A random dynamical systems approach to filtering in large-scale networks

The paper studies the problem of filtering a discrete-time linear system observed by a network of sensors. The sensors share a common communication medium to the estimator and transmission is bit and power budgeted. Under the assumption of conditional Gaussianity of the signal process at the estimator (which may be ensured by observation packet acknowledgements), the conditional prediction error covariance of the optimum mean-squared error filter is shown to evolve according to a random dynamical system (RDS) on the space of non-negative definite matrices. Our RDS formalism does not depend on the particular medium access protocol (randomized) and, under a minimal distributed observability assumption, we show that the sequence of random conditional prediction error covariance matrices converges in distribution to a unique invariant distribution (independent of the initial filter state), i.e., the conditional error process is shown to be ergodic. Under broad assumptions on the medium access protocol, we show that the conditional error covariance sequence satisfies a Markov-Feller property, leading to an explicit characterization of the support of its invariant measure. The methodology adopted in this work is sufficiently general to envision this application to sample path analysis of more general hybrid or switched systems, where existing analysis is mostly moment-based.