On the Robustness of Deep Learning Based Beamforming for MU-MISO Systems
Ezequias de Santana Jr, Igor M. Guerreiro, Yuri C. B. Silva, Charles Casimiro Cavalcante

DOI: 10.14209/sbrt.2021.1570730492
Evento: XXXIX Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2021)
Keywords: Deep learning WMMSE MU-MISO Beamforming
Abstract
In this paper, we address the problem of robust beamforming in multi-user multiple-input single-output (MU-MISO) systems. To this end, we assess the performance of a deep neural network (DNN)-based beamforming strategy under imperfect channel state information (ICSI) conditions in a MU-MISO scenario. A robustness analysis is provided by adding a controlled additive error term in the fast fading component of the channel. Such an additive error is independent of the channel realization and does not change the statistics of the channel. The DNN-based beamforming strategy is compared with the classical weighted minimum mean square error (WMMSE) solution in terms of the impact of ICSI on the average sum rate. Numerical results show that ICSI affects in a similar way both the DNN-based strategy and the WMMSE solution. Thus, besides being an alternative method with less computation complexity due to its offline training phase, the DNN-based beamforming strategy can be said to be as robust as the WMMSE solution under ICSI.

Download