

• Orthogonal / Non-orthogonal cooperation

• Coded / Uncoded cooperation

Page 6: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Analysis of Basic 3 Node Scenario

Performance metrics

• Outage

• Power consumption

• Diversity

• BER (Coded/Uncoded)

d

s2

Tw

o so urces

s1

h1d

h2d

h12

S1 transmits S2 transmits

d receives d receivesConventionalmodel

Tx

Rx

S1 tx S2 repeats S2 tx S1 repeats

d, S2 rx d rx d,S1 rx d rx Cooperative source model

Tx

Rx

[Laneman & Wornell, IEEE Trans. on Inf. Theory, 2004]

[Stefanov, Erkip, IEEE Trans. on Communications, 2004]

Page 7: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Outage Analysis: Amplify and Forward

[1][1]

[2] [2]

sd dd

d rd sr rd r d

h wyx

y h h h w wβ β

= + +

2

0

r

sr s

P

h P Nβ ≤

+

2 22

2 2

SNR SNRlog 1 SNR

SNR SNR

sr rd sr rdAF sd sd

sr sr rd sr

h hI h

h h

÷= + + ÷+

d

r

s hsd

hrd

hsr

xyd

yr = hsr x + wr

( ) ( ) 222 2

2 2 2 2

2 11( , ) Pr

2 SNRsr

sd sr

Rrd

out AFrd

P SNR R I Rσ σ

σ σ σ−+

= < ≈

Relay power

constraint:

Tx. rate

Outage prob.

Diversity order = 2

Page 8: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Outage Analysis: Decode and Forward

Case 1: Destination can decode only if relay decodes

ˆrx x= ˆd rd dy h x w= +

( ) ( )2 2 21min log 1 , log 12DF sr sd rdI SNR h SNR h SNR h = + + +

( )2

2

1 2 1( , ) Pr

R

out DFsr

P SNR R I RSNRσ

−= < ≈

(Assume codeword level decoding)

Diversity order = 1

Page 9: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Outage Analysis: Adaptive Decode and Forward

Case 2: Source forwards to destination instead of relay if SR channel is

poor

ˆrx x= ˆd rd dy h x w= +

( )( )

22 2

2 2

1 2 1log 1 2 ,2

1log 1 , else2

R

sd sr

DF

sd rd

SNR h hSNRI

SNR h SNR h

−+ <= + +

( ) ( ) 222 2

2 2 2 2

2 11( , ) Pr

2

R

sr rdout DF

sd sr rd

P SNR R I RSNR

σ σσ σ σ

−+= < ≈

(Similar results apply for non-orthogonal scheme in which source transmits

to destination in both time slots, and relay repeats in second time slot)

Page 10: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

DF Coded Cooperation: An Explicit Example

• Codeword of N bits divided into two parts: N1 and N2

• In next frame:

– S2 relays N2 bits of S1 if it can decode it correctly

– Else, S2 sends its own N2 bits

[Hunter & Nosratinia, IEEE Trans. on Wireless Commn., 2006]

S1 bits S2 bits relay Inactive

Inactive S2 bits S1bits relay

S1

S2 Rx S1 bits

Rx S2 bits

N1 bits N2 bits N1 bits N2 bits

Page 11: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Analysis: Pairwise Codeword Error Probability

• Slow fading

1 1 2 2

1 1 1( )

2 1 1d d

P dd SNR d SNR

= ÷ ÷+ +

( )1 1 2 2( ) 2 2d dP d Q d dγ γ= +

• Fast fading

1 2

1 2( ) 2 ( ) 2 ( )d dn n

P d Q n nη η

γ γ∈ ∈

= + ÷ ÷

∑ ∑1 2

1 1

1 1 1( )

2 1 1

d d

d d

P dSNR SNR

≤ ÷ ÷+ +

Diversity order = 2

Diversity order = Hamming distance

(Same for non-cooperation case)

