Deep Learning-Driven Beamforming Vector Design and Data Detection in IRS-Enabled Communication

In this paper, we present two deep neural network (DNN) architectures for an intelligent reflecting surface (IRS)-enabled communication scenario. The first DNN is designed to obtain an optimal beamforming weight vector which maximizes the achievable rate over the IRS-aided network. The beamforming vector is designed based on a pilot-based channel estimation procedure, which requires only a small fraction of the available reflecting surfaces on the IRS to be active. The obtained vector is fed to the second DNN, which is designed for detecting information symbols from the transmitter. Our simulation study shows that the achievable rate obtained using the designed beamforming vector approaches the maximum achievable rate quickly, as the number of active elements increases. Further, we compute the computational complexities of the proposed architectures. The impact of the number of active elements and the beamforming vector on the performance of the detector is also studied, in terms of the bit error rate.