UWB-Based Indoor Error Correction and Fingerprint Localization Across Diverse Environments: A Unified Framework

Ultra-Wideband (UWB) technology is widely used in indoor positioning systems (IPS) due to its high accuracy and low cost. However, in Non-Line-of-Sight (NLOS) environments, multipath effects make it challenging for traditional geometric localization methods to achieve sufficient accuracy. Although deep learning techniques can improve positioning performance under NLOS conditions, models often rely on environment-specific data, making it difficult to generalize to new scenarios. To address this problem, we propose a Convolutional Neural Network (CNN)-based transfer learning (TL) framework with UWB Channel Impulse Response (CIR) to optimize error correction and fingerprint localization models across diverse environments. Experimental results show that the TL model achieves an error correction accuracy of 80.81 mm with only 50 samples from the new environment and a fingerprint localization accuracy of 61.24 cm using only 10% data from the new environment. Compared to models trained from scratch, our TL approach improves error correction accuracy by 40.67% and fingerprint localization accuracy by 16.99%. Consequently, it achieves high accuracy in UWB ranging error correction and fingerprint localization with less data collection in challenging NLOS environments.