1. Construction and Analysis of Non Systematic
Codes on Graphs for Redundant Data
Amira A LLOUM
Ecole Nationale Sup´ rieure des T´ l´ communications
e ee
Telecom Paris Tech
September 5th, 2008
2. Presentation Outline
Introduction and Motivations
Part I: Non Systematic LDPC Codes Constructions
Part II: Density Evolution Analysis for Split-LDPC Codes
Part III: Exit Chart Analysis for Split-LDPC Codes
Part IV : EM for Joint Source-Channel Estimation
Conclusions and Future Work
3. Introduction
The Non Uniform Assumption
0.5
0 1
P (s = 1) = 0.5
Source Encoder Channel Encoder Channel Channel Decoder Source Decoder
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 1 / 35
4. Introduction
The Non Uniform Assumption
Source Encoder Source Decoder
0.9
0 Channel
0.1 Channel Encoder Channel Decoder
1
P (s = 1) = µ
The Uniform Assumption is not valid anymore When
1 It is not worth to compress (bad channel conditions)
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 1 / 35
5. Introduction
The Non Uniform Assumption
Source Encoder Source Decoder
0.9
0 Channel
0.1 Channel Encoder Channel Decoder
1
P (s = 1) = µ
Source Encoder Source Decoder
The Uniform Assumption is not valid anymore When
1 It is not worth to compress (bad channel conditions)
2 Using sub-optimal compression (highly redundant sources)
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 1 / 35
6. Introduction
Channel Coding Strategies for Non Uniform Sources
0.9
P (s = 1) = µ
0 0.1
1
Source Channel Encoder Channel Channel Decoder Sink
Shannon has intuited in his 1948 Paradigm
Any Redundancy in the source will usually help if it is utilized at the receiveing
...This redundancy will help to combat noise
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 2 / 35
7. Introduction
Channel Coding Strategies for Non Uniform Sources
0.9
P (s = 1) = µ
0 0.1
1
Source Channel Encoder Channel Channel Decoder Sink
SCCD
µ
Channel Coding for Redundant data follows the following strategies
1 Source Controlled Channel Coding (Hagenauer 1995).
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 2 / 35
8. Introduction
Channel Coding Strategies for Non Uniform Sources
0.9
P (s = 1) = µ Systematic Coding
0 0.1
1 Info Parity
Source Channel Encoder Channel Channel Decoder Sink
Parity
µ Non−Systematic Coding µ
Best Constructions
Channel Coding for Redundant data follows the following strategies
1 Source Controlled Channel Coding (Hagenauer 1995).
2 Non Systematic Encoding Structures (Shamai and Verdu 1997)
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 2 / 35
9. Introduction
Information Theoretical Limits wih Redandancy: AWGN Channel
Capacity limit Versus Source Entropy, Coding Rate = 0.5, AWGN Channel
2
0
Minimum Achievable Eb/N0 (dB)
-2
-4
-6
-8
-10
Systematic code, BPSK input
Non-Systematic code, BPSK input
Gaussian input
-12
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Source Entropy (bits)
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 3 / 35
10. Introduction
Information Theoretical Limits wih Redandancy: AWGN Channel
Capacity limit Versus Coding Rate , Source Entropy= 0.5, AWGN Channel
6
Systematic codes, BPSK input
5 Non-Systematic code, BPSK input
Gaussian input
4
3
Minimum Achievable Eb/N0 (dB)
2
1
0
-1
-2
-3
-4
-5
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Coding Rate
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 4 / 35
11. Introduction
Achieving the Theoretical Limits 1
In the presence of Redundancy
The Theoretical Limits of Information Theory are mooving to better
regions.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 5 / 35
12. Introduction
Achieving the Theoretical Limits 1
In the presence of Redundancy
The Theoretical Limits of Information Theory are mooving to better
regions.
To Attain these Challenging limits :
1 Building Non Systematic Capacity Achieving Encoding Structures In the
Codes On graphs Family.
2 Using Source Controlled Channel Decoding with Iterative Algorithms In
the Sum-Product Algorithms Family.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 5 / 35
13. Introduction
Related Work and Motivation
Non−Systematic
Codes On Graph
Turbo Codes LDPC
Non−Systematic MN Codes for BSC
Alajaji Codes et.al. LDPC Codes
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 6 / 35
14. Introduction
Main System Assumptions
0.9
P (s = 1) = µ
0 0.1
1
Source Channel Encoder Channel Channel Decoder Sink
Lossless source coding or no source coding.
