IEEE International Symposium on Information Theory (ISIT)
Ternary Min-Sum Algorithm
Ting-Wei Yang, Qi-Yue Yu
Abstract
Ternary low-density parity-check (LDPC) codes over GF(3) are an appealing non-binary option for transmission with high bandwidth efficiency, but iterative decoding is often dominated by computationally intensive check node (CN) processing. This paper proposes a ternary min-sum (TMS) decoder by deriving an explicit low-complexity CN update that avoids trellis-based forward-backward dynamic programming (DP). Under the min-sum approximation, we apply an a posteriori probability (APP)-centered normalization and reformulate the CN update as a reduced-dimensional kernel minimization. We then prove that the minimizing relative-symbol vector can always be chosen to have Hamming weight at most two. This leads to a DP-free CN update that uses only comparisons and additions, with complexity that scales linearly with the CN degree. Compared to the extended min-sum (EMS) with selection size ne=2, simulation results show that the proposed TMS outperforms EMS by about 0.35 dB at BER near 10−5, and reduces the simulated CN complexity by 85.2% for the considered ternary LDPC code.