Integrating Geometry-Driven and Data-Driven Positioning via Combinatorial Data Augmentation

Due to the emergence of various wireless sensing technologies, numerous positioning algorithms have been introduced in the literature, categorized into \emph{geometry-driven positioning} (GP) and \emph{data-driven positioning} (DP). These approaches have complementary limitations, e.g., a non-line-of-sight issue for GP and the lack of a real-time training dataset for DP, calling for combining the two for practical use. To this end, this paper aims at introducing a new principle, called \emph{combinatorial data augmentation} (CDA), a catalyst for the two approaches' tight integration. Specifically, GP-based datasets augmented from different combinations of positioning entities can be used as DP's inputs and labels. We confirm the CDA's effectiveness from field experiments based on WiFi \emph{round-trip times} (RTTs) and \emph{inertial measurement units} (IMUs) by designing several CDA-driven positioning algorithms. First, we show that CDA enables us to quantify the given position estimates' uncertainties. Then, we can filter out unreliable ones for WiFi RTT positioning and compute the covariance matrix of a Kalman filter to integrate two position estimates derived by WiFi RTT and IMUs. The mean and standard deviations of the resultant positioning error are $1.65$ (m) and $1.01$ (m), respectively. Next, we use the above position estimate to its real-time label for \emph{fingerprint-based positioning} (FBP), shown to provide an acceptable positioning accuracy, say the average positioning error of $1.51$ (m) with a standard deviation of $0.88$ (m). Lastly, we discuss the proposed CDA's potential from positioning and beyond positioning perspectives.