Binary i.i.d. source with entropy Hs = H2 (µ), where µ = P (si = 1).
Source sequence s = (s1 , s2 , ..., sK ) encoded by a binary channel code of
rate Rc = K/N , dimension K, and length N .
x = (x1 , ..., xN ) denotes the codeword (channel input).
Transmitted information rate R = Hs × Rc bits per channel use.
Any symmetric binary-input channel can be considered, mainly BEC,
BSC, and AWGN.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 7 / 35
16. Encoding Structures Decoding Strategy Simulation Results
Scrambling-LDPC Encoding Structure
s u c(u,v)
Cs G
Scramble LDPC
c = G × Cs × s
v
(1−R).N β (1−R).N
v db
LDPC
Cs : sparse matrix of dimension
β K × K. In the regular case row and
u
db
R.N dc
u
column weight are ds
u: systematic bits for the inner
LDPC.
s α
ds v: parity bits for the inner LDPC.
R.N s Cs α R.N
ds
s: source bits.
s α
Scrambler
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 8 / 35
17. Encoding Structures Decoding Strategy Simulation Results
Splitting-LDPC Encoding Structure
s u c(u,v)
C−1
s G
Splitter LDPC
−1 × s
c = G × Cs
v
(1−R).N β (1−R).N
v
LDPC dc
db Cs : full rank sparse matrix of
u
β dimension K × K. In the regular
R.N case row and column weight are ds
u
Cs u: systematic bits for the inner
ds
ds
Splitter
LDPC.
s α
v: parity bits for the inner LDPC.
s α
R.N R.N s: source bits.
s α
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 9 / 35
18. Encoding Structures Decoding Strategy Simulation Results
Source Controlled Sum-Product Decoding
The bitnode Rule: v
β
v
LLRci →pcei =LLRtype + LLRpcej →ci LLR0+ Extrinsic Information
∗
pcej ∈Sc
i
β
u
LLR0 if bitnode ∈ {u, ϑ}
LLRtype = u
LLRs if bitnode ∈ {s}
s α
LLR0 is the channel observation LLR.
1−µ s α
LLRs = log( ) is the source LLR .
µ LLRs+ Extrinsic Information
s α
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 10 / 35
19. Encoding Structures Decoding Strategy Simulation Results
Source Controlled Sum-Product Decoding
The bitnode Rule:
v
β
LLRci →pcei =LLRtype + LLRpcej →ci
∗
v
pcej ∈Sc
i
LLR0 if bitnode ∈ {u, ϑ} β
LLRtype = u Extrinsic Information
LLRs if bitnode ∈ {s}
u
The checknode Rule:
s α
LLRcj →pcei
LLRpcei →ci = 2 tanh−1 tanh( ) s α
∗
2
cj ∈Spce
i
s α
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 10 / 35
20. Encoding Structures Decoding Strategy Simulation Results
Source Controlled Sum-Product Decoding
The bitnode Rule:
v
β
LLRci →pcei =LLRtype + LLRpcej →ci
∗
v
pcej ∈Sc LLR0+ Extrinsic Information
i
LLR0 if bitnode ∈ {u, ϑ} β
LLRtype = u
LLRs if bitnode ∈ {s}
u Extrinsic Information
The checknode Rule:
s α
LLRcj →pcei
LLRpcei →ci = 2 tanh−1 tanh( ) s α
∗
2 LLRs+ Extrinsic Information
cj ∈Spce
i
s α
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 10 / 35
21. Encoding Structures Decoding Strategy Simulation Results
Scrambling or Splitting: Information Theoretical Comparaison
Mutual information vs. Eb /N0 for Hs = 0.5 and coding rates Rc = 0.5 (AWGN Channel).
