Graph-based Simultaneous Localization and Bias Tracking for Robust Positioning in Obstructed LOS Situations

Abstract

This paper presents a neural-enhanced probabilistic model and corresponding factor graph-based sum-product algorithm for robust localization and tracking in multipathprone environments. The introduced hybrid probabilistic model consists of physics-based and data-driven measurement models capturing the information contained in both, the line-of-sight (LOS) component as well as in multipath components (NLOS components). The physics-based and data-driven models are embedded in a joint Bayesian framework allowing to derive from first principles a factor graph-based algorithm that fuses the information of these models. The proposed algorithm uses radio signal measurements from multiple base stations to robustly estimate the mobile agent's position together with all model parameters. It provides high localization accuracy by exploiting the position-related information of the LOS component via the physics-based model and robustness by exploiting the geometric imprint of multipath components independent of the propagation channel via the data-driven model. In a challenging numerical experiment involving obstructed LOS situations to all anchors, we show that the proposed sequential algorithm significantly outperforms state-of-the-art methods and attains the posterior Cramér-Rao lower bound even with training data limited to local regions.

A Neural-enhanced Factor Graph-based Algorithm for Robust Positioning in Obstructed LOS Situations

“…MP-SLAM provides high-accuracy position estimates even in OLOS situations. However, it requires specular, resolved MPCs caused by environments consistent with the VA model, i.e, flat surfaces of sufficient extend [21]- [23]. Other methods exploit cooperation among individual agents [3], [24], [25], or perform robust signal processing against multipath propagation and clutter measurements in general.…”

Offsite Evaluation of Localization Systems: Criteria, Systems, and Results From IPIN 2021 and 2022 Competitions

“…3) Team SPSC: Team SPSC used a two-step method similar to the algorithm presented in [50], [51]. First, a snapshot-based parametric channel estimation and detection algorithm extracts delays and corresponding amplitudes of multipath signal components out of the received base-band signal.…”

“…In the following, the probabilistic system model of Team SPSC's algorithm is briefly discussed. a) Channel Estimation and Detection Algorithm: The channel estimation and detection algorithm presented in the supplementary material of [51] was applied to the base-band signal vector at each time n and for each anchor j independently. It provides a measurement vector z are subject to data association uncertainty, i.e., it is not known which measurement originates from the LOS, from multipath or from a false-alarm.…”

“…Analyzing the results of Track 7 of the IPIN 2021 challenge their proposed algorithm compares well to the proposed machine-learning based approaches, consistently showing robust behaviour for all data sets and being only slightly outperformed in terms of accuracy. An important aspect of the methods presented in [50], [51] is the non-uniform NLOS likelihood, which allows the information contained in multipath to be fused in a soft, Bayesian manner [55], i.e., it constrains the PF-based position estimate using information contained in the NLOS measurements. However, when there is a LOS connection to at least two anchors and these anchors provide enough directional diversity for accurate positioning, the NLOS information has virtually no effect (especially when the mirror sources due to walls in the environment setup are far away).…”

“…Due to a large number of anchors, there were only partial obstructed LOS situations, i.e., the LOS to all anchors was never obstructed simultaneously over significant time intervals. Therefore, unlike [50], [51], the method described in Section IX-D.3 does not involve a non-uniform NLOS likelihood. The team assumed that machine learning-based methods that learn a nonlinear regression model that maps to the agent position p n can, in turn, exploit small differences in the imprint of the provided radio signal data, leading to the observed gain in estimation accuracy.…”