Code Design For Relay Channels
Code Design For Relay Channels
Behnaam Aazhang
Credits
Arnab Chakrabarti
Alexandre de Baynast
Ashutosh Sabharwal
Basic Idea
Network-Channel
Explore all available dimensions
Signal
Space
Network
S
D
R
X
1
U
1
U
2
D
Focus
The gap
Information theory results
Algorithms
Protocols
Synchronization--CSI
Coding
Implementation
S
D
R
S
D
Assumptions
Power constraint
AWGN channels with collinear path loss
P P P t tP
MAC MAC BC
R S S
s + + ) )( 1 (
X
1
X
2
Y
1
= h
12
X
1
+ Z
1
Y
0
= h
10
X
1
+ h
20
X
2
+ Z
0
Source
h
10
Z
1
Z
0
h
12
h
20
Relay
Destination
1 ,
) 1 (
1
,
1
=
= =
SD RD SR
d d
o o
Assumptions
Fading
AWGN relay codes could be used!
Channel (network) state information
Power control [Ahmed, et. al. 04]
Rate control [Ahmed, et. al. 05]
Relay Operation
Full duplex
Half duplex
Relay will not receive and transmit same time
and same frequency band
S
D
R
S
D
R
1
st
time slot
2
nd
time slot
Broadcast
Multiple access
Relay Function
Fixed relaying
Decode and forward
Estimate and forward
Amplify and forward
Adaptive relaying
Selection
Incremental
}
decode Not to
Decode
)
`
Relay Channel
Achievable rates
BSC, DMC, Binary AWGN, AWGN, fading
Code design
Turbo, LDPC
Convolutional with CRC
S
D
R
X
1
Y
1
Y
0
X
2
Coding for Relay Channels
BC mode
Relay can decode
Destination may not
MAC mode
Relay re-encodes/transmits
Source transmits
New information
Repeats previous information
Destination decodes
correlation
Low SNR: Stay with Binary
Binary relay
Lessons at Low SNR
Binary
Coding
Half duplex
Correlation 1
LDPC
Binary linear block codes
Information vector u is 1xK
Code vector c is 1xn
The kxn generator G produces codes c=uG
Sparse parity check matrix Hc
T
=0
LDPC on bipartite graph
Variable nodes
Check nodes
LDPC
Codes with sparse parity-check nodes
Cycles
Belief propagation
Degree profiles
Large n, only the distribution of node degrees
determines performance
Coding Methodology
Variable nodes sent by R for MAC mode
Check nodes for relay MAC code
Variable nodes for BC mode
Check nodes for BC mode SR code
Additional check nodes for BC mode SD code
Variable nodes sent by S for MAC mode
Check nodes for source MAC code
LDPC Codes
Structural relationship between
SR code in BC
SD code in MAC
SR subgraph of SD
Joint optimization of degree profiles for the two
codes
Additional constraints on degree distributions
Check nodes
Variable nodes
Modified density evolution.
LDPC Codes for Relay
Modeling the densities of messages in
belief propagation
Gaussians
Noise thresholds
Code profiles
High rate
Low rate
Beamforming
First phase of
successive decoding
More iterations
Challenges
SR link in BC mode high rate
Check node large degree
SD link in MAC mode low rate
Check node smaller degree
Variable nodes for BC mode
Check nodes for BC mode SR code
Additional check nodes for BC mode SD code
Subgraph
Challenges
Belief propagation
Speed of convergence
Variance of degree profile
Reliability factor in messages
Weighted belief propagation
TAP: A Mesh Network
Research Platform
Board at Work
Progress
RF board--Daughtercard
2.4GHz ISM band
Half duplex
Baseband processing board
FPGA basedmodular
Front end
Single carrier and OFDM
LDPC coding and decoding
Coherent combing!
Superposition decoding!
Estimate and Forward
Relay transmits an
estimate of
information in MAC
mode
Destination decodes
SD in MAC mode
with side information
Estimate and Forward
Side information for D
Vector quantization
LDPC