1.8
1.6
Gaussian input
1.4 Non-Systematic code, BPSK input
Scrambled ds=5
Scrambled ds=3
1.2 Systematic codes, BPSK input
Mutual information
1
0.8
0.6
0.4
0.2
0
-10 -5 0 5 10
Eb/N0(dB)
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 11 / 35
22. Encoding Structures Decoding Strategy Simulation Results
Finite-length Performance
1E-01 1E+00
(3,6) systematic LDPC without SCCD (3,6) systematic LDPC without SCCD
(3,6) systematic LDPC with SCCD (3,6) systematic LDPC with SCCD
scrambler ds=3 scrambler ds=3
MN code db=3 MN code db=3
splitter ds=4 splitter ds=4
1E-02 1E-01
Frame Error Rate
Bit Error Rate
1E-03 1E-02
1E-04 1E-03
1E-05 1E-04
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 -2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5
SNR SNR
Bit (left) and word (right) error probabilities vs. signal to noise ratio for codes with rate
Rc = 1/2 length N = 2000 and non uniform source distribution µ = 0.1 : systematic (3,6)
LDPC with and without SCCD, split-LDPC with ds = 4, scramble-LDPC ds = 3 and MN
Codes db = 3.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 12 / 35
23. Encoding Structures Decoding Strategy Simulation Results
Finite-length Performance (2/2)
1E+00
(3,30) systematic LDPC with SCCD
scrambler ds=5
splitter ds=7
1E-01
Frame Error Rate
1E-02
1E-03
1E-04
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3 3.5 4
SNR
Word error probabilities vs. signal to noise ratio Eb /N0 for codes with rate Rc = 0.9 length
N = 2000 and non uniform source distribution µ = 0.1 : systematic (3,30) LDPC with SCCD,
scramble-LDPC with ds = 5 and split-LDPC with ds = 7.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 13 / 35
25. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
Motivation and Contribution
The problem:
How close do Split-LDPC structures approach the Challenging
asymptotical limits ?
Exploring the Split-LDPC asymptotical convergence behaviour.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 14 / 35
26. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
Motivation and Contribution
The problem:
How close do Split-LDPC structures approach the Challenging
asymptotical limits ?
Exploring the Split-LDPC asymptotical convergence behaviour.
The proposal: Density Evolution Analysis
1 Deriving a Density Evolution algorithm for Split-LDPC codes.
2 Investigating the stability issues related to the decoder convergence.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 14 / 35
27. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
Density Evolution Assumptions
1 Concentration and the local tree Assumption.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 15 / 35
28. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
Density Evolution Assumptions
1 Concentration and the local tree Assumption.
3 types of message distribution =⇒ Local tree assumption over 3 types of trees
Message oriented Density Evolution.
db dc p3(x)
ϑ β ϑ β
3 dc
db
p2(x)
u ds u
2 ds
p1(x)
s
s
1 1 1
α
α
Messages Distributions
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 15 / 35
29. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
Density Evolution Assumptions
1 Concentration and the local tree Assumption.
2 Symmetry Conditions (Channel, variable nodes, Checknodes).
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 15 / 35
30. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
Density Evolution Assumptions
1 Concentration and the local tree Assumption.
2 Symmetry Conditions (Channel, variable nodes, Checknodes).
3 The all-zero codeword restriction.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 15 / 35
31. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
Density Evolution Assumptions
1 Concentration and the local tree Assumption.
2 Symmetry Conditions (Channel, variable nodes, Checknodes).
3 The all-zero codeword restriction.
0 0 0 0 0 /
∈B
A Typical Set
C[A]=2k
B
C(B)=2K×Hs
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 15 / 35
32. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The DE statement: the Checknode level
u 0 0 0 0
{
0
s 0
s Averaging
p1(x) α 1
u 1 0 0 0
{
ds − 1 0
u u
pm (x) = Rc ps (x), (1 − µ) q1 (x) + µ q1 (−x)
α
m m
where:
1−µ
ps (x) = δ(x − log ) = δ(x − s)
µ
q1 (x) = ρα (pm (x))
m
1
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 16 / 35
33. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The DE statement: the Checknode level
β
R p2(x) + (1 − R) p3(x)
dc − 1
1
2
u + 1ϑ
2
1
2
u + 1ϑ
2
pm (x) = ρ Rc pm (x) + (1 − Rc ) pm (x)
β 2 3
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 16 / 35
34. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The DE statement: the Checknode level
s
β
α p1(x) R p2(x) + (1 − R) p3(x)
ds − 1 dc − 1
u u 1
u + 1ϑ 1
u + 1ϑ
2 2 2 2
pm (x) = Rc ps (x), (1 − µ) q1 (x) + µ q1 (−x)
α
m m
pm (x) = ρ Rc pm (x) + (1 − Rc ) pm (x)
β 2 3
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 16 / 35
35. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The DE statement: the variable-node level
p1 α
u
ds − 1 db
η(x) xλ(x)
α α β β
p1 R p2 + (1 − R) p3
ds − 1 dc − 1
1 1 1
u u
2
u + 2
ϑ 2
u + 1ϑ
2
p1 (x) at the (m + 1)th iteration
pm+1 (x) = p0 (x) ⊗ λ1α pm (x)
1 α ⊗ λ1 pm (x)
β
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 17 / 35
36. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The DE statement: the variable-node level
β
p2
u
ds db − 1
xη(x) λ(x)
α α β β
p1 R p2 + (1 − R)p3
ds − 1 dc − 1
u u 1
+ 1 1
+ 1ϑ
2u 2ϑ 2u 2
p2 (x) at the (m + 1)th iteration
pm+1 (x) = p0 (x) ⊗ λ2α pm (x)
2 α ⊗ λ pm (x)
β
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 17 / 35
37. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The DE statement: the variable-node level
β
p3
ϑ
db − 1
R p2 + (1 − R)p3 β β β
dc − 1
1 1 1
2u + 2ϑ 2u + 1ϑ
2
p3 (x) at the (m + 1)th iteration
pm+1 (x) = p0 (x) ⊗ λ pm (x)
3 β
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 17 / 35
38. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
2
lambda(x)=0.32660x+0.11960x^2+0.18393x^3+0.36988x^4
rho(x)=0.78555x^5+0.21445x^6
0
Minimum Achievable Eb/N0 (dB) -2
-4
-6
-8
Systematic code, BPSK input
-10
Non-Systematic code, BPSK input
Gaussian input
split-LDPC code, DE thresholds
-12
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Source Entropy (bits)
Minimum achievable Eb /N0 versus source entropy Hs for a regular ds = 3 splitter
concatenated to an irregular LDPC code of rate Rc = 1/2 over a BIAWGN channel.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 18 / 35
39. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The asymptotical behaviour of the decoder in the neighborhood of δ∞
1−R
ϑ ϑ β
CHANNEL P0
u u
Pu0
R
s BSC (µ δ−s + (1 − µ) δs )⊗ ds
µ
ds
s BSC
µ
s BSC
µ
Pu0 = p0 (x) ⊗ (µ δ−s + (1 − µ) δs )⊗ dS
Proposition 1
In the neighborhood of δ∞ , the message density given by the splitter to the core LDPC
(from node α to node u) is equivalent to the initial message density of ds parallel
concatenated BSC with a crossover probability µ.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 19 / 35
40. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The asymptotical behaviour of the decoder in the neighborhood of δ∞
1−R
ϑ ϑ β
CHANNEL P0
u u
Pu0
R
s BSC (µ δ−s + (1 − µ) δs )⊗ ds
µ
ds
s BSC
µ
s BSC
µ
Pu0 = p0 (x) ⊗ (µ δ−s + (1 − µ) δs )⊗ dS
Proposition 2
For non-uniform sources,type-1 message distribution (from node u to node α) shows a
permanent stability around the fixed point δ∞ .
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 19 / 35
41. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The asymptotical behaviour of the decoder in the neighborhood of δ∞
1−R
ϑ ϑ β
CHANNEL P0
u u
Pu0
R
s BSC (µ δ−s + (1 − µ) δs )⊗ ds
µ
ds
s BSC
µ
s BSC
µ
Pu0 = p0 (x) ⊗ (µ δ−s + (1 − µ) δs )⊗ dS
Proposition 3
When close to zero error rate, stability of the LDPC constituent is not disturbed by the
splitter.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 19 / 35
42. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The asymptotical behaviour of the decoder in the neighborhood of δ∞
1−R
ϑ ϑ β
CHANNEL P0
u u
Pu0
R
s BSC (µ δ−s + (1 − µ) δs )⊗ ds
µ
ds
s BSC
µ
s BSC
µ
Pu0 = p0 (x) ⊗ (µ δ−s + (1 − µ) δs )⊗ dS
If the embedded LDPC is stable the Split-LDPC would be so.
The inverse is not true !
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 19 / 35
43. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
Stability Condition for Split-LDPC
CHANNEL Systematic LDPC
p0(x)
The General Stability Condition for a systematicLDPC:
B(p0 )λ (0) ρ (1) ≤ 1
+∞
where B(p0 ) = p0 (x)e−x/2 dx is the Bhattacharyya constant of the channel.
