Timely Target Tracking in Cognitive Radar Networks

We consider a scenario where a fusion center must decide which updates to receive during each update period in a communication-limited cognitive radar network. When each radar node in the network only is able to obtain noisy state measurements for a subset of the targets, this means that the fusion center may not receive updates on every target during each update period. If the set of nodes which are available to give updates in each update period is limited to the nodes with interesting updates, the problem is further constrained. The solution for the selection problem at the fusion center is then non-stationary in time, and is not well suited for sequential learning frameworks where rewards have high temporal correlation. The important parameters become the age of the most recent update for every track, and the measurement quality each node provides. We derive an Age of Information-inspired track sensitive metric to inform node selection in such a network and compare it against less-informed techniques such as a multi-armed-bandit and random selection.