Title: IMNet: A Learning Based Detector for Index Modulation Aided MIMO-OFDM Systems

Abstract: Index modulation (IM) brings the reduction of power consumption and complexity of the transmitter to classical multiple-input multiple-output orthogonal frequency division multiplexing (MIMO-OFDM) systems. However, due to the introduction of IM, the complexity of the detector at receiver is greatly increased. Furthermore, the detector also requires the channel state information at receiver, which leads to high system overhead. To tackle these challenges, in this paper, we introduce deep learning (DL) in designing a non-iterative detector. Specifically, based on the structural sparsity of the transmitted signal in IM aided MIMO-OFDM systems, we first formulate the detection process as a sparse reconstruction problem. Then, a DL based detector called IMNet, which combines two subnets with the traditional least square method, is designed to recover the transmitted signal. To the best of our knowledge, this is the first attempt that designs the DL based detector for IM aided systems. Finally, to verify the adaptability and robustness of IMNet, simulations are carried out with consideration of correlated MIMO channels. The simulation results demonstrate that the proposed IMNet outperforms existing algorithms in terms of bit error rate and computational complexity under various scenarios.