Robust Simultaneous Localization of Nodes and Targets in Sensor Networks Using Range-Only Measurements

TL;DR

This paper introduces robust algorithms for simultaneous localization and tracking in sensor networks that effectively handle both Gaussian noise and outliers, improving accuracy and computational efficiency.

Contribution

It develops a novel incremental Majorization-Minimization approach for SLAT that manages outliers and reduces computational load in time-recursive scenarios.

Findings

Algorithms outperform existing methods under outlier conditions.

Performance is comparable to existing methods with Gaussian noise.

Proposed methods improve accuracy and robustness in practical sensor network applications.

Abstract

Simultaneous localization and tracking (SLAT) in sensor networks aims to determine the positions of sensor nodes and a moving target in a network, given incomplete and inaccurate range measurements between the target and each of the sensors. One of the established methods for achieving this is to iteratively maximize a likelihood function (ML), which requires initialization with an approximate solution to avoid convergence towards local extrema. This paper develops methods for handling both Gaussian and Laplacian noise, the latter modeling the presence of outliers in some practical ranging systems that adversely affect the performance of localization algorithms designed for Gaussian noise. A modified Euclidean Distance Matrix (EDM) completion problem is solved for a block of target range measurements to approximately set up initial sensor/target positions, and the likelihood function is then iteratively refined through Majorization-Minimization (MM). To avoid the computational burden of repeatedly solving increasingly large EDM problems in time-recursive operation an incremental scheme is exploited whereby a new target/node position is estimated from previously available node/target locations to set up the iterative ML initial point for the full spatial configuration. The above methods are first derived under Gaussian noise assumptions, and modifications for Laplacian noise are then considered. Analytically, the main challenges to be overcome in the Laplacian case stem from the non-differentiability of $\ell_1$ norms that arise in the various cost functions. Simulation results confirm that the proposed algorithms significantly outperform existing methods for SLAT in the presence of outliers, while offering comparable performance for Gaussian noise.