Optimal stochastic routing in low duty-cycled wireless sensor networks

We study a routing problem in wireless sensor networks where sensors are duty-cycled. When sensors alternate between on and off modes, delay encountered in packet delivery due to loss in connectivity can become a critical problem, and how to achieve delay-optimality is non-trivial. For instance, when sensors’ sleep schedules are uncoordinated, it is not immediately clear whether a sensor with data to transmit should wait for a particular neighbor (who may be on a short route) to become available/active before transmission, or simply transmit to an available/active neighbor to avoid waiting. To obtain some insight into this problem, in this paper we formulate it as an optimal stochastic routing problem, where the randomness in the system comes from random duty cycling, as well as the uncertainty in packet transmission due to channel variations. Similar framework has been used in prior work which results in optimal routing algorithms that are sample-path dependent, also referred to as opportunistic in some cases. We show such algorithms are no longer optimal when duty cycling is introduced. We first develop and analyze an optimal centralized stochastic routing algorithm for randomly duty-cycled wireless sensor networks, and then simplify the algorithm to work with only local information. We further develop a distributed algorithm utilizing only local sleep/wake information of neighbors. This algorithm is shown to perform better than some existing distributed opportunistic routing algorithms such as ExOR.