Improving UWB Ranging Accuracy via Multiple Network Model with Second Order Motion Prediction

Abstract

With the wide application of ultra-wideband (UWB) ranging technology in industry, production and aerospace, how to improve the accuracy of UWB ranging has become a research hotspot. If the UWB ranging data is treated as sequence data, it is possible to improve the ranging performance by sequence analysis, which has low computation complexity compared to direct UWB signal processing. However, as the UWB ranging data has its inherent properties, existing sequence analysis methods may not achieve good performance on UWB ranging data. In this paper, a two-path deep-learning framework to process, namely the Multiple Network model with Second order motion (MNS), is proposed the improve UWB ranging performance via ranging sequence analysis. The proposed method fuses the target motion prediction via Newton's laws of motion and error estimation via GRU, LSTM and Bayes. To evaluate our algorithm, we also proposed a method to collect both UWB ranging data and the accurate answer via laser ranging. The collected dataset DATA_TS, the proposed MNS algorithm, and the trained model are all open sourced to the community to help researchers for further research,please visit https://github.com/xiaojiuwotongxue/data-store.git. Experiments on DATA_TS shows our proposed method outperforms traditional regresion methods significantly.