Deep Unfolding-Powered Analog Beamforming for In-Band Full-Duplex

Abstract

In-band full-duplex (IBFD) can double the spectral efficiency of wireless communications systems. However, its major drawback is the self-interference, which interferes with the desired signal at the terminal. Self-interference cancelation can be operated in antenna, analog, and digital domains. In the antenna domain, beamformers are designed to minimize the leaked signal entering the system. This paper proposes an efficient analog beamforming design to maximize the channel achievable rate of massive multiple-input multiple-output (MIMO) IBFD systems. Specifically, we propose a low-complexity deep unfolding technique that provides a significantly better achievable rate than the state-of-the-art to deal with the non-convex design objective and constraints. More concretely, a complexity reduction of 75% is achieved for some scenarios, and of 50% is reached for the rest.

Keywords

• Analog beamforming
• deep learning
• deep unfolding
• in-band full-duplex
• PGA
• POCS