Efficient early termination schemes for belief-propagation decoding of polar codes
Yuanrui Ren, Chuan Zhang, Xing Liu, Xiaohu You · 2015
Polar codes, the first class of codes that can provably achieve the capacity of symmetric binary-input discrete memoryless channels (B-DMCs), have drawn intensive attentions recently. In general, polar codes can be decoded by either successive cancellation (SC) algorithm or belief propagation (BP) algorithm. Unlike SC decoding, BP decoding shows high-parallelization capability and its performance largely depends on the number of iterations. For better compromise of performance and complexity, the efficient metrics for convergence checking are of great importance. To this end, this paper devotes itself to offering efficient early termination schemes based on those metrics. First, the convergence property of BP polar decoding is explored based on the LLR messages. A low-complexity LLR-magnitude aided (LMA) early termination scheme is proposed accordingly. Numerical results have shown that for Imax = 30, the BP decoder with the proposed LMA scheme can achieve as high as 72.6% iteration reduction at Eb/N0= 4.0 dB compared to the BP decoder without LMA scheme. For better sensitivity and lower complexity, the cyclic redundancy check (CRC) aided (CA) early termination scheme is also proposed. Thanks to its even higher sensitivity, for Imax= 30, the BP decoder with CA scheme can achieve as high as 84.5% iteration reduction at Eb/N0=4.0 dB compared to the conventional one.