When RSSI encounters deep learning: An area localization scheme for pervasive sensing systems

Abstract

Localization has long been considered as a crucial research problem for pervasive sensing systems, especially with the arrival of big data era. Various techniques have been proposed to improve the localization accuracy by leveraging common wireless signals, such as radio signal strength indication (RSSI), collected from sensors placed in pervasive environments. However, the measured signal value can be easily affected by noise caused by physical obstacles in such sensing environment, which in turn compromises the localization performance. Hence, we present a novel RSSI-based area localization scheme using deep neural network (DNN) to explore the underlying correlation between the RSSI data and the respective sensor placement to achieve a superior localization performance. Moreover, to cope with the sensor data loss issue that commonly occurs during wireless sensor network (WSN) operation, an algorithm is designed to reconstruct the missing data for respective sensors in order to preserve the performance of DNN localization model. The effectiveness of the proposed scheme is verified with a real-world WSN testbed deployed inside an office building. The results demonstrate that the proposed scheme provides satisfactory prediction accuracy in area localization for pervasive sensing systems, regardless of the data loss issue that occurs with the respective sensors.