A Multipath Separation Method for Network Localization via Tensor Decomposition

High-accuracy position awareness is essential to a variety of applications, including navigation, Internet-of-Things, and autonomous vehicles. Network localization is a promising positioning service provider with merits of wide coverage and low cost, while its precision degrades in complex multipath propagation environments due to mixed coherent signals. In this paper, we propose a tensor-based multipath estimation method for network localization, which fully explores the inherent structure in the measurements and the uniqueness of tensor factorization. Specifically, we represent the observation as a low-rank tensor and separate different propagation paths based on the sparsity in the spatio-delay domain via tensor factorization. Simulation results show that compared with conventional algorithms, the proposed tensor-based method effectively reduces the multipath effect and achieves the centimeter-level localization accuracy.