Cost-efficient Beam Alignment in mmWave Communications enlightened by Image Restoration

Given the importance of millimeter wave (mmWave) communication in 5G and beyond (B5G) wireless technology, its unique features—such as narrow beam characteristics, high frequency, and substantial bandwidth—are extensively leveraged to achieve the high speed, low latency, and high connection density required by 5G networks. Beamforming techniques are employed to mitigate severe propagation losses in mmWave frequency bands through highly directional transmissions; however, these techniques also increase the cost and complexity of the beam alignment process. To address this challenge, this paper formulates the optimal beam pairs prediction task in massive MIMO networks as an image restoration problem and proposes an cost-efficient Convolutional Autoencoder (CAE) based beam alignment algorithm. Simulation results demonstrate that the proposed method achieves beam alignment accuracy comparable to exhaustive beam searching schemes while significantly reducing overhead. Furthermore, the proposed methodology strikes a satisfactory balance between computational complexity and prediction accuracy when compared to previous AI-based beam alignment methods.