Application of complex-valued convolutional neural network for next generation wireless networks

A novel Complex-Valued Convolutional Neural Network (CV-CNN) model is proposed for the detection process in Multiple Input Multiple Output next generation wireless networks. The proposed model is used a as an auto-encoder based CV-CNN which is based on using the absolute value as the activation function and the L1-Norm as the loss function instead of the Euclidean norm. The proposed model is used for the joint detection of the index of the transmitted symbols and the index of the spatial channel matrix of the active antennas that are used for the high spectrum efficient Multi-Symbol Generalized Spatial Modulation systems. Simulation results show that the performance of the proposed CNN measured in terms of the Bit Error Rate (BER) is at most 0.5dB less than the performance of the Maximum Likelihood (ML) detector with the advantage of reducing the computational complexity at the receiver. The computational complexity is reduced by 14.71% which is achieved by the reduction of the search space at the output of the maximum pooling layer.