CAREER: Scaling Laws and Measure-Matching in Sensor Networks

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

Sensor networks create new challenges in the understanding of communication and information processing. This research addresses two of the key issues: the scaling behavior of the networks, and the fact that sensor networks involve both sources and channels and hence generally perform both data compression and data transmission. The "scaling behavior" of the network denotes its key properties and characteristics as a function of the number of nodes. For low-cost, low-power, and dense scenarios such as sensor networks, this is the most relevant characterization, and it is theoretically important and successful since the exact analysis of network performance appears to be a very hard problem. By contrast, a scaling law focuses on how the performance depends on the number of nodes. Preliminary scaling-law results of the investigators suggest that qualitatively different laws and insights apply to large networks as compared to small networks. This is of fundamental importance to both research and education: different code constructions, algorithm designs, and signal processing techniques have to be developed, and a different intuition has to be taught to cover such networks. The second key issue of this research concerns the fact that the sensed world is often analog, requiring both data compression and data transmission. This communication problem is unsolved to date: The omnipresent separation of the coding strategy into a compression and a transmission stage can lead to dramatically suboptimal performance, implying that the well-established understanding of information as "bits" needs to be revised. This research investigates a new and different understanding of communication as "matching" the source to the channel. In a scaling-law sense, this can lead to spectacular gains, as preliminary results have shown.