Multipath Fingerprinting: Exploiting Multipath for Accurate Indoor Localization

Localization of an emitter in rich multipath environments, indoors and outdoors, is a challenging problem. In such environments there is typically no line-of-sight propagation between the emitter and the receivers, and as a result the traditional localization techniques based on line-of-sight propagation between the emitter and the receivers are ill-suited. To cope with this problem, a new localization technique based on machine learning techniques has been developed. The basic premise of this technique, referred to as multipath fingerprinting, is that in rich multipath environments the characteristics of the multipath signals – the directions-of-arrival and the differential-time-delays – provide a unique "fingerprint" of the emitter's location. The fingerprint is extracted from the spatial-temporal covariance matrix of the signals received by the antenna array, and is founded on the lower dimensional subspace, known as the signal subspace, capturing the dominant multipath signals. The performance of this technique, as demonstrated by both simulated and real data, is highly superior to existing techniques – achieving 1 meter accuracy in typical indoor environments using only a single Access Point (AP).