Abstract
It is known that the detection of a random signal s(t) in the presence of an additive background noise n(t) is performed on the basis of the cross-correlation a1 of an input signal x(t)=s(t)+n(t) and the output signal of a linear filter driven by x(t).
This paper presents a new method to determine an impulse response g(t) of the linear filter in such a signal detection system; that is, g(t) is matched to the auto-correlation function of s(t) under the condition that g(t) is orthogonal to the auto-correlation function of n(t). Therefore, a1 depends only on the power of s(t), but does not depend on n(t) even if the level of n(t) varies. As a result, by this method it is possible not only to detect the presence or absence of the random signal hurried in the background noise, but also to measure the power of the signal.
As an application of this method, the detection of knock in a gasoline engine is attempted. The level of an engine vibration under knock free conditions, which is regard as the background noise from the knock-detecting point of view, always varies according to the engine speed. The experiment shows that this method is applicable to detect the knock and to quantify the intensity of the knock under these variable running conditions of the engine.
