Abstract:
To solve the problem of poor performance of min-sum (MS) decoding algorithm on parity check codes, an improved MS algorithm is proposed. If the symbol of the new variable node message is different from the previous variable node message, the new variable node message and the previous variable node message are dynamically weighted to modify the variable node message in the iterative process, so as to reduce the adverse effect of MS overestimation. The decoder implemented by the deep learning method can not only suppress the influence of MS approximation, but also suppress the adverse influence of the loop in the code structure. Simulation results show that, compared with MS algorithm, the improved algorithm achieves significant improvement in decoding performance with almost no increase in complexity, and the decoding performance in short and medium codes is better than the classical belief propagation (BP) algorithm.