Home > Research > Publications & Outputs > Adaptive hybrid compression for wireless sensor...

Associated organisational unit

Links

Text available via DOI:

View graph of relations

Adaptive hybrid compression for wireless sensor networks

Research output: Contribution to journalJournal article

Published
Close
Article number53
<mark>Journal publication date</mark>1/12/2015
<mark>Journal</mark>ACM Transactions on Sensor Networks
Issue number4
Volume11
Number of pages36
Publication statusPublished
Original languageEnglish

Abstract

Wireless Sensor Networks (WSNs) are often deployed to sample the desired environmental attributes and deliver the acquired samples to a central station, termed as the sink, for processing as needed by the application. Many applications stipulate high granularity and data accuracy that results in high data volumes. However, sensor nodes are battery powered, and sending the requested large amounts of data rapidly depletes their energy. Fortunately, environmental attributes (e.g., temperature, pressure) often exhibit spatial and temporal correlations. Moreover, a large class of applications such as scientific analysis and simulations tolerate high latency for sensor data collection. Hence, we exploit the spatiotemporal correlation of sensor readings while benefiting from possible data delivery latency tolerance to minimize the amount of data to be transported to the sink. Accordingly, we develop a fully distributed adaptive hybrid compression scheme that exploits both spatial and temporal data redundancies and fuses both temporal and spatial compression for maximal data compression with accuracy guarantees. We present two main contributions: (i) an adaptive modeling technique that allows frugal and maximized temporal compression on resource-constraint sensor nodes by exploiting the data collection latency, and (ii) a novel model-based hierarchical clustering technique that allows for maximized spatial compression resulting into a hybrid compression scheme. Compared to the existing spatiotemporal compression approaches, our approach is fully decentralized and the proposed clustering scheme is based on sensor data models rather than instantaneous sensor data values, which allows merging nearby nodes with similar models into large clusters over a longer period of time rather than specific time instances. The analysis for computation and message overheads, the analysis for theoretical compressibility, and simulations using real-world data demonstrate that our proposed scheme can provide significant communication/energy savings without sacrificing the accuracy of collected data. © 2015 ACM 1550-4859/2015/05-ART53 15.00.