Jumps in observation noise statistics: effects on parameter estimates

In this paper the case of parameter estimation in the presence of jumps in the observation noise statistics is considered. These jumps may or may not be accompanied by jumps in parameter values. The estimation algorithm considered here is an extension (proposed earlier by the authors) of the Kalman filter in which statistical hypothesis testing with a three-state decision rule is used. It is pointed out that if the parameters undergo jumps, and the observation noise variance increases (decreases) at that time, then the probability of correct detection is increased (decreased). Results concerning changes in the expected value of the noise are also presented