Optimization for data-driven wireless sensor scheduling

Modern large-scale distributed systems such as cyber-physical systems and the Internet of Things often consist of components that communicate/interact over shared networks of limited bandwidth and operate with minimal delay. One way to model this constraint is to assume that, at any time instant, only a single packet can be reliably transmitted over the network to its destination. In order to coordinate access to this limited communication resource, it is common to use some form of medium access control (MAC) scheme. The goal herein is to optimize the MAC to maximize performance. The design of scheduling policies for sensors making arbitrarily distributed random observations is considered. It is shown that the scheduler design can be expressed as a difference-of-convex functions optimization problem, which can be solved using the convex-concave procedure. Furthermore, this approach leads naturally to the application of data-driven approaches, where the design of scheduling policies where the joint probability density of the observations is unknown can be achieved by using an approximate sub-gradient method to solve an empirical risk minimization problem.