Enhanced Emitter Identification Using Scaled Conventional Pulse and Intrapulse Parameters
Abstract
Neural networks trained only on intrapulse (IP) parameters can yield a comparable level of emitter identification (EID) accuracy to what is currently achieved by experienced human analysts. We have extended this study using 314 collects from 42 emitters of the same model with the addition of three conventional parameters: pulse repetition interval, radio frequency, and pulse width. We examined the effects of pulse averaging using statistically derived distance measures. The two neural classifiers used outperformed the currently used matching algorithms. We find that using ail three conventional parameters increases accuracy from 71.9% to 93.5%. Training using single-pulse representation was computationally expensive and revealed no difference in accuracy gained over pulse-averaged collects. Overall classification accuracy of greater than 95% is achieved using Mahalanobis distance measure. These results collectively show that EID can be significantly improved with the combined use of scaled unconventional and conventional parameters.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 08, 1999
- Accession Number
- ADA367856
Entities
People
- David A. Stenger
- Geoffrey L. Barrows
- John C. Sciortino Jr.
- Shah-an Yang
- Vuayanand C. Kowtha
Organizations
- United States Naval Research Laboratory