Machine Learning for Wideband Localization

Thang Van Nguyen, Youngmin Jeong, Hyundong Shin, Moe Z. Win

IEEE Journal on Selected Areas in Communications · 2015 · 146 citations · 64 references

Concepts

TL;DR

Wireless localization is critical for commercial, service, and military applications, yet indoor NLOS conditions severely limit accuracy, making NLOS identification and mitigation essential. The authors propose a robust, efficient ultrawide‑band TOA localization framework that uses relevance vector machine techniques to detect and mitigate NLOS signals and introduces a fast‑converging two‑step iterative algorithm, extended to centralized cooperative scenarios. The framework employs RVM‑based classification and regression within the TSI algorithm, implements a distributed cooperative scheme via variational Bayesian inference on factor graphs with refined Gaussian variational message passing to reduce complexity, and validates the approach using a stochastic localization network.

Abstract

Wireless localization has a great importance in a variety of areas including commercial, service, and military positioning and tracking systems. In harsh indoor environments, it is hard to localize an agent with high accuracy due to non-line-of-sight (NLOS) radio blockage or insufficient information from anchors. Therefore, NLOS identification and mitigation are highlighted as an effective way to improve the localization accuracy. In this paper, we develop a robust and efficient algorithm to enhance the accuracy for (ultrawide bandwidth) time-of-arrival localization through identifying and mitigating NLOS signals with relevance vector machine (RVM) techniques. We also propose a new localization algorithm, called the two-step iterative (TSI) algorithm, which converges fast with a finite number of iterations. To enhance the localization accuracy as well as expand the coverage of a localizable area, we continue to exploit the benefits of RVM in both classification and regression for cooperative localization by extending the TSI algorithm to a centralized cooperation case. For self-localization setting, we then develop a distributed cooperative algorithm based on variational Bayesian inference to simplify message representations on factor graphs and reduce communication overheads between agents. In particular, we build a refined version of Gaussian variational message passing to reduce the computational complexity while maintaining the localization accuracy. Finally, we introduce the notion of a stochastic localization network to verify proposed cooperative localization algorithms.

References

64