System design issues in time-driven sensor networks based on stochastic sampling

Abstract

This paper considers the type of sensor networks devoted to continuous monitoring. Typically, these networks constitute the hardware support for many environmental applications, where some physical phenomenon must be regularly monitored over a certain region. In line with the usual random deployment of sensor networks, the distributed sense-and-transmit process can be viewed as an extension of a technique from computer graphics called stochastic sampling, which provides an alternative to the uniform deployment imposed by multidimensional sampling theory. In addition, this paper proposes extending the randomness of the sampling process to the time domain, in order to avoid the strong synchronization requirements that sensor nodes should satisfy if they had to generate and transmit their readings periodically. The focus of the paper is system-oriented. Thus, it is assumed that packet transmissions are governed by a previously proposed MAC protocol, which can efficiently support the random nature of data with hardly detriment to the quality of the reconstructed signal. In particular, the paper focuses on the case of exponential randomization, since it allows for analytical treatment. Then, under these assumptions, closed-form expressions that relate the most relevant system-design parameters (application requirements, deployment and network configuration parameters and system behavior) are derived, and the main design tradeoffs are highlighted.