LSTM-Based Error Correction for Reducing UWB Measurement Errors

This paper proposes an Long Short-Term Memory (LSTM)-based error correction method for enhancing Ultra-Wideband (UWB) measurement accuracy and minimizing localization errors. Initially, a high-precision Real-Time Kinematics (RTK) technique is employed to localize pre-calibrated measurement points, with the first point selected as the reference point. Using the reference point as the origin, distances between the reference point and other points are calculated and considered as ground truth values. Subsequently, the UWB base station is placed at the origin, and UWB tags are sequentially placed on the measurement points to obtain the measured UWB distances between the points and the origin. It is crucial to maintain consistent vertical heights of the UWB base station and tags with respect to the ground for ensuring accuracy. The ground truth values are used as the output of the training dataset, while the corresponding measured UWB distances serve as the input values. The dataset is then employed to train an LSTM model, resulting in the desired model. Finally, a test dataset is acquired using the same procedure, and the trained LSTM model is utilized for prediction. Experimental results demonstrate that the proposed method significantly reduces UWB measurement errors and enhances both UWB measurement accuracy and localization precision.