Pre-Trained Hybrid-Damping Message Passing Detector for Multiple MIMO Scenarios

Among various multiple-input multiple-output (MIMO) detection algorithms, message passing algorithms (MPAs) have been widely considered to balance performance and complexity. Damping factors can greatly affect MPAs' convergence behavior and stability. Deep neural network (DNN) based detectors with learnable damping factors have shown improved convergence performance in MIMO systems but require retraining in different system configurations. In this paper, a pre-trained hybrid-damping (PHD) message passing detection (MPD) for multi-scenarios using multi-objective evolutionary algorithm (MOEA) is proposed, which only requires one pre-training step and can be adapted to multiple scenarios. Numerical results indicate that the proposed PHD scheme can provide better convergence and flexibility than unified damping (UD) and DNN-based ones. Furthermore, the proposed PHD scheme can also be applied to other MPAs.