Error-Based Filtering and Adaptive Weighting for Real-Time DL-TDoA Localization

Downlink time-difference-of-arrival (DL-TDoA) localization using ultra-wideband (UWB) is well suited for supporting large-scale tagging systems, but its accuracy degrades severely under non-line-of-sight (NLOS) conditions. Existing approaches often rely on high-dimensional channel impulse response (CIR) data for NLOS mitigation, yet the associated processing overhead and latency limit their practicality for real-time applications. This paper proposes an error-based filtering and adaptive weighting strategy for DL-TDoA localization that operates without CIR data. The filtering step identifies a subset of potentially erroneous TDoAs by examining the inconsistency between Chan- and LS-based estimates and the most recent valid position, refining the measurement set by selectively removing measurements and recomputing positions to confirm reliable TDoAs. The weighting step adjusts each measurement’s contribution based on the ToF-based ranging error between anchor pairs, as well as the RSSI of the received signal. Experimental evaluations in representative indoor NLOS scenarios demonstrate that the proposed method outperforms both the conventional Chan+Taylor and CIR-Stat algorithms in localization accuracy, while maintaining a relatively low computational overhead, making it suitable for real-time applications.