Error analysis of Jacobi derivative estimators for noisy signals (Q639349): Difference between revisions

Numerical differentiation by integration computes the derivatives of a given function \(x\). A simple example is the Lanczos generalized derivative \[ D_Tx(t_0) = \frac{3}{2T}\, \int_{-1}^1 \tau\, x(t_0 + T \tau)\, d\tau = x'(t_0) + {\mathcal O}(T^2) \] with \(T>0\). In this paper, the authors study a parametric estimation technique introduced by \textit{M. Mboup, C. Join} and \textit{M. Fliess} [Numer. Algorithms 50, No.~4, 439--467 (2009; Zbl 1162.65009)], which estimates the \(n\)-th derivative by an integral of a given noisy signal \(y = x + \omega\), where Jacobi polynomials are used. Here \(\omega\) denotes a noise. These Jacobi derivative estimators contain a bias term error and a noise error. In order to reduce these errors, the authors extend the parameter domains used in the Jacobi derivative estimators. If \(\omega\) is a continuous parameter stochastic process with finite second moments and if \(x\) is sufficiently smooth, then the Jacobi derivative estimators exist and their statistical properties (mean value, variance and covariance) are analyzed. Finally, numerical experiments are presented.