抄録
A new method of generating signals whose amplitude distributions are Gaussian and whose power spectra can be given arbitrarily is proposed. Gaussian distributed random numbers generated by adding uniformly distributed random numbers are placed in an inverse Fourier transform region where each number is considered to represent a frequency component of a random time signal. By varying random numbers, the inverse Fourier transform is repeated. By specifying the variance of the random numbers as a function of frequency, the power spectra of the time function can be given arbitrarily.
The Gaussian signal whose spectrum is limited in a certain band and whose autocorrelation function is triangle is considered as an example. When the ensemble averages are considered, their amplitude distributions are Gaussian, but in considering finite time length sample functions, their band widths cannot be made too narrow, because a sufficiently large number of independent sample points should be taken from one sample function.
The inverse Fourier transform is carried out by FFT algorithm in a computer and the obtained values are converted to analog signals by a D-A converter. To obtain smooth analog signals, an interporation method based on the sampling theorem equation is developed.
To generate long time signals, an interrupt method is used, where signal output and computation are performed in parallel. In this method the frequency of obtained signal is limited to a maximum by computation time. Up to 500Hz signals could be generated by the authors' experiments. A generating method of higher frequency signals is also devised.
Gaussian signals whose spectra are band-limited and whose autocorrelation functions are triangular shape are generated by a minicomputer. Their amplitude distributions, power spectra and autocorrelation functions are measured and they are shown to have sufficiency good qualities.
