Research on Receiver Optimization Algorithm for SCMA System Based on Deep Learning

Sparse code multiple access (SCMA) is an important non-orthogonal multiple access technology for 5G and subsequent communication technologies, and its performance optimization is of great significance to wireless communication systems. However, existing SCMA receiver optimization methods (such as maximum likelihood detection and message passing algorithm) have the problems of high detection complexity and limited performance improvement under complex channel conditions. This paper proposes a SCMA receiver optimization algorithm based on deep convolutional self-attention network (DCSAN-SCMA). The algorithm combines CNN and self-attention mechanism to construct feature extraction and global dependency capture modules, and optimizes signal detection accuracy and interference suppression capability through multi-task loss function. The SCMA system is simulated under different channel conditions (AWGN channel, Rayleigh fading channel). The result shows that DCSAN-SCMA algorithm has better performance than traditional methods both in BER rate and detection complexity. When SNR reaches 10 dB, BER decreases to 0.0025%, 25% more than conventional MPA algorithm; detection complexity reduces 30%, achieves real time detection ability. In addition, the algorithm shows good robustness in multi-user interference suppression and dynamic channel adaptability. Studies have shown that the DCSAN-SCMA algorithm can not only significantly improve the signal detection performance of SCMA receivers, but also has good scalability and can be applied to other non-orthogonal multiple access communication technologies, providing support for the development of future communication technologies.