摘要： In the real-world, it is virtually impossible to have non-causal knowledge of future events. Research in energy harvesting (EH) systems that assumes knowledge of future energy arrivals falls short in terms of practical utility, pointing to the need for online strategies. In addition, the modeling and analysis for EH transmitter and receiver are inherently different. Compared with EH transmitter, EH receiver has received less attention. In this paper, we formulate Markov decision process problems and perform online optimization to maximize the number of bits decoded for an EH receiver with a time-switching architecture, which harvests energy from both a dedicated transmitter and other sources. We consider both infinite and finite horizon scenarios. For the infinite horizon, we provide an upper bound on the average expected reward. Then, we find an optimal policy which can achieve performance arbitrarily close to this bound. For the finite horizon, we first provide a policy obtained from standard backward induction with space quantization. Its performance can be close to optimal online performance as the number of quantization intervals increases, at the cost of relatively high computational complexity. Then, by carefully restricting the state space, we present a computationally efficient policy, which achieves comparatively good performance.