−∞
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 20 / 35
44. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
Stability Condition for Split-LDPC
p0 (x) 1−R
ϑ CHANNEL ϑ β
Averaging
EQUIVALENT
u u
CHANNEL
R
ds
Pu0 = p0(x) ⊗ (µ δ−s + (1 − µ) δs )⊗
GLOBAL EQUIVALENT CHANNEL Systematic CORE−LDPC
Peq = R pu0 + (1 − R) p0(x)
The General Stability Condition for Split-LDPC:
B(Peq )λ (0) ρ (1) ≤ 1
where Peq = Rc Puo (x) + (1 − Rc ) p0 (x) is the initial message density of the global
equivalent channel; and B(Peq ) is the Bhattacharyya constant of the global equivalent
channel.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 20 / 35
45. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The Splitter Asymptotical Properties:
Proposition 4
For uniform sources, the threshold and the stability condition of the
split-LDPC code are the same for the CORE-LDPC
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 21 / 35
46. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The Splitter Asymptotical Properties:
Proposition 4
For uniform sources, the threshold and the stability condition of the
split-LDPC code are the same for the CORE-LDPC
Example
Split-LDPC Stability Condition for BEC Channel:
1 1
λ (0) ρ (1) < ×
ε [(1 − Rc ) + Rc (2 µ(1 − µ))ds ]
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 21 / 35
47. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The Splitter Asymptotical Properties:
Proposition 4
For uniform sources, the threshold and the stability condition of the
split-LDPC code are the same for the CORE-LDPC
Example
Split-LDPC Stability Condition for BSC Channel:
1 1
λ (0) ρ (1) < ×
2 λ(1 − λ) [(1 − Rc ) + Rc (2 µ(1 − µ))ds ]
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 21 / 35
48. DE Motivation and Assumption DE statement Simulation Results Stability Analysis
The Splitter Asymptotical Properties:
Proposition 4
For uniform sources, the threshold and the stability condition of the
split-LDPC code are the same for the CORE-LDPC
Example
Split-LDPC Stability Condition for AWGN Channel::
1 1
λ (0) ρ (1) < e 2σ2 ×
[(1 − Rc ) + Rc (2 µ(1 − µ))ds ]
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 21 / 35
50. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Motivation and Proposal
The problem
High computational complexity for Density Evolution
Low Complexity Approaches based on Gaussian Approximation
Lower Complexity and one-dimensional
More Insightful
Less Accurate
How to build capacity achieving Split-LDPC ?
The Proposal:
Message oriented Bi-dimensional low complexity approach based on a more
accurate Exit Chart method.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 22 / 35
51. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Assumptions and Notations:
Message-oriented local tree assumption.
p1 α
u
ds − 1 db
η(x) xλ(x)
α α β β
p1 R p2 + (1 − R) p3
ds − 1 dc − 1
1 1 1
u u
2
u + 2ϑ 2
u + 1ϑ
2
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 23 / 35
52. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Assumptions and Notations:
Message-oriented local tree assumption.
Consistency Condition is realized on all types of message distribution.
f (x) = f (−x) ex for all x ∈ R+
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 23 / 35
53. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Assumptions and Notations:
Message-oriented local tree assumption.
Consistency Condition is realized on all types of message distribution.
Gaussian approximation is applied at the bitnodes output based on equal
mutual information.
Uout
U0
Uout f
f −1 Uin
U0
Uin
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 23 / 35
54. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Assumptions and Notations:
Message-oriented local tree assumption.
Consistency Condition is realized on all types of message distribution.
Gaussian approximation is applied at the bitnodes output based on equal
mutual information.
We display the error probability as a measure of knowledge U .
Uout
U0
+∞
σ 1 t2
Perr = Q( ) = √ e− 2 dt
2 2π σ/2
Uout f
f −1 Uin
U0
Uin
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 23 / 35
55. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
The EXIT Chart Statement for Split-LDPC : Message Combining
β ∗
P2 = (1 − R)P3 + RP2 β
P2 P3
u
ϑ
ds db − 1
db − 1
xη(x) λ(x)
Pin2
∗ β β β
α α β β
Pin1
∗
dc − 1
ds − 1 dc − 1
1 1 1
u u 1
u + 1ϑ 1
u + 1ϑ 2u + 2ϑ 2u + 1ϑ
2
2 2 2 2
∗ ∗ ∗
