ASA 125th Meeting Ottawa 1993 May

3aEA12. How much random data is enough?

John C. Burgess

Dept. of Mech. Eng., Univ. of Hawaii, 2540 Dole St., Honolulu, HI 96822

There is a relationship between the amount of data taken from a random signal and its statistical reliability. The usual assumptions are that the signal is stationary, white, and Gaussian. While procedures for analysis of a finite amount of random data after acquisition are well known, little information appears to be available about planning before data acquisition. With the usual assumptions, an upper limit can be found for the amount of random data required to meet a specified statistical reliability. This can be done for both time and frequency domain applications.