JOPCNN-based hybrid precoding/combining algorithm for millimeter wave communication systems

To enhance the wideband hybrid precoding/combining (HPC) performance, this paper presents an antenna selection-based joint optimization HPC algorithm, namely JOPCNN. In the proposed algorithm, a JOPCNN framework is constructed to obtain the optimal analog precoder and combiner. Specially, the JOPCNN includes antenna selection convolutional neural network(ASCNN) and hybrid precoding/combining convolutional neural network (HPCCNN) subnetworks. Firstly, to improve spectral efficiency(SE) while reducing hardware cost and complexity of the system, the purpose of the ASCNN is to get the best antenna arrays. Subsequently, by employing the analog phase matrix as a label, the HPCCNN subnetwork is able to output the best analog precoder and combiner while adhering to the constant modulus restriction. Meanwhile, in order to better extract the amplitude, phase, and carrier-related information of the channel, a 3-dimensional convolutional neural network(3DCNN) is introduced into the proposed JOPCNN model. Finally, the digital coder can be solved by employing the resulting analog coder. Simulation findings indicate that the suggested approach performs better in SE than comparable algorithms.