Two-Stage Dilated Convolutional Neural Network-Based Detector for OFDM-IM

Abstract

As a key emerging green communication technology, signal detection based on deep learning can improve communication performance for orthogonal frequency division multiplexing with index modulation (OFDM-IM). However, it may lead to an increase in the bit error rate (BER) when the index and carrier are detected as a whole. To tackle this problem, a two-stage dilated convolutional neural network based on OFDM-IM (TS-DCNN-IM) is presented to improve signal detection performance in this paper. Through the two-stage design, the index and carrier can be processed separately by different subnetworks, thereby achieving better detection performance. In the first stage, an index subnetwork based on CNN is designed to obtain the index information of the carriers. Specifically, a dilated convolution module is introduced into the index subnetwork to better extract the carrier features, which is achieved by enlarging the receptive field without adding the network parameters. In the second stage, a deep neural network is constructed to predict the transmitted signal bits. Finally, the well-trained TS-DCNN-IM model is used to directly output the transmitted signal bits. Simulation results show that compared to the related algorithms, the TS-DCNN-IM algorithm can achieve better BER performance and higher computational efficiency.