Abstract
It becomes often necessary to extract an adequate linear model out of the observed response data of an unknown system. The usual process for this is first to assume the model transfer function with a certain number of unknown parameters and then fix their numerical values so that the agreement of the model response with the observed one is optimum in a certain sense.
Even though the skill in the first step, the assumption of the model form, is often more crucial effect than the other on the quality of the model performance, very little to rationalize this step has been done yet so far.
In this report, a new method for this purpose is proposed which enables us to estimate quantitatively such things as the width of the significant frequency band, the equivalent lag order within the above frequency band and the approximate distribution of the poles of the observed system, All of these can bedone through a straightforward data reduction routine by a digital computer without resorting to any kind of human judgement, even though the scope of application of the method is limited to monotone or nearly monotone systems.