This paper develops a control framework for a network of energy harvestingnodes connected to a Base Station (BS) over a multiple access channel. Theobjective is to adapt their transmission strategy to the state of the network,including the energy available to the individual nodes. In order to reduce thecomplexity of control, an optimization framework is proposed where energystorage dynamics are replaced by dynamic average power constraints induced bythe time correlated energy supply, thus enabling lightweight and flexiblenetwork control. Specifically, the BS adapts the packet transmissionprobability of the "active" nodes (those currently under a favorable energyharvesting state) so as to maximize the average long-term throughput, underthese dynamic average power constraints. The resulting policy takes the form ofthe packet transmission probability as a function of the energy harvestingstate and number of active nodes. The structure of the throughput-optimalgenie-aided policy, in which the number of active nodes is known non-causallyat the BS, is proved. Inspired by the genie-aided policy, a Bayesian estimationapproach is presented to address the case where the BS estimates the number ofactive nodes based on the observed network transmission pattern. It is shownthat the proposed scheme outperforms by 20% a scheme in which the nodes operatebased on local state information only, and performs well even when energystorage dynamics are taken into account.
展开▼