SNR in first frame

SNR in second frame

Page 12: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Other Cooperation Schemes

• Estimate and forward

– [Cover & El Gamal, IEEE Trans. Inf. Theory, 1979]

• Non-orthogonal transmission schemes

– Perform better at the expense of a more complicated destination

receiver [Nabar, Bolczkei, Kneubuhler, IEEE JSAC 2004]

Page 13: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Cooperation in Ad Hoc Networks

• Basic 3 node scenario

• Multiple sources/relays case

Page 14: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Extension to Multiple Node Scenarios

Non-orthogonal schemesOpen-loop scenario • Each relay that decodes

chooses its column of a pre-specified ST code matrix

(e.g., Orthogonal ST design)[Chakrabarti, Erkip, Sabharwal, Aazhang, IEEE Sig. Proc. Mag., 2007]

• Relay subset selection

Closed-loop scenario

• Relays that decode beamform together to destination

2 Repeats 11 Tx 3 Repeats 1 …... N Repeats 1

1 Repeats 22 Tx 3 Repeats 2 …... N Repeats 2

1 Repeats 33 Tx 2 Repeats 3 …... N Repeats 3

1 Repeats NN Tx 3 Repeats N …... N-1 repeats N

time

freq

uen c

y

Orthogonal scheme

[Laneman & Wornell, IEEE Trans. on Inf. Theory, 2003]

1 Tx D(1) subset repeats

2 Tx D(2) subset repeats

N Tx D(N) subset repeats

time

freq

uen c

y

Non-orthogonal scheme

Page 15: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

C

22 2

2

22

Cooperative Beamforming and its Feasibility

• Relays phase align and power control transmit signal

• Equivalent to a multi-antenna array at transmitter

• Two important practical issues

– CSI needs to be acquired

– Beamforming nodes need to be synchronized

1

1

1

1

1

1

C

Inter-cluster communications

[Ochiai, Mitran, Poor & Tarokh, IEEE Trans. Sig. Proc. 2005]

Page 16: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Acquiring CSI in Cooperative Beamforming

s

r1

tr2

r3

r4

r5

xx

1. Broadcast data 2. Acquire CSI 3. Select relays

[Madan, Mehta, Molisch, Zhang, To appear in IEEE Trans. Wireless Commn., 2008]

• Acquiring CSI requires extra energy and time

s

r1

tr2

r3

r4

r5

Relay subset

selection by

destination

g1

g3

g2

h1

h2

h3

h5

s

r1

tr2

r3

r4

r5

4. Beamform data

α |g1|/(|g1|+|g3|)

α |g3|/(|g1|+|g3|)

x

Page 17: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Trade-offs and Design Goals

• Broadcast power:

– Less power: Signal reaches fewer relays, lose out on diversity

– More power: Signal reaches more relays, but increases relay

training overhead

• Relay selection by destination:

– Select few relays: Lose out on diversity when transmitting data

– Select many/all relays: More feed back energy spent to reach less

and less useful relays

• Questions:

– Optimum relay subset selection rule (subject to outage constraint)?

– Energy savings achieved by cooperative beamforming?

Page 18: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Average Energy Consumption: Including Cost of CSI

As a function of number of relays who decode message

Total energy consumed: Effect of relay selection rule

• Rule of thumb: Broadcast to reach 3-4 (best) relays, some of then beamform upon selection

Page 19: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Synchronization for Cooperative Beamforming

• Performance robust to imperfect synchronization

• Example: Two equal amplitude signals from two

transmitters. Signals are offset by a phase w

– Resulting amplitude: |1+ ejω| = 2 cos(ω/2)

– Even if ω = 300, amplitude = 1.93 (instead of 2) – Off by only 4% !

[Mudumbai, Barriac & Madhow, IEEE Trans. Wireless Commn. 2007]

• General case:

[ ] [ ]

22

1

2

1.

