Artificial Neural Network Based Adaptive Spatial Modulation

Abstract

Adaptive spatial modulation (ASM) is a closed loop feedback transmission technique for multiple-input multiple-output (MIMO) systems, where different modulation orders can be assigned to the transmit antennas based on the available channel conditions. However, the conventional optimal modulation order selection (MOS) schemes in ASM have high computational complexity. In this paper, a supervised learning aided feed-forward artificial neural network (ANN) is proposed to design the MOS in ASM and achieve an effective tradeoff between the system computational complexity and the bit error rate (BER) performance. Specifically, the proposed ANN is utilized to transform the MOS problem in ASM to a multiclass classification problem based on a low search classification method and predict the optimal MOS candidate which maximizes the minimum Euclidean distance. Simulation results reveal that, for a given spectral efficiency (SE), the proposed ANN based ASM scheme outperforms the classical SM scheme and retains the advantages of the conventional ASM scheme but with lower system computational complexity.