Battery-Aware Selective Transmitters in Energy-Harvesting Sensor Networks: Optimal Solution and Stochastic Dual Approximation

Abstract:
Energy-harvesting devices alleviate the problem of sensor nodes being powered by finite-capacity batteries. Since harvested energy is limited and stochastic, development of energy-efficient schemes to operate the network is well motivated. We investigate a scenario where, at every time instant, sensors have to decide whether to transmit or discard a message based on: its importance, the transmission energy cost, and their battery level. First, we formulate the problem as a sequential decision optimization and rely on dynamic programming to solve it. Second, we use dual stochastic methods to develop a suboptimal scheme that entails much lower computational complexity. This scheme leverages a relation between the battery level and the Lagrange multiplier associated with a long-term energy-conservation constraint. This finding is general and can be used in setups different that the considered here.