Transmit Antenna Selection Using CNN-Based Multiclass Classification with Linear Interpolation of Wideband Channels

Abstract

This study proposes a transmit antenna selection (TAS) method. The proposed TAS selects a transmit antenna based on the predicted channel quality by using a convolutional neural network (CNN)-based multi-class classification. The designed CNN directly determines the transmit antenna index based on the past signal-to-noise ratio (SNR), which is obtained through the received signals before the transmission. Since the channel states vary over time, the future SNRs are implicitly predicted through the CNN, and the predictive antenna index is explicitly determined. Here, the channels in the receiving and transmitting periods are symmetric, i.e., a time-division duplex (TDD) system is assumed. Further, various interpolation methods are examined to fill the missing received SNRs. Based on numerical results, it is verified that the proposed CNN-based TAS outperforms two conventional benchmarking methods: i) a TAS method based on the previous SNR and ii) a TAS method based on the average SNR.