Deep Neural Network Parameterization for Channel Estimation in MUSA Systems

Multi-user shared access is a non-orthogonal multiple access technique that has been considered as a potential solution for 5G and beyond wireless networks. However, its performance is affected by the propagation of the channel estimate error in the SIC algorithm in the multi-user detector. Deep neural networks can be used to improve the initial channel estimate and this paper analyzes how the adjustment of hyperparameters and randomness affect the performance of the proposed estimator.