12. 2 ( 1) cos

i

Nj

R ii

R i

P g e

E P N EN

ω

ω

==

= + −

Page 20: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Receive Power Distribution

Phase uniformly distributed between [-π/10, π/10]

[Mudumbai, Barriac & Madhow, IEEE Trans. Wireless Commn. 2007]

Page 21: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Relay Selection: Relays Help Even When ‘Not Used’

• Full diversity achieved by just selecting single best relay

– Well understood classical result

• [Win & Winters, IEEE Trans. Commn. 1999]

• E.g., Antenna selection, Partial Rake CDMA receivers

– Simple to implement

Page 22: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Relay Selection: Selection Criteria and Mechanisms

s

r1

dr2

r3

r4

h1

h2

h3

h4

g1

g2

g3

g4

Selection criteria:

• Depends on SR and RD channels

• Criteria: ( )2 2

2 2

2 2

1. min ,

2.

i i i

i ii

i i

h g

h g

h g

µ

µ

=

=+

[Blestsas, Khisthi, Reed & Lippman, IEEE JSAC, 2006; Luo et al, VTC 2005;

Lin, Erkip & Stefanov, IEEE Trans. on Commn., 2006]

• Multiple access relay selection mechanism:

– Relays overhear a RTS (request to send) from source, and

CTS (clear to send) from destination to estimate channels

– Each relay sets a timer with expiry 1/i it µµ

Page 23: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Opportunistic Relay Selection and Cooperation Using Rateless Codes

• Rateless codes (e.g., digital fountain codes)– Convert a finite-length source word into an infinitely long

bitstream

– Receiver decodes successfully when received mutual information exceeds the entropy of the source word

– Receiver only needs to send a 1-bit ACK

• Ideal ‘binning’ properties of rateless codes1. Order in which bits received doesn’t matter

2. If destination receives data streams from N nodes, it accumulates mutual information from all N nodes

[Shokrollahi, ISIT 2004; Mitzenmacher, ITW 2004; Luby, FOCS 2002; Palanki & Yedidia, ISIT 2004;

Erez, Trott & Wornell, CoRR 2007]

Page 24: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Asynchronous Cooperation With Rateless Codes

s

r1

dr2

r3

r4

s

r1

dr2

r3

r4

s

r1

dr2

r3

r4

Broadcast Best relay receives packet and starts transmitting to

destination

Second best relay also receives packet and starts transmitting to destination

[Molisch, Mehta, Yedidia, Zhang, IEEE Trans. Wireless Commn,

2007]

Time taken for best relay to decode packet: ( )( )2log 1 max i i

Bt

h=

+

h1

h4

h2

h3

Page 25: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Performance: Transmission Energy & Time

Mean transmission time and energy usage

Energy usage statistics

Performance primarily depends on inter-relay link strength

Mea

n tx

.

ener

gy

Mean tx .

time

Number of

relays

CD

F (

tx.

time)

Tx. time (normalized)

Page 26: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Cooperation in Infrastructure-Based Networks

Page 27: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Cooperation in Infrastructure-Based Networks

• Downlink

– Base station cooperation

– Relay cooperation

• Uplink

– Similar to schemes we have seen thus far

• [Lee & Leung, IEEE Trans. Vehicular Technology, 2008]

Page 28: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Base Station (BS) Cooperation

• Much more capable base stations (source nodes)

– Each base station possesses multiple transmit antennas

• CSI shared between base stations

– Extreme case: Full CSI at all BSs

• Benefit: Significantly better co-channel interference

management

BS1 BS2

MS1

MS2

Page 29: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Giant MIMO Array: Transmission Techniques

• Linear precoding

– Generalized Zero Forcing (GZF)

– SLNR criterion based designs

– Sum rate criterion based designs

• Non-linear techniques

– Dirty paper coding

Page 30: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Base Station Cooperation: Is It Giant MIMO?

No!

BS1 BS2

MS1

MS2

1H 2HSuper BS

MS1

MS2

12 , HH

Page 31: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Interference is fundamentally asynchronous

• Even with perfect timing-advance!

