Unsupervised Transfer NLOS Identification Model Based on Adversarial Discriminative Domain Adaptation in UWB Positioning

Non-line-of-sight (NLOS) propagation significantly impacts the performance of ultra-wideband (UWB) positioning. Identifying and mitigating or suppressing NLOS propagation is crucial for improving UWB positioning accuracy, with NLOS identification being a key component. Current NLOS identification methods limited to single scenarios, lacking adaptability to dynamic environmental changes and cross-scenario transferability. Existing transfer methods heavily rely on annotated data, increasing data annotation costs and implementation difficulty. Therefore, this letter proposes an unsupervised transfer NLOS identification model based on adversarial discriminative domain adaptation (ADDA) in UWB positioning: ADDA-NLOS. Experiment results show existing models have poor transferability in NLOS identification. ADDA-NLOS achieves high-performance unsupervised cross-scenario NLOS identification transferability and outperforms supervised models under limited data.