Optimum power allocation for sensor networks that perform object classification

This publication analyzes the power allocation problem for a distributed sensor network. We consider a network that may have power-limited sensor nodes and is used for target object classification. In the classification process, the absence, the presence, or the type of a target object is observed by the sensor nodes independently. Since the observations are noisy, and are thus unreliable, they are fused together as a reliable global decision in order to increase the overall classification probability. The global decision is performed at a remotely located fusion center, after combining the local observations. The combiner uses the best linear unbiased estimator in order to estimate the reflection coefficient of the present object accurately. By using the proposed system architecture, we are able to optimize the power allocation analytically in order to maximize the classification performance if the total power of the sensor network is limited. Two different cases of power constraints are discussed and compared with each other. The corresponding results are valid for additive white Gaussian channels as well as for frequency-flat slow-fading channels.