Signal delay estimation in the presence of corrupting parameters (Q918098)

Let \(\{X_ t\}\), \(t\in [0,T]\), be a process of the form \(X(t)=S(t- \theta)+\eta (t)\) where \(\eta\) is a Gaussian white noise with variance \(\epsilon\) and S belongs to \({\mathbb{L}}^ 2({\mathbb{R}})\). The author studies the estimation of \(\theta\) when only \[ X^{\epsilon}(t)=h_{\epsilon}^{-1}\int^{T}_{0}K((t- u)/h_{\epsilon})X(u)du \] is observed. An estimator of \(\theta\) is given by \[ {\hat \theta}_{\epsilon}=\arg \min_{\theta \in \Theta} \inf_{S\in \Sigma}\int^{\infty}_{-\infty}[s(t)- X^{\epsilon}(t+\theta)]^ 2dt. \] It is shown that if K satisfies a number of regularity conditions, \({\hat \theta}{}_{\epsilon}\) exists and tends to \(\theta_ 0\) when \[ \lim_{\epsilon \to 0}h_{\epsilon}\epsilon^{-1}=\infty \text{ and } \lim_{\epsilon \to 0}h_{\epsilon}^{4(1+\delta)}\epsilon^{-1}=0. \] The risks of the estimators are computed and their asymptotic normality is shown.