Resource Allocation in Energy-Harvesting Sensor Networks

Abstract

We consider an energy-harvesting sensor network, possibly a subnetwork of a larger system, with a central unit that acts as information sink, and also recharges periodically the sensors. At each stage, a system supervisor allows the central unit to manage a certain amount of energy, which is to be apportioned amongst the sensors for their sensing tasks. How should energy be apportioned? And how successful can the network be? We relate this sensing and optimization task to log-optimal portfolio theory, and find that the underlying model can give rise to two quite different regimes. One is characterized by unbounded information accumulation and a nontrivial optimal apportionment strategy, and the other by information stagnation and a best strategy of winner-take-all.