Adaptive Threshold Learning for Robust First-Path Detection in UWB Ranging

Ultra-wideband (UWB) signals are widely used for indoor localization systems, yet its accuracy critically depends on reliably detecting the first path (FP) in the channel impulse response. Commodity receivers employ fixed or heuristic leading-edge thresholds, which often fail under multipath and NLoS conditions, introducing a First-Path Misprediction Degree (ΔFMD) that biases ToA/ToF estimates. This work presents a data-driven framework that adaptively predicts optimal LDE thresholds using PCA-compressed CIR features and gradient-boosted regression (XGBoost). Evaluations on public datasets show that the method reduces ΔFMDand improves ranging precision compared to statistical baselines and vendor heuristics, offering a robust alternative for indoor environments.