Deep Learning-Based Adaptive Beamforming for Interference Cancellation in V2I Scenarios

The fifth-generation (5G) mobile networks rely on advanced smart antenna systems to achieve high accuracy and low latency. In vehicular-to-infrastructure (V2I) scenarios, adaptive beamforming methods play a critical role in enhancing network throughput. This paper proposes a deep learning-based adaptive beamforming technique using long short-term memory (LSTM) networks as beamformers to determine complex weights for the antenna array, thereby mitigating interference in multiuser environments. Unlike conventional minimum variance distortionless beamformers (MVDR) that require knowledge of the direction of arrival (DoA) for the desired signal, the proposed LSTM-based beamformer estimates the desired signal in the presence of interference and noise without DoA knowledge. The LSTM network is trained to predict the angles between user equipment (UE) and roadside units (RSU) using complex time series input data, resulting in a beamformed output. Simulation results demonstrate that the proposed LSTM-based beamforming approach achieves comparable performance in terms of throughput, making it a promising solution for interference cancellation in 5G V2I scenarios.