This paper introduces a general method for a particle solution to optimal nonlinearudestimation in signal processing . We deal here with the class of discrete timeudMarkov processes, to which the estimation problems of RADAR/SONAR signaludprocessing belong.udThe main feature of particle resolution is that it generates a global picture ofudthe probability space and therefore provides all desirable estimators (maximumudlikelyhood, minimum variance, etc . . . ). Its algorithmic principle relies onuda dynamic version of Monte-Carlo principles and is independant of dynamicudcomplexity (in particular the nature of non-linearities) . It is on the number of noiseudvariables that the size of the number of particles depends, according to resolutionudaccuracy. Convergence is inconditionaly stable under simple hypotheses.udAs an example, two important non-linear problems which arise in RADAR/SONARudsignal processing are dealt with, using this method.
展开▼
