Target parameter estimation with heterogeneous sensor networks

A distributed network of sensors leverages its performance by aggregating information gathered by individual sensors. This process is referred to as sensor fusion. The primary goal of sensor fusion is to process and progressively refine information from multiple sensors to eventually create situation awareness (SA). Sensor fusion requires sensors to exchange data and information over a network with limited capacity. Consequently, fusing raw sensor data may not be desirable, since transporting large amounts of data from sensor nodes will consume a lot of bandwidth, which is in short supply in a typical wireless sensor network. More importantly, when a network has heterogeneous sensors with widely varying signal characteristics, it may not even be possible to mix sensor data from different sensor modalities. To minimize network bandwidth requirements and to deal with the multiple sensing modalities, we propose an alternative approach in which sensors compute their individual estimates, which are then sent to a fusion center that generates global estimates by optimally aggregating individual estimates. Since sensors do not have any prior knowledge about the value of the parameters to be estimated, each sensor independently computes its maximum likelihood estimates of the unknown parameters, based on the limited sensed data gathered in its local vicinity. These estimates along with their distributions are then communicated to a fusion center. The global estimator computes the final maximum likelihood estimates by maximizing a new likelihood function using the distribution of the individual estimates provide by different sensors. Since the distribution of an estimate typically requires a small number of parameters or moments, the amount of data that a sensor needs to communicate over the network is significantly reduced. The performance of the global estimator is evaluated by computing the Cramer-Rao lower bound of the variance of the estimates.