Efficient training for fingerprint based positioning using matrix completion

Fingerprint localization methods are extensively used in many location-aware applications in pervasive computing. In this paper, we propose a new framework in order to reduce the exhaustive calibration procedure during the training phase in fingerprint-based systems. In particular, we minimize the number of Received Signal Strength (RSS) fingerprints by sensing a subset of the available channels in a WLAN. We exploit the spatial correlation structure of the RSS fingerprints to reconstruct the signature map. The problem is formulated according to the recently introduced Matrix Completion (MC) framework, which provides a new paradigm for reconstructing low rank data matrices from a small number of randomly sampled entries. Analytical studies and simulations are provided to show the performance of the proposed technique in terms of reconstruction and location error.