We show that the real-valued system can be modeled as spectral signal convolution on graph, and propose graph convolution-enhanced expectation propagation (GCEPNet), a graph convolution-enhanced EP detector.
|-- EP_time # code for measuring CUDA enabled EP
|-- GEPNet #code for GEPNet
|-- SpectralGNN_k #code for GCEPNet
|-- classic_main.py #test traditional methods
|-- classic_solver.py #code for traditional methods
|-- data.py #generate MIMO data
|-- helper.py #utils functionsGCEPNet incorporates data-dependent attention scores into Chebyshev polynomial for powerful graph convolution with better generalization capacity.
Both GCEPNet and GEPNet introduce additional computations to EP, which originate from their GNN modules. Comparing with EP, GEPNet does not scale with the problem size as a result of the inefficient GNN aggregation, while GCEPNet effectively resolves the bottleneck with the newly proposed graph convolution.
Parts of the code are based on
MIT