Deep-Variational-Inference-Learning Detection for Cell-Free Massive MIMO With Quantization Error

A deep variational inference learning (DVIL) framework is proposed for data detection for cell-free massive multiple-input multiple-output (MIMO). The unknown model of the superimposed noise of quantization error and the environment noise is extracted based on the mixed Gaussian (MG) model, in order to make the proposed method have greater adaptability over variable scenarios. An iterative solution is obtained using VI. After that, the proposed algorithm is divided into two parts including the VI part and the trainable projected gradient (TPG) part. The TPG part is used to calculate the variable which has the highest complexity induced by matrix inversion. The numerical results show the merits of the proposed algorithm over the traditional algorithms.