Offloading over Time-Varying Channels in Wireless Powered Networks

In this work, we consider a network of energy harvesting devices served by an access point with a single antenna that is used for both wireless power transfer (WPT) and information reception. The objective is to maximize the sum throughput of the network over a finite horizon when the channel state information is only available causally. The devices are energized by WPT for a certain duration, which is subject to optimization, and then, their tasks are offloaded to the AP until the end of the time horizon by employing optimal dynamic power allocation. First, a single client case is investigated by modeling the joint optimization problem as a dynamic programming problem. Based on the characteristic of the problem, we prove that a time dependent threshold type structure exist for the optimal WPT duration, and we obtain closed form solution to the dynamic power allocation in the offloading period. Then, we consider the case of multiple energy harvesting devices and propose both a centralized and a distributed policy to determine the optimal stopping time of WPT. Unlike the centralized policy, the distributed policy does not require the complete network state information, and the decision to stop WPT is given by the AP based on the limited feedback from the devices sent only when they observe that a threshold is exceeded. The optimal number of feedback needed to make the transition is determined by a policy-search reinforcement learning method. We demonstrate numerically that the distributed policy is able to achieve very good performance with respect to the centralized policy.