Simulation Code for: Residual-Aided CSI-free End-to-End Learning for Multiuser MIMO

Abstract

Complete simulation code for reproducing all results in the paper "Residual-Aided CSI-free End-to-End Learning for Multiuser MIMO" published in PLOS ONE (2026). Implements the Deep Unfolding Successive Over-Relaxation (DU-SOR) framework for CSI-free multi-user MIMO detection. Includes user-side encoders, base station decoder with sparse Graph Transformer, channel models (Rayleigh, Rician, 3GPP UMi, Kronecker), curriculum learning, MAML meta-learning, baseline comparisons (MMSE, DeepRx, OAMP-Net, GNN-Detector), and figure generation scripts. Requirements: Python 3.9+, PyTorch 2.1+. Usage: python run_experiments.py