Temporal Convolutional Neural Network UWB Positioning Method Based on SimCLR-CIR-SC Autonomous Classification

Ultrawideband (UWB) technology has garnered considerable research interest for indoor positioning applications owing to its temporal resolution and strong signal penetration capabilities. However, the traditional classification of line-of-sight (LoS) or non-line-of-sight (NLoS) channel condition fails to characterize the complex indoor environments, and the existing models are trained on the entire dataset without considering the heterogeneity of the data, resulting in suboptimal positioning accuracy and reliability. To overcome these limitations, we propose an innovative autonomous classification method called SimCLR-CIR-SC that combines contrastive learning principles from the SimCLR architecture with spectral clustering (SC) techniques for enhanced feature extraction from channel impulse response (CIR) measurements. Building on the results of autonomous classification, we develop a temporal convolutional network with attention (TCN-A) architecture to discriminate between diverse channel state categories. For each identified channel condition, a dedicated TCN-A model is deployed to predict ranging errors, which subsequently perform distance calibration through error compensation and adaptive weight assignment in the weighted least-square (WLS) positioning algorithm. Experimental evaluations reveal that the proposed SimCLR-CIR-SC method achieves superior autonomous channel state classification and labeling performance compared to three clustering algorithms. Notably, the TCN-A classification model attains an accuracy of 98.16%, surpassing five existing classification models. Furthermore, the proposed positioning method achieves an average error of 0.57 m with three anchors, enhancing the positioning accuracy by at least 31.3% compared to four benchmark methods. Obviously, the positioning accuracy can be further improved as the number of anchors increases, and the average error is 0.278 m with seven anchors.