(1)2H

(1)1H

(2)1H

(2)2H

(1)1τ

(2)2τ(1)

(2)1τ

BS1 BS2

MS2

MS10 0

(1) (1)2 1τ τ−

(2) (2)2 1τ τ−

[Zhang, Mehta, Molisch & Zhang, IEEE Trans. Wireless Commn. 2008]

Page 32: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Implications on Fundamental System Model

( ) ( ) ( ) ( ) ( )

1 1 1

( ) ( ) ( )B K B

b b b b bk k k k k k jk k

b j b

m m m= = =

= + + ∑ ∑ ∑y H T s H T i n

( ) ( ) ( ) ( )

1 1 1

( ) ( ) ( ) ( )B K B

b b b bk k k k k j j k

b j b

m m m m= = =

= + + ∑ ∑ ∑y H T s H T s n

Changes the basic model!

Should be:

Was:

Generalized zero forcing constraint is no longer sufficient

Channel from BS b to MS k

Precoding at BS b for MS k

Page 33: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Asynchronous Interference-Aware Precoding

• Linear precoding design methods

1. Sum rate maximization (CISVD)

– Non-trivial, non-convex

– Game theoretic approach in DSL: [Yu, Ginis, Cioffi ’02]

2. Mean square error minimization (JWF) – [Zhang, Wu, Zhou, Wang ‘05]

3. Signal to leakage plus noise ratio criterion (JLS) – [Tarighat, Sadek, Sayed ‘05][Dai, Mailaender, Poor ’04]

Page 34: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Modeling Asynchronicity Helps

-5 0 5 10 15 200

2

4

6

8

10

12

Transmit SNR per User(dB)

Ave

rag

e S

pe

ctru

m E

ffici

en

cy P

er

Use

r(b

ps/

HZ

)

JWFJWF: Ignoring async. intf.JLSJLS: Ignoring async. intf.CISVDCISVD: Ignoring async. intf.

• Rate penalty for ignoring asynchronicity is significant

JWF

JLS

CISVD

Transmit SNR per user [dB]

Ave

. spe

ctra

l eff i

cien

cy

(bits

/s/H

z)

2 cell, 2 UE set up

Page 35: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Relay Cooperation System Model

1 11 21 11 21 1 1

2 12 22 12 22 2 2

Y h h b b U N

Y h h b b U N

= +

Received signals

BS-MS channel

Linear precoding

Information symbols

AWGN

• Linear precoding at relays

Page 36: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Asymmetric Relaying Arises Naturally

• Optimal asymmetric linear precoder is unknown!

• Can reduce the dimensionality of the optimization problem considerably

Page 37: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Cross Layer Aspects of Cooperation

Page 38: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Cross-Layer Aspects of Cooperation

• Cooperative MAC

– [Liu, Lin, Erkip, Panwar, IEEE Wireless Commn., 2006]

• Cooperative Hybrid ARQ

– [Zhao & Valenti, IEEE JSAC 2005]

• Cooperative routing

– General routing problem

– Progressive accumulative routing

• Queued cooperation

– [Mehta, Sharma, Bansal, Submitted, 2008]

• Impact of physical layer non-idealities

Page 39: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Cooperative Multi-Hop Routing

• Which relay subset should cooperate in which step?

• Number of possibilities/step: 2N instead of N

• Channel fading: Drives how local the cooperation can be

s

r1

tr2

r3

r4

r5

r6

r7

r9

[Khandani, Abounadi, Modiano & Zheng, Allerton

2003]

Page 40: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Reducing Problem to Conventional Routing Problem

• Only allow nodes k edges/hops apart to cooperate

• Construct hyper graph of neighbour nodes

• Determine optimal cooperation/non-cooperation scheme to transmit between

neighbours

• Assign energy cost to each edge in hyper graph

• Distributed conventional routing algorithms now applicable to determine best

multihop route from source to destination, e.g., Belman-Ford routing

[Madan, Mehta, Molisch, Zhang, Allerton 2007]

