An Efficient Deep Network for Modulation Classification in Impaired MIMO-OFDM Systems

In this paper, an efficient automatic modulation classification for MIMO-OFDM signals is proposed for next generation wireless networks by exploiting cutting-edge deep learning techniques. Particularly, we design a deep network, namely OFDM modulation classification network (OMCNet), with asymmetric depthwise separable convolution, residual connection, and attention mechanism in a sophisticated design of processing blocks to reduce the overall complexity without sacrificing learning efficiency. Relying on the simulations, our deep network achieves over 92 for different delay spread models demonstrates the robustness of modulation classification under various channel impairments. Remarkably, compared with a baseline model, OMCNet reduces the network size by four times and computation cost by two times with the asymmetric depthwise separable while achieving a competitive accuracy thanks to residual connection and attention mechanism. © 2023 IEEE.