Scalable Data Collection in Sensor Networks

Abstract

Dense sensor deployments impose significant constraints on aggregate network data rate and resource utilization. Effective protocols for such data transfers rely on spatio-temporal correlations in sensor data for in-network data compression. The message complexity of these schemes is generally lower bounded by n, for a network with n sensors, since correlation is not collocated with sensing. Consequently, as the number of nodes and network density increase, these protocols become increasingly inefficient. We present here a novel protocol, called SNP, for fine-grained data collection, which requires approximately O(n − R) messages, where R, a measure of redundancy in sensed data generally increases with density. SNP uses spatio-temporal correlations to near-optimally compress data at the source, reducing network traffic and power consumption. We present a comprehensive information theoretic basis for SNP and establish its superior performance in comparison to existing approaches. We support our results with a comprehensive experimental evaluation of the performance of SNP in a real-world sensor network testbed.