Pout2 = G(Pin1 , Pin2 )
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 24 / 35
56. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
The EXIT Chart Statement for Split-LDPC : Message Combining
β ∗
P2 = (1 − R)P3 + RP2 β
P2 P3
u
ϑ
ds db − 1
db − 1
xη(x) λ(x)
Pin2
∗ β β β
α α β β
Pin1
∗
dc − 1
ds − 1 dc − 1
1 1 1
u u 1
u + 1ϑ 1
u + 1ϑ 2u + 2ϑ 2u + 1ϑ
2
2 2 2 2
db db
G(x, y) = Rc × λj fds +1,j (x, y) + (1 − Rc ) × λj gj (y)
j=2 j=2
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 24 / 35
57. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
The EXIT Chart Statement for Split-LDPC : Message Combining
∗ α
P1
u
ds − 1 db
η(x) xλ(x)
α α β β Pin2
∗
Pin1
∗
ds − 1 dc − 1
u u 1
u + 1ϑ 1
u + 1ϑ
2 2 2 2
∗ ∗ ∗
Pout1 = F (Pin1 , Pin2 )
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 24 / 35
58. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
The EXIT Chart Statement for Split-LDPC : Message Combining
∗ α
P1
u
ds − 1 db
η(x) xλ(x)
α α β β Pin2
∗
Pin1
∗
ds − 1 dc − 1
u u 1
u + 1ϑ 1
u + 1ϑ
2 2 2 2
db
F (x, y) = λj fds ,j+1 (x, y)
j=2
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 24 / 35
59. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
The validity Range for the Bidimensional Dynamical System
∗ ∗
The mixtures of Gaussian densities associated to inputs Pin1 and Pin2 must
satisfy the two equalities:
db
∗ 1 1
Pin1 = λj erf c m0 + (ds − 1)mα + jmβ
j=2
2 2
db
∗ 1 1
Pin2 = Rc × λj erf c m0 + ds mα + (j − 1)mβ
j=2
2 2
db
1 1
+ (1 − Rc ) × λj erf c m0 + (j − 1)mβ
j=2
2 2
m0 : Mean of messages from the channel.
mα : Mean of messages from the splitter checknodes α.
mβ : Mean of messages from the LDPC checknodes β.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 25 / 35
60. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Trajectory of error probability near the code threshold
Es
Right to the threshold: = −5.58dB, Threshold=−5.68dB. Final fixed point is 0.
N0
Transfer Function F(x,y)
Trajectory of Pout1
Pout1 0.20
0.25 0.15
0.20
0.10
0.15
0.10 0.05
0.05
0.00
0.00
0.20
0.25
0.15
0.20
0.15 0.10 Pin1
0.10
0.05
Pin2 0.05
0.00
Illustration for an irregular Split-LDPC code.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 26 / 35
61. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Trajectory of error probability beyond the code threshold (2)
Es
Left to the threshold: = −5.78dB, Threshold=−5.68dB. Final fixed point is non-zero.
N0
Transfer Function F(x,y)
Trajectory of Pout1
Pout1 0.20
0.15
0.25
0.20 0.10
0.15
0.05
0.10
0.05 0.00
0.00
0.20
0.25 0.15
0.20
0.10 Pin1
0.15
0.10 0.05
Pin2 0.05
0.00
Illustration for an irregular Split-LDPC code.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 27 / 35
62. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Open Tunnel for the Bi-dimensional EXIT Chart
Open tunnel obtained by plotting the trajectory of error probability and its z = x plane
Es
reflection at = −5.58dB .
N0
Transfer Function F(x,y)
Pout1 Trajectory of Pout1
Inverse of Trajectory of Pout1
Plan z=x
0.18
0.25
0.16
0.20 0.14
0.12
0.15 0.10
0.08
0.10 0.06
0.04
0.05 0.02
0.00
0.00
0.25
0.20
0.15
0.10 Pin2
0.00 0.05 0.10 Pin1 0.15 0.20 0.05
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 28 / 35
63. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Design Irregular Split-LDPC codes
The Design linear Program:
1 Maximize λi /i
i≥2
2 subject to : λi ≥ 0, λi = 1
i≥2
∗ ∗
3 and ∀(P1in , P2in ) ∈ T (S)
db ∗ ∗
j=2 λj fds ,j+1 (P1in , P2in∗ ) < P1in
db ∗ ∗ ∗ ∗
j=2 λj [Rc × fds +1,j (P1in , P2in ) + (1 − Rc ) × gj (P2in )] < P2in
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 29 / 35
64. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Design Irregular Split-LDPC codes
The Design Process:
Select a regular code with the desired coding rate.
∗ ∗
Use the regular code degree sequence for mapping (P1in , P2in ) to the
appropriate input Gaussian mixture densities.
Find the EXIT charts surfaces for different variable degrees.
Find a linear combination with an open EXIT chart that maximizes the
rate and meets all the required design criterion.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 29 / 35
65. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Design Irregular Split-LDPC codes
The Design Process:
Select a regular code with the desired coding rate.
∗ ∗
Use the regular code degree sequence for mapping (P1in , P2in ) to the
appropriate input Gaussian mixture densities.