Page 41: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Progressive (Energy) Accumulative Routing

s

r1

tr2

r3

r4r6

• Nodes do not discard previous transmissions in a route

• Energy-efficient unicast, multicast and broadcast

Unicast: [Yim, Mehta, Molisch & Zhang, IEEE Trans. Wireless Commn., 2008]Broadcast/Multicast routing: [Maric & Yates, IEEE JSAC 2002, 2005]

Page 42: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

1st Relay Addition: Necessary & Sufficient Conditions

• A node r helps if and only if

(Any eligible node can

overhear source to

destination transmission)

• Source (s) and relay (r) transmit powers for maximal power savings

s t

hrt > hst

(Relay doesn’t help)

hsr > hst

(Relay doesn’t help)

hst < min{hsr,hrt} (Relay saves power)

Page 43: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Progressive Accumulative Routing: Protocol Designs

r

t

s

r

t

q

s

r

t

q

s t

u v

l

w

• Update routes without tearing them down

• Sufficient conditions to add a relay turn out to be nice!

• Packet header fields can be designed so that only local CSI is needed

• How to select optimal relays?

• Optimal relay transmission power?

Page 44: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

s t

u v

l

w

s t u v whwt hwv

MSrc

MDest

RSrc RDest RelayID

GainD GainR

Ready to cooperate packet

Data Packet and Cooperation Packet Structures

PAR Protocol q

s t u v hst/hsq + hqt/hqu hut huv

MSrc

MDest

RSrc RDest FracDelivered GainD GainR

Data

Local CSI info

u to v

w to u

1 1 1wt ut

uw uw uv

h h

h h h

>

+ <

Sufficient conditions to be a useful relay

Energy accumulated thus far

Page 45: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Simulations: Gains from PAR

• 100 nodes distributed uniformly

in a grid of size 20 x 20 grid

• Source at (5,10) and destination

at (15,10)

• Total power consumption

decreases from 100% to 13.6%

to 2.84% to 1.47% and 1.35% in

5 iterations.

Box plot

Number of iterations

Tot

al p

o wer

con

s um

ed

Page 46: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Other Aspects

• Network lifetime maximization and cooperation

– [Himsoon, Siriwongpairat, Han & Liu, IEEE JSAC 2007]

• Distributed detection and estimation using cooperation in

sensor networks

– [Nayagam, Shea & Wong, IEEE JSAC 2007]

• Cognitive radios and cooperation

– [Ganesan & Li, IEEE Trans. Wireless Commn 2007]

Page 47: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Summary and Conclusions

• Cooperation effectively exploits three essential wireless characteristics:– Physical layer spatial diversity

– Broadcast advantage

– Multiple access characteristics of wireless

• Affects physical layer and higher layer design

• Some key problems: – General multihop scenarios

– Cross-layer design with cooperation

– Robust synchronization schemes

– Infrastructure-based cooperation in next generation wireless

Page 48: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

General Case: Multiple Relays (Between Two Relays)

• Sufficient condition for inclusion: Not conducive to distributed implementation

• Only two nodes adjust transmit powers

1 1and lt wt

uw uv wvuv uw lt ut

h hh h h

h h h h

−> − > ÷ −

1 1 1

uw wv uvh h h+ <• Weaker condition:

uuw

Ph

γ= 1 ( ) ut ut wtl

uw uv uw wv

h h hP A l

h h h h

γ = − + − − ÷

Energy accumulated

at last node (l)

s t

u v

l

w

Add node between two relays

(not after last relay)

(Parent

relay)

(Last

relay)

Page 49: Cooperative Comm v3-1

Indian Institute of Science, Bangalore

Master-Slave Architecture for Phase Synchronization

[Mudumbai, Barriac & Madhow, IEEE Trans. Wireless Commn.

2007]