Deep Learning Based Decentralized Beamforming Methods for Multi-Antenna Interference Channels

This paper develops deep learning (DL) based beamforming approaches for multi-antenna interference channels where several base stations (BSs) individually optimize their own beamforming vectors in a decentralized manner. By exploiting the optimal beam structure, we propose an efficient method for beam decisions and coordination among BSs based solely on local information. Moreover, we show that the proposed approach allows a scalable design with respect to the number of users. We also present novel training strategies for the proposed deep neural networks, validating its potential as an innovative decentralized beamforming methodology. Consequently, the proposed DL based decentralized beamforming framework can achieve various optimal beamforming strategies. Numerical results demonstrate the advantages of the proposed framework over conventional methods.