A Bayesian Inference Approach for Location-Based Micro Motions using Radio Frequency Sensing

The Bayesian inference is leveraged for target tracking from radio signals collected from access points (APs). The density of the moving object as well as its distance from the transmitted wireless signal substantially affect the signal strength arriving at the receiving end. The target tracking objective is formulated as an inference problem, by which we show how the Bayesian framework can be exploited to infer the parameters of interest for a given physics model. The channel state information (CSI) is collected from wireless APs, on which the inference is carried out. We employ a non-linear forward physics model of propagation, where we differentially infer the location, the velocity, and the fractional area of the moving surfaces in 3D space, versus time. The optimization is conducted via a Levenberg–Marquardt algorithm with analytically derived Jacobian and prior. The proposed model easily scales for any given number of access points. Experiments are conducted on Cisco wireless 4800 Access Point series; operating at 5 GHz radio frequency, and the probabilistic results for position and effective surface area estimates are provided, as well as numerical results for the point spread function from the statistics of the surface location.