Neural Network Aided Impulsive Perturbation Decoding for Short Raptor-Like LDPC Codes

As short packet communications become more important, multi-round belief propagation decoding with impulsive perturbation (MBPD-IP) was proposed to improve the decoding of short low-density parity-check (LDPC) codes. To operate the MBPD-IP effectively, it is crucial to select proper symbols to be perturbed in as few trials as possible. Unfortunately, existing perturbation symbol selection criteria are heuristic and not effective for special codes such as raptor-like (RL) LDPC codes. In this letter, a neural network (NN) based perturbation symbol selection scheme for MBPD-IP is proposed where the symbols to be perturbed are selected from a pre-trained NN. It is shown that the proposed scheme performs significantly better than existing schemes for the RL LDPC code of 5G new radio and still works well for plain code structures.