Testing one-sided hypotheses for the mean of a Gaussian process (Q1181134)

The author considers the problem of testing one-sided hypotheses w.r.t. the real-valued parameter \(\theta\) in observation models of the form \(X(t)=S(\theta,t)+Z(t)\), \(0\leq t\leq T\), where \(S\) is a known non-random function and \(Z\) is a continuous zero-mean Gaussian process with known covariance function. Under smoothness and integrability conditions on \(S\) he derives a locally most powerful test which is asymptotically uniformly most powerful against local alternatives when \(T\) tends to infinity. As an illustration, several examples (sinusoidal signal transmission in white and coloured noise, linear stochastic differential systems) are studied.