A Multi-Stage Machine Learning and Post-Processing Pipeline for Indoor Ranging and Localization Using Bluetooth 6.0 Channel Sounding

This paper presents a multi-stage hybrid system combining Bluetooth 6.0 Channel Sounding (BCS) with machine learning and a post-processing pipeline for high-precision indoor ranging and localiza tion. The BCS technology provides two complementary distance measurements: Round-Trip Timing (RTT) and Phase-Based Ranging (PBR).

We first characterize the behavior of these measurements across multiple experi mental scenarios, including Line-of-Sight (LOS) and Non Line-of-Sight (NLOS) conditions, both indoors and out doors, and in an anechoic chamber. A Kalman filter is then applied to reduce noise and detect outliers, fol lowed by an XGBoost regression model that predicts the real distance from the RTT and PBR features. A post processing consensus algorithm is designed to stabilize predictions and converge toward the real distance.