Evaluating Adaptive Prediction Filters for Efficient Data Gathering in Wireless Sensor Networks

Data gathering in wireless sensor networks is one of the essential tasks that has to be performed efficiently due to the sensors´ limited processing, storage, and communication capabilities. When sensor nodes continuously sense and wirelessly transmit raw sensor readings, predicting such readings might be a promising approach to save energy. This paper examines two adaptive prediction algorithms called as Least Mean Square and Recursive Least Square, integrated in a data gathering framework. A comprehensive simulation study of these algorithms assuming Gaussian processes shows that significant communication savings while guaranteeing a user-defined maximum error can be achieved. Especially, low processing costs and memory usage favor these algorithms for practical sensor node implementations. Finally, we prove the wide applicability of our data gathering framework investigating different kinds of real world sensor traces.