A Model-Driven Network Based on ISTA for Massive MIMO Signal Detection

Massive multiple-input multiple-output (MIMO) is a key technology in the fifth-generation (5G) wireless communication, and signal detection is a crucial module in the physical layer. However, signal detection for massive MIMO system suffers from high computational complexity due to its large scale. To address this problem, a model-driven neural network based on the iterative soft-thresholding algorithm (ISTA) called detection ISTA (DISTA) is proposed. We reformulate the MIMO detection problem into a least absolute shrinkage and selection operator (LASSO) problem and solve it by DISTA effectively. In the proposed approach, all the hyper-parameters are adaptively learnable and a convolutional neural network (CNN) is used as a sparse operator to transform the transmitted signals into sparse vectors. Simulation results demonstrate that the proposed approach can achieve near-optimal performance. Moreover, compared with other state-of-art deep learning based model-driven methods, DISTA has much lower computational complexity.