Optimal sensing using query arrival distributions

Abstract

We examine optimal strategies for sampling and querying a sensing system when energy and data freshness need to be balanced. This approach is useful for planning algorithms utilizing data from vehicular networks, for example. These algorithms may be robust to some data staleness and this robustness can be used to save energy. Our model relies on the statistical distribution of user queries depending on which we develop sensor sampling schedules while optimizing system cost. For Poisson arrivals of user queries, we develop an optimal data sampling strategy which samples the network at regular intervals. For hyper-exponential query inter arrivals, we discuss methods to find an optimal sampling strategy. We show that optimal strategies can be discovered using dynamic programming techniques but the process is highly computational. Due to this reason, we suggest suboptimal sampling strategies which are nearly as efficient as the optimal strategy. We carefully design the cost function for the sensing system such that it is truly representative of most platforms we want to optimize for. Our model is generic and can be used to model any system that aggregates information which is then queried in real-time by users.