A Light-weight Learning Framework for RIS-assisted Beamforming Design with Mobile Edge Computing

Reconfigurable intelligent surface (RIS) serves as a promising paradigm for improving the spectral efficiency of wireless networks. However, the ubiquitous RISs in the future networks cannot afford the heavy central computations of their passive beamforming at central computational unit. To alleviate the stress, a light-weight learning framework with mobile edge computing (MEC) is considered for beamforming design, where expensive computational cost of reflecting elements is offloaded to MEC node with the light-weighted learning network. Specifically, a light-weighted neural network is proposed for RIS’s phase shift predictions, where unsupervised learning and image-structured samples are devised for boosting network training. Simulation results show that the proposed method outperforms the conventional deep learning in terms of the signal-to-noise ratio while it also reduces the computational complexity compared to the existing mathematical iterative optimization methods.