Statistical evaluation of set-up errors in the ARC/SPAR system

The correlation coefficient, r, between the acoustic return signal and the reference signature of an underwater nuclear safeguard seal, is used to discriminate between intact and broken seals. Ideally, correlation coefficients for intact seals are distributed close to 1.0, while those for broken seals are distributed about 0, with no overlapping of the two distributions. One source of variation that affects these distributions and, therefore, the probabilities of making correct decisions with respect to tampering, is the use of different fixtures. Data consisting of 640 digitized amplitudes obtained with each one of ten fixture set-ups were used to investigate the effect of set-up error on correlations for intact seals. These measurements are slow and difficult to make, so ''bootstrap'' sampling was used to simulate the process. Compared to previous simulations of a single set-up, the standard deviation of the correlation coefficients for intact seals increased almost three-fold, due to variation among set-ups, while the minimum r decreased from 0.96 to 0.91. These changes do not significantly increase the probabilities of incorrect decisions. Bootstrapping has proved to be useful where measurements are difficult to obtain