LIPS: Link Prediction as a Service for data aggregation applications

Abstract A central component of the design of wireless sensor networks is reliable and efficient transmission of data from source to destination. However, this remains a challenging problem because of the dynamic nature of wireless links, such as interference, diffusion, and path fading. When the link quality gets worse, packets will get lost even with retransmissions and acknowledgments when internal queues become full. For example, in a well-known study to monitor volcano behavior (Werner-Allen et al., 2006), the measured data yield of nodes ranges from 20 % to 80 % . To address this challenge brought by unreliable links, in this paper, we propose the idea of LIPS, or Link Prediction as a Service. Specifically, we argue that it is beneficial for applications to be aware of link layer variations, so that they can take into account the future link quality estimates based on past measurements. In particular, we present a novel state space based approach for the link quality prediction, and demonstrate that it is possible to integrate this model into operating system interfaces, so that higher layer data aggregation protocols can directly exploit these interfaces to improve their performance. Our intensive evaluation indicates the state space based approach is accurate, stable and lightweight comparing to other strategies such as the autoregressive model (Liu and Cerpa, 2010). We carry out experiments based on the commonly used sensor node hardware including both link layer and operating system measurements.