We propose RIDA, a novel robust information-driven data compression architecture for distributed wireless sensor networks. The key idea is to determine the data correlation among a group of sensors based on the data values to significantly improve compression performance rather than relying solely on spatial data correlation. A logical mapping approach assigns virtual indices to nodes based on the data content, which enables simple implementation of data transformation on resource-constrained nodes without any other information. We evaluate RIDA with both discrete cosine transform (DCT) and discrete wavelet transform (DWT) on publicly available real-world data sets. Our experiments show that 30% of energy and 80–95% of bandwidth can be saved for typical multihop data networks. Moreover, the original data can be retrieved after decompression with a low error of about 3%. In particular, for one state-of-the-art distributed data compression algorithm for sensor networks, we show that the compression ratio is doubled by using logical mapping while maintaining comparable mean square error. Furthermore, we also propose a mechanism to detect and classify missing or faulty nodes, showing accuracy and recall of 95% when half of the nodes in the network are missing or faulty.
Spatial data compression via adaptive dispersion clustering