神经网络辅助的非二进制量子LDPC码置信传播译码研究OA

Belief propagation(BP)algorithm has made many advances in decoding quantum low-density parity check(LDPC)codes.A new recurrent neural network-based belief propagation(RNBP)decoding scheme is proposed for the problems of quantum short-cycle and error degeneracy under non-binary conditions.Specifically,in the loss function section,RNBP combines the cross-entropy loss applicable to classical error patterns,with the logical loss applicable to quantum error patterns,and sets a hyperparameter for this loss function to better accomplish this multi-objective task.Simulation results show that the RNBP decoding scheme mitigates the quantum short-cycle problem by detecting degeneracy errors with a high probability.On the standard CSS form quantum codes tested in this article,the RNBP scheme decodes better than the conventional BP,suppressing the logic error rate from 10-2 to between 10-3 and 10-4.The RNBP also considers asymmetric channel decoding,which also performs better than the conventional BP.On the XZZX form twisted quantum codes,the logic error rate achievable by the conventional BP is suppressed from 10-2 to around 10-3.