This paper considers the problem of the numerical solution of continuous system equations when the excitation is white noise. Many signal generation models consist of a linear system excited by white noise, and in simulating the performance of these systems it is usually necessary to solve the system differential equations numerically. The simplest representation of white noise is to represent it by independent random samples at the discrete time instants used in the numerical process. However it is shown in this paper that this is not appropriate when the time step is further subdivided in the numerical integration process, and doing so can lead to significant errors in the solution. Some simple examples are included to illustrate the difficulties.
展开▼
