Nonlinear filtering and optimal phase tracking (Q639248)

This monograph is a concise introduction to the theory of nonlinear filtering. It sets itself apart from the majority of existing books on this subject by being very mathematically rigorous and, on the other hand, by always keeping applications in mind. The Zakai equation and the Kushner equation are derived, but the mathematics is never unnecessarily complicated. The basic difficulties in nonlinear filtering are demonstrated in classical benchmark problems, and it is pointed out that even new mathematical tools will not be able to escape these problems. As the author states, there are three conceptual differences between this monograph and others. For one, most of the time the optimality criterion for filters is the ``mean time to lose lock'' (MTLL), rather than the usually considered ``conditional mean square estimation error'' (CMSEE). The second difference is that, as far as possible, everything is discretized in time by an Euler scheme, the results are derived for the discrete model, and then the continuous time case is obtained in the limit. This allows the author to use relatively elementary arguments, as he states ``measure theory plays practically no role in this approach''. And finally, the emphasis is usually on small observation noise, and analytic methods are used to derive the asymptotics. Chapters 1 and 2 introduce the necessary prerequisites on Brownian motion and diffusions, on their relation to partial differential equations and on numerical approximations to diffusions, which can be used to approximate the solutions of partial differential equations. It is remarkable that the introduction of Brownian motion is carefully motivated by physical considerations, and that the relation to white noise is clearly worked out. Chapter 3 introduces several optimality criteria for filters, among them the CMSEE. The Zakai equation and Kushner's equation are derived. But also the case of smoothing (that is, estimating the signal given information about its past and its future) is considered, and the partial differential equations for the smoothing CMSEE are derived. In Chapter 4, the low noise case is studied, and asymptotic results are derived. Chapters 5, 6 and 7 study the MTLL as optimality criterion. The MTLL of various filters is calculated, loss of lock in radar and synchronization is studied, and, finally, it is shown that the minimum noise energy estimator, a variational filter that leads to a stochastic Hamilton-Jacobi-Bellman equation, is the filter that minimizes the MTLL. This is then applied to calculate the minimal MTLL in a range of benchmark problems. Throughout the text there are many exercises. Everything is applied to practically relevant examples, and there are nearly 50 figures. The monograph can be used as the basis for a one-semester course on nonlinear filtering. It is also an excellent source for mathematical researchers interested in the applications of nonlinear filtering. For non-mathematicians, the book is interesting because it is written with applications in mind, and because the mathematics are kept as elementary as possible but as rigorous as necessary.