Formulas are derived for the information content criterion K of a pair
of stochastically related signals whose joint probability density is
described by a two-dimensional normal distribution law. A technique fo
r determining closely correlated signal pairs allows the probability o
f error to be reduced in the discrimination. of two adjacent values of
a controllable parameter by a factor of six (or even more) orders of
magnitude. Techniques for rapid elimination of uninformative signal pa
irs and selection of the most informative signal pairs are described.