A general likelihood function decomposition that is linear in target state

Abstract

Likelihood function decomposition is a technique to coordinate deployed fields of multiple diverse heterogeneous sensors and for the automated processing of large volumes of multisensor data. It is an innovative new concept that is potentially useful in many of the kinds of nonlinear problems that arise in sensor fields used for detection, classification, and localization. Algorithms derived via the likelihood decompositionmethod are of interest because they have linear computational complexity in many of the parameters in distributed networked sensors — the number targets, the number of measurements, and the number of sensors. This efficiency is complemented with the ease with which the decompositions can be adapted to important application requirements such as land mass avoidance and ID/classification tags. The decomposition method also provides a natural way to exploit the spatial diversity of a sensor field to enable estimation of the aspect dependent targets. Observed information matrices derived from the likelihood decompositions can be exploited to maintain control of the field. The likelihood function decomposition method also simplifies the unconditional data likelihood function, enabling it to be written as an integral that is independent of the dimension of the target state space. This greatly reduces the computational complexity of the clutter rejection problem