Find the EXIT charts surfaces for different variable degrees.
Find a linear combination with an open EXIT chart that maximizes the
rate and meets all the required design criterion.
Best Approach to Shannon limits is within 0.1 dB !
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 29 / 35
66. Motivation and Assumptions Split-LDPC EXIT Chart Statement Design of Irregular Split-LDPC Results
Bi-dimensional EXIT Chart Accuracy
Error in approximation of threshold (dB) using EXIT chart analysis for various split-LDPC
codes of Rate one-half with ds = 3 , Hs = 0.5. ∆ is the log-ratio quantization step.
Eb /N0 ∗ (dB) Eb /N0 ∗ (dB) Error ∆Eb /N0 ∗
db dc Rate DE ∆ = 0.005 EC ∆ = 0.005 for ∆ = 0.005
3 6 0.5 −2.22 −2.199 0.020
4 8 0.5 −1.12 −1.129 0.009
5 10 0.5 −0.32 −0.369 0.049
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 30 / 35
68. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
Motivation and Proposal
Motivation
Full utilization of systematic and non-systematic LDPC codes requires:
1 Knowledge of the source probability distribution (SSI) at the decoder side.
2 Knowledge of channel parameters (CSI) at the decoder side.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 31 / 35
69. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
Motivation and Proposal
Motivation
Full utilization of systematic and non-systematic LDPC codes requires:
1 Knowledge of the source probability distribution (SSI) at the decoder side.
2 Knowledge of channel parameters (CSI) at the decoder side.
Proposal
Joint Source-Channel Iterative Estimation and Decoding for non-uniform
sources based on the Expectation Maximization Algorithm (EM)
→ No performance loss → Negligible estimation complexity
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 31 / 35
70. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
Brief statement of the EM algorithm
κ : complete data, κ = (x, y)
0.9
x:missing data y: incomplete data (observed)
p(s = 1) = µ
0.1 0
1 x y
Source Channel Encoder Channel Channel Decoder Sink
SSI = µ CSI
Θ=SSI+CSI
set of parameters to be estimated, SSI+CSI ˆ
Θ
E-step: Compute the Auxiliary function Q:
Q(θ|θi ) = E[log p(x, y|θ)|y, θi ]
= log[P (y|x, θ) P (x|θ)] AP Pi (x)
x
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 32 / 35
71. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
Brief statement of the EM algorithm
κ : complete data, κ = (x, y)
0.9
x:missing data y: incomplete data (observed)
p(s = 1) = µ
0.1 0
1 x y
Source Channel Encoder Channel Channel Decoder Sink
SSI = µ CSI
Θ=SSI+CSI
set of parameters to be estimated, SSI+CSI ˆ
Θ
E-step: Compute the Auxiliary function Q:
Q(θ|θi ) = E[log p(x, y|θ)|y, θi ]
= log[P (y|x, θ) P (x|θ)] AP Pi (x)
x
The SSI part in the auxiliary function is:
P (x|θ) ≡ P (s|θ) = µωH (s) (1 − µ)K−ωH (s)
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 32 / 35
72. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
Brief statement of the EM algorithm
κ : complete data, κ = (x, y)
0.9
x:missing data y: incomplete data (observed)
p(s = 1) = µ
0.1 0
1 x y
Source Channel Encoder Channel Channel Decoder Sink
SSI = µ CSI
Θ=SSI+CSI
set of parameters to be estimated, SSI+CSI ˆ
Θ
E-step: Compute the Auxiliary function Q:
Q(θ|θi ) = E[log p(x, y|θ)|y, θi ]
= log[P (y|x, θ) P (x|θ)] AP Pi (x)
x
M-step:
θ i+1 = arg max Q(θ|θ i )
θ
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 32 / 35
73. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
Joint Source-Channel Estimation on Complex BIAWGN
The Complex BIAWGN channel is defined as
φ
√
yi = Ae xi + ηi with = −1
where the three CSI parameters are :
The amplitude A , which is real positive
The phase ambiguity φ, which is uniformly distributed between 0 and 2π.
The Gaussian noise variance σ 2 .
