Sketch-based data recovery in sensor data streams

Abstract

The data imprecision received at a base station is common in mobile wireless sensor networks. In scenarios, data cleaning based on spatio-temporal relationships among sensors is not practical due to the unique, but commonly found, characteristics of sensor networks. As one of the first methods to clean sensor data in such environments, our proposed method deploys a sketch technique to periodically summarize N sensor samples into a fixed size array of memory and manage to recover values of missing or corrupted sensor samples at the base station. Our evaluation demonstrates that, with a small fixed portion of additional data transmission compared to original N data, the proposed method outperforms the existing data cleaning methods which assume the spatio-temporal relationship among sensors.