Autoencoders for variable active antenna spatial modulation in MIMO systems

Abstract

This paper introduces three novel autoencoder-based frameworks for variable active antenna spatial modulation (AE-VASM) in multiple-input multiple-output (MIMO) systems, aimed at enhancing system robustness and power efficiency in highly correlated Rician fading channels. By integrating dynamic antenna selection with the joint transmitter–receiver optimization capabilities of autoencoders (AEs), the proposed frameworks effectively mitigate signal degradation due to antenna correlation. Framework 1 serves as a baseline model that modulates data symbols and detects antenna indices based on channel variations. Framework 2 improves performance by incorporating antenna index information into the transmitted signal using phase-shift keying (PSK)-based signatures, which enhances detection accuracy in correlated channels. Framework 3 further advances this approach by jointly encoding both data symbols and antenna indices within the autoencoder, leading to superior bit error rate (BER) performance and power efficiency across diverse channel conditions. Simulation results confirm that Framework 3 consistently outperforms the baseline and other frameworks. This demonstrates its potential for next-generation wireless systems demanding robust and efficient modulation under complex channel conditions.