Accurate Multiclass NLOS Channels Identification in UWB Indoor Positioning System-Based Deep Neural Network

The accurate distinction between line-of-sight (LOS) and non-line-of-sight (NLOS) propagation channels is paramount for precise distance measurement within ultra-wideband (UWB) indoor localization systems. In complex and dynamic environments, such as those encountered in the indoor positioning of autonomous mobile robots or vehicles, UWB signal propagation is particularly susceptible to NLOS conditions. However, much of the existing literature focuses on binary LOS/NLOS classifications, often overlooking the complexities of real-world environments such as hard-NLOS and multipath conditions. Additionally, a dynamic adaptation model for diverse indoor environments is lacking. This omission impedes the accuracy of UWB localization applications, such as autonomous robotics, where precision is vital in complex indoor settings. This article presents a fast hybrid-lightweight deep neural network model for LOS and NLOS conditions identification named indoor NLOS/LOS detection deep neural network. The proposed model ensures high accuracy and stability of LOS and NLOS identification with low processing time. More importantly, all of this comes with no loss, which is unique in the field of study. A robust verification has been guaranteed through experimental validation of four established databases. Moreover, the proposed framework performance is benchmarked against the recent LOS and NLOS state-of-the-art identification methodologies. The experimental findings underpin the efficacy and robustness of the proposed model in delivering an accuracy of 99.9% within a processing time of one second, which is the ideal recognition outcome validated across multiple databases.