Deep Unfolded Fractional Programming-Based Beamforming in RIS-Aided MISO Systems

Reconfigurable intelligent surface (RIS) is a promising solution to programmable wireless channels. However, most existing RIS phase optimization algorithms, relying on iterative processes, suffer from high latency and poor scalability. To address this issue, we propose a low-complexity scheme based on deep unfolding. Specifically, we consider a downlink MISO system aided by a RIS and aim to maximize users' weighted sumrate (WSR). We employ Lagrangian dual transformation to decouple the original non-convex problem into two sub-problems: transmit beamforming optimization and phase shift design. Then, we introduce a block coordinate descent (BCD) method, which still relies on iterative updates and includes complex operations such as matrix inversion, leading to high computational complexity and latency. To achieve fast solutions, we further propose to unfold the BCD method's iterative process into layers of an interpretable neural network (NN) with a few trainable parameters. The NN is trained offline and deployed online for realtime solutions. Finally, numerical results validate the performance of the proposed scheme in terms of comparable WSR and reduced computational complexity.