Ranging Data Completion for UWB Systems Based on Time-Varying Filter Empirical Mode Decomposition

During the process of trains entering stations and stopping at platforms, ranging data based on Ultra-Wideband (UWB) frequently suffers from severe loss. Traditional interpolation or filtering methods for data completion are inadequate in the terms of accuracy and run off-line. This paper proposes a real-time data completion method to recover train’s trajectory online, which integrates time-varying filter empirical mode decomposition (TVFEMD), encoder-decoder (ED) and long shortterm memory (LSTM) networks. First, TVFEMD is used to perform multi-scale noise suppression and component extraction for non-stationary signals, in order to effectively reduce multipath interference. Then, LSTM together with ED is applied to predict short-term trends and complement missing values in incomplete component sequences, enhancing interpretability and applicability. Experimental results show that, compared with traditional methods such as TVFEMD-LSTM and VMDLSTM, the proposed TVFEMD-ED-LSTM approach significantly improves positioning continuity and accuracy. Compared with the aforementioned algorithms, the proposed algorithm achieves a 32% and 35% improvement in RMSE.