Fine-grained Indoor Localization with Adaptively Sampled RF Fingerprints

Indoor localization is a supporting technology for a broadening range of pervasive wireless appli- cations. One promising approach is to locate users with radio frequency fingerprints. However, its wide adoption in real-world systems is challenged by the time- and manpower-consuming site survey process, which builds a fingerprint database a priori for localization. To address this problem, we propose an adaptive sampling approach that samples a small subset of reference points. Unlike current methods in this domain which are based on sampling and recovery of 2-D data, we model the fingerprint space as a 3-D tensor embedded in a low-dimensional tensor-column subspace using the recently developed algebraic framework for handling 3-D tensors [27, 30], which models a 3-D tensor as a matrix over a commutative ring, that is in turn constructed out of tensor-fibers. In this framework, the proposed scheme adaptively samples the 3-D data to identify reference points, which are highly informative for learning this low-dimensional tensor subspace. We prove that the proposed scheme achieves bounded recovery error and near-optimal sampling complexity. For an $N \times N \times N$ tensor with tensor tubal-rank $r$ [27], our scheme samples $\mathcal{O}(Nrlog^2 N)$ fingerprints compared to the information theoretically optimal sampling complexity of O(Nr). The approach is validated on data generated by the ray-tracing indoor model which accounts for the floor plan and the impact of walls. Simulation results show that, while maintaining the same localization accuracy of existing approaches, the amount of samples can be cut down by 71% for the high SNR case and 55% for the low SNR case.