Role of Deep Learning in Wireless Communications

Traditional communication system design has always been based on the paradigm of first establishing a mathematical model channel, then designing and optimizing according to model. The advent modern machine learning techniques, specifically deep neural networks, opened up opportunities for data-driven optimization. This article draws examples from optimization reconfigurable intelligent surface, distributed channel estimation feedback multiuser beamforming, active sensing millimeter wave initial alignment illustrate that bypasses explicit modeling can often discover excellent solutions problems are otherwise computationally difficult solve. We show by performing an end-to-end training network using large number samples, learning-based approach potentially provide significantly system-level improvements as compared traditional model-based solving problems. key successful applications techniques in choosing appropriate architecture match underlying problem structure.