Randomized incremental least squares for distributed estimation over sensor networks

This paper proposes a randomized incremental algorithm to distributedly compute the least square (LS) estimate of linear systems over sensor networks. By integrating its measurement information, a sensor is randomly activated at every time to incrementally update a diffusion vector, which is also used to recursively estimate the unknown parameters of the system via a temporal average algorithm. Then, the updated diffusion vector is passed to the next activated sensor. The activating process is modeled as an identically and independently distributed process. It is shown that the estimate in each sensor asymptotically converges both in mean and almost surely to the standard LS estimate of the system parameters, which is based on all the sensor information. Simulation is finally included to validate the theoretical results.