The performance of modern wireless communications systems depends critically on the quality of the available channel state information (CSI) at the transmitter and receiver. Several previous works have proposed concepts and algorithms that help maintain high quality CSI even in the presence of high mobility and channel aging, such as temporal prediction schemes that employ neural networks. However, it is still unclear which neural network-based scheme provides the best performance in terms of prediction quality, training complexity and practical feasibility. To investigate such a question, this work compare five neural networks networks in terms of prediction quality. For this prupose, the MLP, CNN, LSTM, GRU and the transformer model are compared to the Kalman filter to invesigate the ability of data-driven approaches in channel prediction.
This repository is the code from implemented in the manuscript "A Comparison of Neural Networks for Wireless Channel Prediction". It should be stressed that this repository is just an illustration of th epredictions. Due to Github's restriction of file sizes, the original data set could not be uploaded. This also caused some minor edits in the prediction scheme. This is just a concetpual demo of the outline of the data and how the predictions were conducted. The neural network parameters are exactly the same as in the original code, but due to the mentioned changes, the results are not identical.
The code inputs simulated data from the 3GPP standardized TDL-A model and outputs predicted channel data on a desired time horizon. The results are measured in mean squared error (MSE) and the code outputs the best obtained MSE for each sceanrio in a csv file once complete.