HAL Id: hal-01849643
https://hal.archives-ouvertes.fr/hal-01849643
Submitted on 26 Jul 2018
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
New criteria for iterative decoding
Florence Alberge, Ziad Naja, Pierre Duhamel
To cite this version:
Florence Alberge, Ziad Naja, Pierre Duhamel. New criteria for iterative decoding. ICASSP, Apr 2009, Taipei, Taiwan. �hal-01849643�
Univ. Paris-Sud, UMR 8506, Orsay
CNRS, Orsay, France
New criteria for iterative decoding
Florence Alberge, Ziad Naja, Pierre Duhamel
International Conference on Acoustic Speech and Signal Processing 09
Laboratoire des Signaux et Systèmes
Supelec, Gif-sur-Yvette, France
Context: Turbo-like algorithms with iterative decoding
Goal: Make the link between iterative decoding and classical optimization tech-
niques. Improve the performance of iterative decoding.
System model (Bit Interleaved Coded Modula-
tion)(Zehavi, 1992; Li, 2002)
Transmission model
Convolutional
Encoder Interleaver Symbol
Mapping Channel
b c d s y
AWGN Channel yk = sk + nk nk N (0, σn2) sk = M apping(dkm+1, dkm+2, ..., d(k+1)m)
BICM-ID receiver with soft-decision feedback
Demapping Deinterleaver SISO
Decoder Interleaver
p(b;O) p(c;O) p(d;I)
p(d;O) p(c;I)
y
APP (Demapping) pAP P(dkm+i = b) = Km′ X
sk∈Ψib
p(yk | sk) Y
j
p(dkm+j; I)
Prior × Channel probability → Marginalization
Extrinsic (Demapping) p(dkm+i = b; O) = Km X
sk∈Ψib
p(yk | sk) Y
j6=i
p(dkm+j; I)
APP (Decoder) pAP P(cl = b) = Kc′ X
c∈Rlb
IC(c) Y
j
p(cj; I) (IC indicator f unction of the code)
Prior × Indicator function → Marginalization
Extrinsic (Decoder) p(cl = b; O) = Kc X
c∈Rlb
IC(c) Y
j6=l
p(cj; I)
Information geometry and iterative decoding
Basic tools and Notations (Walsh, 2006)
(Bi) ∈ {0, 1}N binary representation of integer i gathered into matrix B = (B0, B1, ..., B2N−1)T.
PMF η = (Pr[χ = B0], Pr[χ = B1], ..., Pr[χ = B2N−1])T
Log-coordinates of PMF η (θi)0≤i≤2N−1 = ln(P r[χ = Bi]) − ln(P r[χ = B0])
Bitwise log-probability ratio (λj)0≤j≤N −1 = log(P rP r[χ[χj=1]
j=0])
For factorisable probability measures, the log-coordinates take the form θ = Bλ.
Link with iterative decoding
• Demapping subblock
pAP P(dkm+i = b) = Km′ P
sk∈Ψib p(yk | sk) Q
j p(dkm+j; I)
y
yLog−coordinates
θm Bλ1 pAP P(dkm+i = b) = Km′′ p(dkm+i; I) p(dkm+i; O)
y
yLog−coordinates
Bλ1 Bλ2
The demapping sub-block solves, with respect to λ2, pB(λ
1+λ2) = pBλ1+θm
• Decoding sub-block
pAP P(cl = b) = Kc′ P
c∈Rlb IC(c) Q
j p(cj; I)
y
yLog−coordinates
θc Bλ2
pAP P(cl) = Kc′′ p(cl; I) p(cl; O)
y
yLog−coordinates
Bλ2 Bλ1
The decoding sub-block solves, with respect to λ1, pB(λ
1+λ2) = pBλ2+θc
• Global criterion Let DF D(p, q) = P
j pj ln p
j
qj
+ P
j(1 − pj) ln 1−p
j
1−qj
denote the Fermi-Dirac entropy (Kullback- Leibler distance for bit probabilities).
The demapping sub-block solves the minimization problem minλ2 DF D(pBλ1+θm, pB(λ1+λ2)). The decoding sub-block solves the minimization problem minλ1 DF D(pBλ2+θc, pB(λ1+λ2)).
New criteria
An hybrid proximal point algorithm
Goal: Link the two (independent) criteria using proximal point technique (Luque, 1984).
Demapping min
λ2
Jθm(λ1, λ2) = min
λ2
DF D(pBλ1+θm, pB(λ1+λ2)) + µmDF D(p
B(λ(k)1 +λ(k)2 ), pB(λ1+λ2)) Decoding min
λ1 Jθc(λ1, λ2) = min
λ1 DF D(pBλ2+θc, pB(λ
1+λ2)) + µcDF D(p
B(λ(k)1 +λ(k+1)2 ), pB(λ
1+λ2)) Optimal choice of the step-size :
Choose µm such that Jθm(λ(k)1 , λ(k+1)2 ) ≤ Jθc(λ(k)1 , λ(k)2 ) Choose µc such that Jθc(λ(k+1)1 , λ(k+1)2 ) ≤ Jθm(λ(k)1 , λ(k+1)2 )
Both criteria decrease with the iterations. Convergence towards the same stationary point than the classical iterative decoding.
An hybrid minimum entropy algorithm
Goal: Improve the performance of the iterative decoding.
Rationale: The entropy of the APP gives a measure of the reliability of the decision : EB(λ
1+λ2) → 0 means that the iterative decoding is confident about its decisions (Kocarev, 2006).
Demapping min
λ2 DF D(pBλ1+θm, pB(λ
1+λ2)) + ηmEB(λ
1+λ2)
Decoding min
λ1 DF D(pBλ2+θc, pB(λ
1+λ2)) + ηcEB(λ
1+λ2)
Simulation
The generator polynomial of the encoder is g = [111; 001; 100]. The bits are mapped using subset partitioning to a 8-PSK modulation. The length of the coded bit sequence is Lc = 6000. The
step-sizes ηm and ηc in the HMEA are both chosen equal to 0.05.
4 4.5 5 5.5 6 6.5 7 7.5 8 8.5
10−5 10−4 10−3 10−2 10−1 100
EbN0
BER
Classical iterative decoding Hybrid Proximal Point Hybrid Min Entropy
4 4.5 5 5.5 6 6.5 7 7.5 8 8.5
5 10 15 20 25 30
EbN0
iterations
Classical iterative decoding Hybrid Proximal Point Hybrid Min Entropy
Left: BER versus EbN0 – Right: Iteration number versus EbN0
References
E. Zehavi. 8-PSK trellis codes for a Rayleigh fading channel. IEEE. Trans on Commun., 40:873–883, May 1992.
X. Li, A. Chindapol and J.A. Ritcey. Bit interleaved coded modulation with iterative decoding and 8-PSK signaling. IEEE. Trans on Commun., 50:1250–1257, Aug. 2002.
J.M. Walsh, P.A. Regalia and C.R. Johnson. Turbo decoding as Iterative Constrained Maximum Likelihood sequence Decoding. IEEE Trans on Inform Theory, 52:5426–5437, Dec. 2006.
F.J. Luque. Asymptotic convergence analysis of the proximal point algorithm. SIAM Journal on Control and Optimization, 22(2), pp. 277–293, 1984.
L. Kocarev, F. Lehmann, G.M. Maggio, B. Scanavino, Z. Tatsev and A. Vardy. Nonlinear dynamics of iterative decoding systems: Analysis and Applications. IEEE Trans. on Inform. Theory, 52(4):
1366–1384, 2006.