Active Sensing for Markov Chain Tracking
Arpan Chattopadhyay, Urbashi Mitra · 2018
Motivated by the need for energy-efficient operation in distributed tracking systems, the problem of active sensing for tracking a Markov chain is considered. A collection of wireless sensors form a multi-hop network, and collaboratively estimates the process at each node via measurements made locally and the messages received from neighbouring nodes. The trade-off is between the number of active sensors and the mean-squared error in the estimates averaged over the sensors, in presence of computing and communication constraints. To this end, a low-complexity, distributed sensor subset selection algorithm for tracking a Markov chain is proposed. The algorithm artfully blends three key tools: Gibbs sampling for sensor subset selection, stochastic approximation to meet the sensor activation constraint, and Kalman-consensus filtering for process estimation. The algorithm is numerically evaluated against natural competing algorithms, and is observed to yield significant improvement in the error performance despite using limited side information.