Modern combat aircraft sensor systems such as synthetic aperture radar (SAR) produce highly detailed, information rich displays. The simulation of such displays for training has demanded ever increasing computational resources as well as data sources more detailed than normally available digital feature analysis data (DFAD). By focusing on the correct reproduction of the content of a radar display rather than on a detailed model of radar physics, a novel Digital Radar Land Mass Simulator (DRLMS) for training is briefly described. A prototype of the system reproduces realistic real-beam, Doppler beam sharpened (DBS), and SAR ground maps from readily available data sources.

This radar simulation technique depends upon highly detailed, modified phototexture databases which contain both dimensional and effective radar cross-section information for broad area clutter and specific radar targets. This paper discusses the application of artificial neural networks in generating such databases from readily available data sources including Project 2851 and commercial satellite data. The issues, differences and solution approaches necessary to generate databases from such disparate sources as overhead imagery, DFAD feature data and existing simulator visual system databases are examined.

The techniques discussed have broad applications to the low-cost simulation of imaging sensor displays including millimeter microwave (MMW) and forward looking infrared (FLIR). The approach also drastically reduces the computational needs for a DRLMS system. The prototype, capable of generating SAR maps, was hosted on a single Motorola 68040 processor in a Macintosh personal computer. A simulation of the APG-68 radar, including real beam, expanded and DBS modes, is targeted to run in real time on a single MIPS R-4400 microprocessor.