An improved SCMA detector based on ResNet perception fusion

The received signals used for sparse code multiple access (SCMA) detection are usually contaminated with noise during transmission, which exposes an issue of low decoding efficiency. To address this issue, a novel detector based on a residual network (ResNet) perception fusion framework (RSMPA) is proposed for uplink SCMA system in this paper. Specifically, we first formulate a joint design of perception system and traditional communication module. A perception framework based on ResNet is applied to cancel the noise component and enhance the communication system performance. The ResNet model is designed and trained using the clean and noisy SCMA signal, respectively. Based on the denoised output, information iteration process is executed for multi-user detection. Simulation results indicate that the perception model achieves an excellent denoising performance for SCMA system and the proposed scheme outperforms the conventional detection algorithms in terms of SER performance.