Survey-free indoor localization via Wi-Fi RTT and physics-informed GPR radio map reconstruction

Indoor localization has emerged as a critical enabler for next-generation location-based services and the evolving 6G ecosystem, particularly where satellite localization is unavailable. Among indoor technologies, Wi-Fi fingerprinting offers an attractive trade-off between deployment cost, infrastructure availability, and localization accuracy. Recent advances in IEEE 802.11mc have enabled the use of round-trip time (RTT) measurements, providing more stable and accurate observables than received signal strength (RSS). However, RTT fingerprinting still suffers from two major limitations: high overhead from exhaustive site surveys and the difficulty of maintaining dense radio maps in dynamic environments. This paper proposes a Wi-Fi RTT fingerprinting framework that minimizes offline calibration effort by restricting empirical measurements to a sparse set of reference locations associated with deployed access points (APs). A Gaussian Process Regression (GPR)-based reconstruction model synthetically generates dense virtual reference point fingerprints, while physical-signal fitting models improve RTT estimation during reconstruction. Unlike previous studies focused on RTT/RSS fusion, ranging compensation, or generic interpolation, this work systematically investigates GPR-based synthetic radio map reconstruction for Wi-Fi RTT fingerprinting, evaluating the impact of AP geometry, grid density, temporal averaging, and reconstruction models on localization performance. Experimental results from real-world RTT measurements show that the framework maintains sub-meter localization accuracy in deployments where the target area lies within the convex hull of the APs. Conversely, the observed RMSE is approximately 2 m in asymmetric extrapolation scenarios. Nonetheless, it significantly reduces the calibration complexity and signaling overhead associated with dense RTT fingerprint collection.