ASA 129th Meeting - Washington, DC - 1995 May 30 .. Jun 06

2pSP10. Effect of ergodicity failure on detection performance prediction.

George E. Ioup

Dept. of Phys., Univ. of New Orleans, New Orleans, LA 70148

Lisa A. Pflug

Naval Res. Lab., Stennis Space Center, MS 39529

Juliette W. Ioup

Univ. of New Orleans, New Orleans, LA 70148

Formulas based on the noise variance and the first four signal moments to predict the performance at the minimum detectable level of ordinary and higher order correlation central ordinate threshold detectors have been derived. The passive prediction formulas have been reported and will be published. The active prediction results have also been presented. These prediction formulas agree remarkably well with computer-based Monte Carlo simulations. A breakdown in the agreement has been observed for simulations containing a small number of sample points---in the observation window for the passive case and in the signal for the active case. This breakdown can be traced to the inapplicability of the ergodicity assumption for various time averages over a small number of sample points, since key steps in the derivations contain the interchange of expectations and summations for nonlinear terms. Small sample number time averages lead to correlation central ordinate value probability density functions which are non-Gaussian. Prediction errors in the formulas as great as 2 dB are observed for simulations with 32 sample points. These errors decrease as the number of sample points is increased. Both passive and active examples are given. [Research supported by ONR/NRL-SSC.]