Adaptive Signal Processing: Next Generation Solutions

5

PARTICLE FILTERING

Petar M. Djurić and Mónica F. Bugallo

Stony Brook University, Stony Brook, NY

Many problems in adaptive filtering are nonlinear and non-Gaussian. Of the many methods that have been proposed in the literature for solving such problems, particle filtering has become one of the most popular. In this chapter we provide the basics of particle filtering and review its most important implementations. In Section 5.1 we give a brief introduction to the area, and in Section 5.2 we motivate the use of particle filtering by examples from several disciplines in science and engineering. We then proceed with an introduction of the underlying idea of particle filtering and present a detailed explanation of its essential steps (Section 5.3). In Section 5.4 we address two important issues of particle filtering. Subsequently, in Section 5.5 we focus on some implementations of particle filtering and compare their performance on synthesized data. We continue by explaining the problem of estimating constant parameters with particle filtering (Section 5.6). This topic deserves special attention because constant parameters lack dynamics, which for particle filtering creates some serious problems. We can improve the accuracy of particle filtering by a method known as Rao–Blackwellization. It is basically a combination of Kalman and particle filtering, and is discussed in Section 5.7. Prediction and smoothing with particle filtering are described in Sections 5.8 and 5.9, respectively. ...