Volume 8 deals with the application of distribution-free tolerance regions to pattern recognition and passive detection and tracking using surface- scattered signals. Pattern recognition is needed to identify sonar signatures as to the type of target by which they are generated. Distribution-free methods are desirable since the applicable probability distributions are frequently unknown, and it is desirable to establish some upper bound on at least the expected false-alarm probability. The recognition method developed has actually been applied to the recognition of speech waveforms since these are more easily obtainable than sonar signatures yet possess some of the same characteristics. Signals reflected from the sea surface undergo distortion which limits their detectability and usability for tracking. Two propagation geometries are analysed. The first deals with the crosscorrelation of surface-reflected and direct transmission paths, and the second with the crosscorrelation of surface- scattered signals received at two different locations. This analysis assumes that the signal generated at the target and the background noise are both gaussian random variables. Three models of the scattering mechanism are proposed and two are analysed in detail. Results are presented for low pass signal spectra and are investigated as a function of bandwidth.
Rept. for 1 Jul 1968-30 Apr 1970
Prepared in cooperation with Yale Univ., New Haven, Conn. See also Volume 7, AD0876563.