A new method of data compression in multisensor estimation fusion

Consider the decentralized estimation of an unknown parameter by bandwidth constrained sensor network with a fusion center. Local sensors make observations which are linearly scaled versions of these parameters corrupted by additive noises. For each sensor, the probability distribution function of the noise is known. In this paper, we propose a new approach that converts the estimation fusion problem to the decision fusion problem. With the methods of decision fusion, we find optimal local sensor compress rules which compress sensor observations into bits. The fusion center combines the transmitted bits from all the local sensors to generate a final estimation of the unknown parameter. Numerical examples show the efficiency of the new method.