Design and experiment of centimeter UWB ranging error correction model based on BP neural network

In complex indoor environment, ultra wide band (UWB) ranging error can not be effectively compensated by conventional methods, which seriously restricts its positioning accuracy. Based on the analysis of the distribution characteristics of UWB ranging error in indoor environment, two different BP network error correction models with different structures are designed. Model BP1 inputs the ranging values of a single label and four base stations, and outputs four corresponding ranging errors; Model BP2 inputs the three-dimensional coordinates of a pair of labels and a base station, and outputs one corresponding ranging error. The network is trained with high precision total station measurement result as reference value, and the ranging and positioning accuracy before and after model correction are compared and analyzed. The results show that both models can effectively correct ranging errors and improve positioning accuracy. BP1 ranging and positioning accuracy are improved by 83.0%, 75.9%, and BP2 ranging and positioning accuracy is improved by 91.7%, 93.8% on average. BP2 can improve the ranging and positioning accuracy more effectively than BP1, and the positioning accuracy can be improved from decimeter level to centimeter level.