Kalman filtering with no A-priori information about noise-White noise case: Part II: Indentification of noise means

Two new algorithms are proposed for identifying the noise means required before Kalman Filtering. These algorithms use the results of Part I of this paper, i.e., the identified noise intensities (or the optimal predictor gain if the noise intensities cannot be identified). To be Compatible with the algorithms of Part I, the algorithms described here are also nonrecursive. They are based on the maximum-likelihood approach. The cost function resulting from this approach is considerably simplified before deriving the algorithms, which affords a great insight into the problem at hand. The question of uniqueness of bias estimates is considered in the light of the invertibility of dynamic systems. An example is included to partially illustrate the algorithms. A significant advantage of our approach is the ability of suboptimally identifying the noise means even when the noise intensities cannot be uniquely determined.