Boundary crossing probabilities for \((q,d)\)-Slepian-processes (Q312111)

The signal plus noise process \(Y_t=s(t)+ B_t\), \(t\in [0,d]\), where \(d>0\) is a known constant, \(s\in L^2[0,d]\) is a deterministic signal and the noise \(B\) is a standard Brownian motion, is often used in statistics as model for many different situations. In order to monitor the process \((s(t)+B_t)_{t\in [0,d]}\), it is convenient to consider the path of the process during a moving window \([t-q,t]\) of length \(q\) moving over time \(t\in [q,d]\), where \(0<q<d\) is fixed. Using this information of the windows over time \(t\in [q,d]\), a decision on a hypothesis of interest has to be made. Often it is of interest to test the null-hypothesis \[ H_0:s=c\text{ where }c\in\mathbb R\text{ is a known or unknown constant}. \] A simple statistic for each window is the difference \[ s(t)+B_t-s(t-q)-B_{t-q},\quad t\in [q,d], \] of the observations at the two boundary points of the interval \([t-q,t]\). Under \(H_0\) and by normalizing the variance to 1 one gets the stochastic process \[ W^{[q,d]}_t:=\frac{1}{\sqrt{q}}(B_t-B_{t-1})_{t\in [q,d]}. \] This process is at least useful to detect increasing or decreasing signal \(s\). The authors call this process \((q,d)\)-Slepian-process. They prove an analytical formula for the boundary crossing probability \[ {\mathrm P}\bigl(W^{[q,d]}_t>g(t)\text{ for some }t\in [q,d]\bigr), \] \(q<d<2q\), in the case that \(g(t)\) is a piecewise affine function. This formula can be used as approximation for the boundary crossing probability of an arbitrary boundary.