Nonparametric asymptotically efficient estimation of a signal in the nonlinear case (Q792728)

The statistical problem of estimating an unknown function a(t) from the observation of the process \(y(t)=A S(t,a(t))+n(t)\) without prior information about a(t) is considered. Here, S(t,x) is a nonlinear function and n(t) is white Gaussian noise. A Cramér-Rao bound for estimating a(t) is obtained and two procedures are presented to produce nonparametric estimators on the basis of the likelihood equation in a function space. Consistency, asymptotic efficiency, asymptotic minimaxity and asymptotic sufficiency of the estimator are investigated. This article is a slight revision of the author's Ph D thesis in Mathematics at the Northeastern University under the supervision of Sam Gutmann.