PSEUDO-CODEWORD LANDSCAPE

CHERTKOV, MICHAEL¹  STEPANOV, MIKHAIL¹ 

[1] Los Alamos National Laboratory

关键词: 99;    A CODES;    LINEAR PROGRAMMING;    PERFORMANCE;    SPECTRA;   

DOI  :  10.2172/1000499
RP-ID  :  LA-UR-07-0144
PID  :  OSTI ID: 1000499
Others  :  TRN: US201101%%634

美国|英语

来源: SciTech Connect

【 摘 要 】

The authors discuss performance of Low-Density-Parity-Check (LDPC) codes decoded by Linear Programming (LP) decoding at moderate and large Signal-to-Noise-Ratios (SNR). Frame-Error-Rate (FER) dependence on SNR and the noise space landscape of the coding/decoding scheme are analyzed by a combination of the previously introduced instanton/pseudo-codeword-search method and a new 'dendro' trick. To reduce complexity of the LP decoding for a code with high-degree checks, {ge} 5, they introduce its dendro-LDPC counterpart, that is the code performing identifically to the original one under Maximum-A-Posteriori (MAP) decoding but having reduced (down to three) check connectivity degree. Analyzing number of popular LDPC codes and their dendro versions performing over the Additive-White-Gaussian-Noise (AWGN) channel, they observed two qualitatively different regimes: (i) error-floor sets early, at relatively low SNR, and (ii) FER decays with SNR increase faster at moderate SNR than at the largest SNR. They explain these regimes in terms of the pseudo-codeword spectra of the codes.