Learning-Based MIMO Detection With Dynamic Spatial Modulation

Abstract

In this paper, we investigate signal detection in emerging dynamic spatial modulation (DSM) based MIMO systems, where the existing mapping and detection methods do not work efficiently. In order to address this issue, we begin by proposing a combinatorial mapping-based DSM (CM-DSM) scheme in this paper. The proposed CM-DSM scheme employs a combinatorial 3D mapping to address the detection ambiguity by leveraging the combinatorial nature of DSM. Additionally, this mapping helps construct an appropriate decision tree for the optimal signal detection. By leveraging the resulting tree, we further propose a memory-bounded tree search (METS) algorithm, which efficiently finds the maximum likelihood (ML) estimate. To further enhance detection efficiency, we propose a deep learning boosted version of METS (DL-METS), which efficiently reduces the computational complexity via estimating the optimal heuristic function. Simulation results show that both the proposed METS and DL-METS work well in the considered system. In particular, the proposed DL-METS achieves nearly optimal detection performance while maintaining almost the lowest expected computational complexity, which strongly validates the effectiveness of the proposed algorithm.