A predictive multi-hop routing scheme for battery-free wireless sensor networks

Abstract

Energy-Harvesting Wireless Sensor Networks (EH-WSNs) promise a continuous everlasting network operation, as long as energy is obtained from the environment, but this advantage is at the expense of extremely unpredictable and often limited power budgets. The challenge is further exacerbated in EH-WSNs composed of battery-free nodes. In battery-free deployments, sensor nodes are not equipped with batteries and solely depend on environmental sources for energy; designing a routing scheme for such a network becomes complicated because in such networks, sensor nodes often switch between active and dead states as the energy obtained by the node varies with the harvesting conditions. Traditional shortest path routing algorithms are predominantly energy blind and repeatedly route traffic through a very limited number of geometrically optimal relays, leading to hotspot regions where nodes around that region continuously use up all their energy, and create routing holes, leaving numerous other nodes underutilized. On the other hand, existing energy-aware schemes prefer high-energy nodes, which have the tendency to create long, zig-zag paths and incur high latency, which is unacceptable for systems that demand minimal delay operations. This work presents a predictive multi-hop routing scheme for battery-free wireless sensor networks, which is based on a new predictive detour strategy. The main idea here is that energy can be considered as a time-varying but partially predictable resource. Each node in the network is capable of running a lightweight energy predictor based on an analytical model of the solar diurnal cycle and a smoothed estimate of the cloud attenuation, which predicts the future harvested energy of a node over a short time horizon. The estimated energy is added to the current residual energy of that node to create an effective future capacity, which is further incorporated into a next-hop cost function. This cost function considers the predicted residual energy and geometric progress in the direction of the sink, where neighbors who are geometrically closest to the sink but are estimated to be near energy depletion are penalized or excluded through a safety threshold, whereas energy-rich alternatives that maintain a reasonable forward progress to the sink node are preferred. This causes micro detours around nodes that have been continuously used, with little loss in path optimality...