Abstract
Stacked intelligent metasurfaces (SIMs) can enhance transmission efficiency in holographic MIMO (HMIMO) systems, but their multi-layer structure poses computational challenges. This paper proposes an unfolded gradient projection (GP) method, where the step sizes are trained by neural networks, leading to faster convergence for the SIM-assisted system. Simulation results demonstrate that the unfolded GP method achieves significant computational complexity reduction and faster convergence compared to conventional GP and alternating optimization benchmarks. Furthermore, the method exhibits robustness against channel estimation errors and strong scalability.
Funding Information
  • Scientific Education and Research Foundation (NTF21048)
  • Scientific Education and Research Foundation (2024A1515011541)

This publication has 15 references indexed in Scilit: