Reconfigurable Holographic Surface Beamforming via Unrolled Algorithm

Satellite communication network may plays a important role in the future sixth-generation (6G) wireless communication. Reconfigurable Holographic Surface (RHS) is a promising solution for satellite communication network. However, the RHS induces the problem of high computational complexity. To mitigate this problem, an unrolled gradient projection (GP) method is proposed in this paper. The unrolled GP method uses neural networks to train the step size in the traditional GP method, which significantly speed up the convergence rate. Extensive simulations show that the unrolled GP method runs faster and achieves a higher sum rate than the fractional programming (FP) and phased array methods.