AdaLOS: A Domain Adaptive UWB NLOS Identification for Dynamic Settings

A major challenge to Ultra-Wideband (UWB) positioning, especially in indoor environments, is the prevalent presence of Non-Line-Of-Sight (NLOS) and multipath propagations, degrading localization performance. Moreover, indoor settings tend to constantly change, making it difficult to characterize NLOS signals each time these changes occur. Promising approaches have been proposed to identify NLOS signals before localization; however, their performance is limited to the specific environment where measurements were carried out and cannot be extended to new or different environments. This can be attributed to feature discrepancies causing domain shifts and distribution divergence, owing to differences in environments captured by the Channel Impulse Response (CIR) waveforms. In this paper, we propose a domain Adaptive NLOS (AdaLOS) identification framework for UWB positioning systems. First, adaptation knowledge is obtained to align the source and target domains’ CIRs to mitigate the domain shift problem via cross-domain mappings. The distribution gap of the CIRs is further reduced by minimizing the marginal and conditional distribution divergence between the two domains (environments) by employing the Maximum Mean Discrepancy (MMD) criterion. A joint optimization procedure is then formulated to minimize the domain shift, marginal, and conditional distribution differences between CIRs from the two domains simultaneously. After minimizing the domain difference, Transformed Representative Features (TRF) of the source and target domains are obtained to train a domain-invariant classifier. We perform extensive simulations with CIR information collected from four distinct indoor environments characterized by different relative permittivity, signal distortions, noise, etc. AdaLOS is effective in adapting to new, distinct environments and significantly outperforms state-of-the-art works for NLOS identification.