A Data Augmentation based DNN Approach for Outage-Constrained Robust Beamforming

This paper studies the long-standing problem of outage-constrained robust downlink beamforming in multi-user multi-antenna wireless communications systems. State of the art solutions have very high computational complexity which poses a major challenge to meet the latency requirement in the future communications systems, e.g., the targeted 1 ms end-to-end latency in 5G. By transforming the robust beamforming problem into a deep learning problem, we propose a new unsupervised data augmentation based deep neural network (DNN) method to address the outage-constrained robust beamforming problem with uncertain channel state information at the transmitter. Simulation results demonstrate that our proposed data augmentation based DNN method for the robust beamforming problem is capable to satisfy the required outage probability, and more importantly, compared to the benchmark Bernstein-Type Inequality (BTI) method, it is less conservative, more power efficient and several orders of magnitude faster.