Statistically-efficient filtering in impulsive environments: Weighted myriad filters (Q1348857)

From MaRDI portal
scientific article
Language Label Description Also known as
English
Statistically-efficient filtering in impulsive environments: Weighted myriad filters
scientific article

    Statements

    Statistically-efficient filtering in impulsive environments: Weighted myriad filters (English)
    0 references
    0 references
    0 references
    15 December 2002
    0 references
    A very successful approach to robust statistics in filtering theory is the use of generalized maximum likelihood estimators, or \(M\)-estimators. Given a set of samples \(x_1,x_2,\dots,x_n\), an \(M\)-estimator of location is defined as the value of the parameter \(\beta\) that minimizes a sum of the form \(\sum^N_{i=1} \rho(x_i-\beta)\), where \(\rho\) is the associated cost function. The behaviour of the \(M\)-estimator is completely characterized by the shape of the cost function \(\rho\). The authors of this paper introduce a novel class of \(M\)-filters motivated by the need for a flexible filtering framework with high statistical efficiency in the distribution families which can appear in engineering practice (e.g. \(\alpha\)-stable). The foundation of the proposed new class, called myriad filters, lies in the definition of the sample myriad as an \(M\)-estimator derived from tunable cost functions of the form: \[ \rho(x)= \log[k^2+x^2],\tag{CF} \] where \(k\) is the tunable parameter. Section 2 of the paper presents a rationale for the choice of cost functions such as (CF) and a quick introduction to \(\alpha\)-stable distributions. Section 3 shows that, depending on the tunable parameter \(k\) in (CF), the sample myriad can show drastically different behaviours, ranging from highly resistant mode-type estimators to the familiar (e.g. Gaussian-efficient) sample average. Section 4 proves several fundamental properties of the myriad estimator and shows its optimality in practical impulsive models such as (symmetric) \(\alpha\)-stable and generated-\(t\) distributions. Section 5 introduces the weighted \(M\)-filters, which, when the cost function in (CF) is used, give rise to a powerful structure that inherits the flexibility and statistical efficiency of the sample myriad, called the weighted myriad. Section 6 derives a set of implicit equations that characterizes the optimal weighted \(M\)-filters for a general cost function, and develops a suboptimal methodology for filter tuning and design. Thus, the newly defined weighted myriad filters appear to be a flexible filtering framework that derives important robustness properties by simply varying a tuning parameter, ranging from highly robust modelike filter forms to simple and Gaussian-efficient linear FIR (Finite Impulse Response) filters.
    0 references
    weighted myriad filters
    0 references
    weighted median filters
    0 references
    impulsive noise
    0 references
    heavy tails
    0 references
    Cauchy distribution
    0 references
    phase-locked loop
    0 references
    \(M\)-estimators
    0 references
    weighted \(M\)-filters
    0 references
    \(\alpha\)-stable distributions
    0 references
    robust statistics
    0 references
    filtering theory
    0 references
    generalized maximum likelihood estimators
    0 references
    myriad filters
    0 references

    Identifiers