Joint Optimization of Source Coding and Power Allocation in Sensor Networks

In a sensor network, each sensor makes a local observation of some underlying physical phenomenon, and sends a quantized version of the observation to a central office via communication links. Since the sensors' observations are often partial and correlated, the network performance becomes a complicated and non-separable function of all individual data rates at each sensor. In this paper, we consider a joint optimization of source coding and power allocation in a sensor network. We model the sensor network from an information theoretical perspective, and propose a novel formulation for distributive source coding to characterize the tradeoff among source coding rates. The new formulation is capable of dealing with the case where the physical source is described by a vector of random variables. We further optimize the power allocation strategy among sensors. We show that the joint source coding and sensor power allocation problem can be solved optimally and efficiently via convex programming