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 33 / 35
74. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
Joint Source-Channel Estimation on Complex BIAWGN
The Complex BIAWGN channel is defined as
φ
√
yi = Ae xi + ηi with = −1
E-step: Auxiliary Function
µ
Q(θ|θi ) = log[ ]˜ + K log[(1 − µ)] − N log[2πσ 2 ]
s
1−µ
N N N
1 2 A2 2 A
− yj − xj + R xj ∗ e−
˜ φ
yj
2σ 2 j=1
2σ 2 j=1
σ2 j=1
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 33 / 35
75. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
Joint Source-Channel Estimation on Complex BIAWGN
The Complex BIAWGN channel is defined as
φ
√
yi = Ae xi + ηi with = −1
M-step: Maximizing the auxiliary function for SSI
K
i+1 j=1 sj
˜
µ =
K
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 33 / 35
76. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
Joint Source-Channel Estimation on Complex BIAWGN
The Complex BIAWGN channel is defined as
φ
√
yi = Ae xi + ηi with = −1
M-step: Maximizing the auxiliary function for CSI
N φi
j=1 ˜ ∗
R xj yj e
Ai+1 = N 2
j=1 xj AP Pi (xj ) xj
N
2 (i+1) 1 φi x 2
σ = yj − Ai e j
2N j=1
N
φi+1 = − Arg ˜ ∗
xj yj
j=1
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 33 / 35
77. Motivation and Proposal Expectation-Maximization Complex BIAWGN Channel Simultations Results
100
Systematic + Uniform
Systematic + Non Uniform (Perfect)
Systematic + Non Uniform (EM observation)
Systematic + Non Uniform (EM APP)
Non Systematic + Non Uniform (Perfect)
10-1 Non Systematic + Non Uniform (EM APP)
Frame Error Rate
10-2
10-3
10-4
-2.4 -2 -1.6 -1.2 -0.8 -0.4 0 0.4 0.8 1.2 1.6 2 2.4
Eb/N0 (dB)
Systematic and non-systematic LDPC codes on AWGN, Rc = 1/2.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 34 / 35
78. Conclusions and Open Issues
Conclusions
The theoretical limits for redundant data are mooving to better regions.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 35 / 35
79. Conclusions and Open Issues
Conclusions
The theoretical limits for redundant data are mooving to better regions.
Split-LDPC codes are the best non systematic LDPC constructions.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 35 / 35
80. Conclusions and Open Issues
Conclusions
The theoretical limits for redundant data are mooving to better regions.
Split-LDPC codes are the best non systematic LDPC constructions.
Split-LDPC codes have good asymptotical performance and properties.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 35 / 35
81. Conclusions and Open Issues
Conclusions
The theoretical limits for redundant data are mooving to better regions.
Split-LDPC codes are the best non systematic LDPC constructions.
Split-LDPC codes have good asymptotical performance and properties.
Designed irregular Split-LDPC codes are capacity achieving.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 35 / 35
82. Conclusions and Open Issues
Conclusions
The theoretical limits for redundant data are mooving to better regions.
Split-LDPC codes are the best non systematic LDPC constructions.
Split-LDPC codes have good asymptotical performance and properties.
Designed irregular Split-LDPC codes are capacity achieving.
The 2-D EXIT Chart formalism is lower complexity and as accurate as the
DE.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 35 / 35
83. Conclusions and Open Issues
Conclusions
The theoretical limits for redundant data are mooving to better regions.
Split-LDPC codes are the best non systematic LDPC constructions.
Split-LDPC codes have good asymptotical performance and properties.
Designed irregular Split-LDPC codes are capacity achieving.
The 2-D EXIT Chart formalism is lower complexity and as accurate as the
DE.
The performance using EM is as good as the perfect knowledge case.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 35 / 35
84. Conclusions and Open Issues
Open Issues
Investigating more complex source and channel models.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 36 / 35
85. Conclusions and Open Issues
Open Issues
Investigating more complex source and channel models.
Extending our study to irregular splitting and scrambling LDPC.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 36 / 35
86. Conclusions and Open Issues
Open Issues
Investigating more complex source and channel models.
Extending our study to irregular splitting and scrambling LDPC.
Performing Complexity and Finite-length Optimization.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 36 / 35
87. Conclusions and Open Issues
Open Issues
Investigating more complex source and channel models.
Extending our study to irregular splitting and scrambling LDPC.
Performing Complexity and Finite-length Optimization.
Consider lossy sources assumption .
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 36 / 35
88. Conclusions and Open Issues
Open Issues
Investigating more complex source and channel models.
Extending our study to irregular splitting and scrambling LDPC.
Performing Complexity and Finite-length Optimization.
Consider lossy sources assumption .
Studying and incorporating Split-LDPC in practical communication
systems.
Amira A LLOUM Construction and Analysis of Non-Systematic Codes on Graphs for Redundant Data September 5th, 2008 36 / 35