On-line learning and prediction of link quality in wireless sensor networks

Communication between sensor nodes in a Wireless Sensor Network (WSN) faces energy and bandwidth constraints. The dynamic behavior over time of the wireless channels makes ephemeral the neighborhood relation between sensors. Link quality estimation is critical for many WSN applications because it drastically influences the success of transmissions. In this paper we resort to machine learning methods to predict the short term evolution of link quality, in order to switch the data transmission on a better quality link. The problem is modeled as a game of prediction based on experts advice, using the Link-Quality Indicator (LQI) metric. A decision maker, called forecaster, advised by several experts, predicts the LQI values. The forecaster is able to learn how to adapt its strategy to predict values as close as possible to real LQI values. The proposed learning and prediction model presents a great flexibility: it is a general model that can be easily adapted to different link-quality metrics or prediction methods.