ADAPTIVE THRESHOLD LOGIC FOR IDENTIFICATION OF SIMULATED RADAR TARGETS.
Abstract
The application of adaptive threshold logic is considered as a means of establishing simple data transforms that will permit the automatic classification of patterns whose signal properties are too complex to be amenable to explicit analytic description. In an attempt to study the learning characteristics of adaptive logic systems, monopulse radar signal returns of multiple targets were computer-simulated to generate different classes of signal patterns. Three levels of pattern complexity were examined. Sum- and difference-transformed data patterns were learned more quickly than the ratio transform for one- and two-target patterns. The ratio transform was unsuccessful at the second level of complexity, i.e., the introduction of patterns of three targets one-quarter beamwidth apart versus one target. Difference-transform data produced weight learning plots that exhibited marked oscillation in early trials before converging to a solution. The sum transform showed oscillation in the learning plot similar to the two-target cases, but of no longer duration. At an S/N of 2, sum and difference transforms were successful in discriminating a two-target, one-quarter beamwidth apart pattern from a five-target, one-quarter beamwidth apart pattern, but exhibited a distribution of weighting functions suggesting that only the first and last few intervals of the radar scan provided useful information. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1965
- Accession Number
- AD0477861
Entities
People
- A. Timothy Ewald
Organizations
- General Electric