A Learning-Based Flexible Spatial Modulation Design for Large-Scale RIS-Assisted Systems

doi:10.1109/tccn.2026.3705811

A Learning-Based Flexible Spatial Modulation Design for Large-Scale RIS-Assisted Systems

2026-06-22

0

PRE

AI

DOI：10.1109/tccn.2026.3705811

Original

Original request for help

Save

Abstract

En 中文

The growing demand for wireless communications has substantially intensified the requirements for both enhancing spectral efficiency and extending coverage in next-generation systems. This paper investigates a flexible spatial modulation design for large-scale reconfigurable intelligent surface (RIS) assisted systems, where the RIS is partitioned into independently controllable blocks and the indices of the activated blocks convey additional information bits. To manage control complexity, all elements within a block share the same phase shift, and the number of available phase patterns per block is dynamically determined by its size, striking a balance between complexity and performance. For optimal block selection, we first propose a Euclidean distance aided selection (EDAS) algorithm. To further reduce computational overhead while preserving reliability, a deep neural network (DNN) based capacity optimization aided selection (COAS) scheme is developed. The DNN learns robust activation patterns that maximize inter-pattern distinguishability and inherently avoid ambiguous combinations, thereby ensuring consistent performance under diverse channel conditions. Simulation results demonstrate that the proposed designs achieve significant improvements in both bit error rate (BER) and misclassification rate compared to various baseline schemes. Moreover, comprehensive analysis reveals that through careful adjustment of block numbers and phase modes, the DNN-based solution can be effectively scaled for practical large-scale RIS deployments.

Keywords:

Reconfigurable intelligent surface (RIS)

flexible spatial modulation (FSM)

Euclidean distance aided selection (EDAS)

capacity optimization aided selection (COAS)

deep neural network (DNN)

A Learning-Based Flexible Spatial Modulation Design for Large-Scale RIS-Assisted Systems

Abstract

Journal

Organization