Deep Learning Supported Path Prediction and Channel Estimation for MIMO-OTFS System With High Delay Resolution

The orthogonal time frequency space (OTFS) is one promising approach for the future wireless system with high-mobility users. This paper proposed a channel estimation for a multiple-input multiple-output (MIMO) OTFS system with high delay resolution in high-mobility environment. Shifts of the path indices and path appearance/disappearance are studied in this work. Due to the high mobility of the user and high delay resolution, the studied system can be more sensitive to index shift of paths. Considering the fractional components in OTFS channel, only the significant CSI elements are processed with minimum performance loss. A Deep Learning supported 3-phase scheme is developed. An auto-encoder (AE) is first deployed for compressed channel features, followed by a recurrent neural network (RNN) based scheme that provides a rough channel prediction. The indices of significant elements in the predicted channel are then extracted using the decoder function of the AE process. Finally, the values of the significant elements are reconstructed via the Least Squares based method. Analysis is provided on erroneous path predictions, i.e., missing existing paths or detecting non-existent paths. Simulation results demonstrate that the proposed 3-phase scheme can outperform the existing channel prediction schemes with a much better accuracy and lower bit error rate in high-mobility use cases.