BISPECTRAL ANALYSIS OF RADAR SIGNALS WITH APPLICATION TO TARGET CLASSIFICATION

Ismail  I. JOUNY

This dissertation has investigated the use of bispectral processing methods for radar backscatter processing and for radar target identification. Bispectral processing methods were adapted to the radar signature analysis problem, and resulted in the so-called birange profile. The state of the art in time‑domain radar signal processing was briefly described in Chapter II. Research on time‑domain signature analysis is focused on the extraction of the major scattering components from the impulse response of a radar target.

In Chapter III we studied the birange profile as a tool for detecting multiple interactions. A definition for the bispectrum of complex radar signals was introduced. Also, a birange estimation technique based on the "indirect bispectral estimation method" with slight modifications and suitable for radar signals was proposed and used in this dissertation. An interpretation of the birange based on a specified scattering model, which is basically a sum of complex exponential with implicit couplings, is presented. Examples on the birange of hypothetical targets were also given. The concept of birange processing was tested using experimental measurements of scattering from a tilted flat plate and a generic aircraft model with removable parts, then using real aircraft models such as Boeing 707, 727,747,DC 10, and the Concord. Finally, a feature extraction algorithm based on Gaussian fitting was used to locate the interactions in the birange profiles.

A birange profile estimation algorithm based on AR modeling of radar signatures was introduced in Chapter IV. This algorithm was tested using both synthesized scattering data and real data. The Fourier‑based birange profile of the synthesized target was reproduced using this algorithm, but the Fourier‑based birange profiles of real data were slightly incompatible with those obtained using this algorithm.

Chapter V considered the effectiveness of birange profiles for target classification. Tests were conducted using non‑Gaussian noise, azimuth ambiguity, and extraneous scatterers. In general, classification results using the birange profile only were inferior to those using the impulse response or nearest neighbor classifiers. There were some cases where the birange classifier outperformed other methods. In the first case the additive noise was non‑Gaussian and in the second case the responses of three extraneous point scatterers were added to the measured target frequency response.

The birange as a time‑domain profile is not an alternative to the impulse response but an augmentative display of target signatures. If the additional knowledge, provided by the birange, is weighted properly, then it will have positive impact on the radar target recognition problem. One may argue that, because both the transient response and the birange profile are derived from the same data source, then neither one will enhance the identification process beyond what is already achieved using the data prior to any signal processing. The validity of this argument diminishes when the parameters of the underlying probability density function are not known a priori. Therefore, features extracted from the birange profile will potentially enhance the identification process under special circumstances.