Abstract
In this paper, we propose a method of model reduction for SISO continuous systems via block-pulse functions.
A set of block-pulse functions is well known as an orthonormal system. But it is not complete in L2-space. In this paper, we clarify that it has a property which corresponds to the completeness, when the space is restricted to bounded, piecewise-continuous functions. And it is pointed out that multiple-integration in functional space is reduced to simple algebraic operations on block-pulse expansion coefficients.
Based on these facts, a method of model reduction is proposed. In first step, block-pulse expansion of input is calculated. And in second step, that of output is calculated. In third step, the parameters of reduced model are calculated via least square algorithm.
The main characteristics of this method are (1) simplicity of algorithm, (2) time-domain approximation, and (3) applicability of any polynomial type input.
