Learning-Based Hybrid Beamforming Design for Full-Duplex Millimeter Wave Systems

Millimeter Wave (mmWave) communications with full-duplex (FD) have the potential of increasing spectral efficiency, relative to those half-duplex. However, residual self-interference (SI) from FD and high pathloss inherent mmWave signals may degrade system performance. Meanwhile, hybrid beamforming (HBF) is an efficient technology enhance channel gain mitigate interference reasonable complexity. conventional HBF approaches for systems are based on optimization processes, which either too complex or strongly rely quality state information (CSI). We propose two learning schemes design systems, i.e., extreme machine (ELM-HBF) convolutional neural networks (CNN-HBF). Specifically, we first alternating direction method multipliers (ADMM) algorithm achieve SI cancellation beamforming, then use a majorization-minimization (MM) joint transmitting receiving optimization. To train networks, simulate noisy channels as input, select beamformers calculated by proposed algorithms targets. Results show that both can provide more robust performance at least 22.1% higher efficiency compared orthogonal matching pursuit (OMP) algorithms. Besides, online prediction time almost 20 times faster than OMP scheme. Furthermore, training ELM-HBF about 600 CNN-HBF 64 antennas.