Low complexity deep neural network based transmit antenna selection and signal detection in SM-MIMO system

Abstract

In this paper, with the investigation of the conventional Euclidean distance based transmit antenna selection (ED-AS) and the maximum likelihood detection (MLD) in spatial modulation (SM) multiple-input multiple-output (MIMO) system, we propose the deep neural network (DNN) based transmit antenna selection (DNN-AS) and signal detection (DNN-SD), respectively, to effectively balance the system performance and the complexity. For the proposed DNN-AS, we transform the existing problem of AS into a prediction problem and design a low dimension multi-output classifier to achieve the low complexity solution of AS. For the DNN-SD, we present two sub-DNNs to recover the transmitted SM signal. Numerical results reveal that the proposed DNN-AS scheme gets a better average bit error rate (ABER) performance than the typical AS schemes in SM-MIMO system with lower complexity. In contrast to the ED-AS approach, it attains the optimal and suboptimal ABER performance at low-and-moderate signal-to-noise ratio (SNR) region and high SNR region, respectively. The proposed DNN-AS scheme achieves obvious SNR gains compared with the conventional SM system and gets about 3.5 dB gains over the typical maximum-norm based AS (Norm-AS) algorithm. Furthermore, the proposed DNN-SD scheme obtains a superior detection performance compared with the conventional linear detection methods and provides the same ABER performance as the optimum MLD scheme in the presence of correlated noise.