Deep-Unfolded Iterative Soft-Input Soft-Output SIC Receiver for Coded MIMO Systems

This article proposes a novel deep learning (DL)based enhancement of iterative detection and decoding technique in multiple-input multiple-output (MIMO) wireless systems. The receiver integrates a classical soft-input soft-output (SISO) minimum mean squared error successive interference cancellation (SIC) algorithm into its detector and an optimized low-density parity-check channel decoder within a deep-unfolded iterative (DUI) framework. Termed DUI-SISO-SIC, this receiver offers two significant contributions. Firstly, it introduces learnable parameters to enhance the detector’s a-priori information, thereby improving overall performance. Secondly, it employs a data-driven DL approach to address the computational complexity associated with the crucial MMSE-based symbol ordering process during the SIC step in detection. The simulation results demonstrate the receiver’s superiority over state-of-the-art counterparts in a 5G-compliant coded multi-user MIMO orthogonal frequency-division multiplexing system, regardless of perfect channel knowledge availability. Furthermore, the symbol ordering enhancement during detection demonstrates a substantial reduction in its asymptotic upper bound, particularly with larger MIMO dimensions, compared to the classical MMSE-based approach.