A Nonlinear Representation for Parametrically Similar Signals.
Abstract
The use of a finite linear orthogonal basis is a familiar technique for representing a family of signals which are quite similar, but representation of a large family of signals with subtle differences is prone to errors which make pattern sorting difficult. The representation of signals by high-dimensional bases is restricted to small families of signals by computer weight, speed, and storage considerations. The dilemma thus created may be avoided if advantage is taken of certain underlying structures often found in families of radar and pulse communication signals. For pattern sorting purposes, an essential feature of the representation space is that relative signal correlations be the same as those in the observation space. In this report, an iterative procedure is developed for mapping a portion of the surface of a high-dimensional sphere onto the whole surface of a sphere of lower dimensionality. The governing criterion for this mapping is the preservation of these relative signal correlations. The results of this mapping is a low dimensional configuration of vectors whose inner products closely match the correlations between pairs of signals in the observation space. The evolution of this iterative technique is discussed, and the final version of the process is used for mapping both simulated pulse data and actual radar waveforms.
Document Details
- Document Type
- Technical Report
- Publication Date
- May 01, 1970
- Accession Number
- AD0870619
Entities
People
- R. S. Bennett
Organizations
- Johns Hopkins University