Sparse Code Multiple Access Scheme Based on Variational Learning
Quan Yuan,Zhenyong Wang,Dezhi Li,Qing Guo,Wei Xiang
DOI: https://doi.org/10.1109/tcomm.2022.3214236
IF: 6.166
2022-12-21
IEEE Transactions on Communications
Abstract:Sparse code multiple access (SCMA) technology has been widely studied thanks to its outstanding overload performance, which provides greater device access on limited time-frequency resources. We propose a variational learning based end-to-end SCMA network model termed as V-SCMA. In our model, SCMA codebooks are generated by a deep neural network (DNN) encoder. Motivated by the idea of variational learning, we derive the posterior probability of SCMA codeword send by each user from the point view of binary coding using the variational method. Then, we design an SCMA decoding network to learn this process of approaching the real posterior probability. The SCMA decoding network designed by us is a truncated recurrent neural network, which is simpler than other networks. In addition, a novel loss function is proposed to optimize V-SCMA. Compared with previous works, the new loss function is a tighter variational lower bound which improves the BER performance. Simulation results show that the proposed V-SCMA offers a better bit error rate (BER) performance and a lower computational complexity than conventional schemes.
telecommunications,engineering, electrical & electronic