Deep neural network for MIMO-SCMA detection

Abstract

This article introduces deep learning into the multiple-input multiple-output (MIMO) sparse code multiple access (SCMA) system and proposes a MIMO-SCMA detection scheme based on deep neural networks (DNN) to improve bit error rate (BER) performance. The DNN learns the codebook of each user through channel feature learning on different transmission antennas. The fully connected DNN is designed as the decoder at the receiving end, which does not require traditional multi-antenna detection and multi-user detection, and can obtain user data with one decoding operation. The encoder and decoder are trained using an end-to-end training method. All learning models of the DNN are generated offline and the learned models are used for online testing. In this model, the received signal and channel coefficients are set as input data, and the label corresponding to the transmitted symbol is set as output data for offline learning. After offline learning is completed, the model can be deployed online with fixed weights and biases. Through simulation experiments, the proposed DNN encoder-decoder method can reduce the BER and computational complexity of the receiver in the MIMO-SCMA system.