We propose low-rank adaptation (LoRA) for machine learning-aided hybrid beamforming (HBF) in episodically dynamic millimeter-wave multiple-input multiple-output (MIMO) systems. This approach introduces low-rank trainable matrices and uses a small buffer with recent channel samples, making it ideal for real-time adjustments. Evaluated for a large MIMO HBF system across both an environment-specific channel using ray tracing and clustered delay line channel models, simulation results show that rank-2 LoRA achieves efficient retraining with only 6% of the original network's parameters and 128 samples, improving average achievable information rate (AIR) by over 45% compared to the pre-trained model in both scenarios. The method significantly outperforms transferlearning with full-model online fine-tuning and model-agnostic meta-learning with its "almost-no-inner-loop" variant.

We introduce a deep learning-based hybrid beamforming (HBF) strategy for millimeter-wave transmission systems, specifically addressing the challenges posed by phase noise of local oscillators. Our approach utilizes a deep neural network to optimize precoding and combining matrices based on channel state information. We incorporate the symbol index through an adaptive attention mechanism and employ a self-supervised learning approach with a phase-noise-aware loss function to mitigate the effects of phase noise. While primarily focused on phase noise, our method also accommodates other practical constraints, such as limited-resolution phase shifter and imperfect channel estimation. Simulation results demonstrate that our design outperforms traditional and deep-learning based HBF methods in terms of data rate both in scenarios impacted only by phase noise and compounded distortion scenarios including low-resolution phase shifters and channel estimation errors.