previous | contents | next

applications currently in the programming process at Goodyear involve sonar, electronic warfare and large scale data management systems. These will be reported as results are achieved.

The Fast Fourier Transform (FFT) is a basic operation in digital signal processing which is being widely used in the real-time processing of radar and sonar signals. The structure of the FFT algorithm is such that it can be segmented into many similar concurrent operations. Parallel implementation of the FFT can provide orders of magnitude speed increases over sequential computer execution times. The organization of STARAN lends itself to efficient manipulation of data in the FFT.

The Air Force supplied real radar data (on tapes) to GAC to be transformed by the STARAN system. A 512-point, 16-bit FFT was performed on this real data in 2.7 milliseconds using only two MDA arrays. A 1024-point transform on real input data could be performed in about 3.0 milliseconds using all four arrays available at GAC's STARAN evaluation and training facility. For comparison purposes, the following is a list of reported execution times for a 1024-point, real input, FFT:

Sensor data processing can be split into two major categories- signal processing and post-processing. Signal processing is the area of the system where operations such as the FFT are performed; post-processing involves the sorting and editing of the signal processor output data to determine tactical information such as whether a real target is in the coverage area and where the target is.

The job of sorting the spectral lines that result from the FFT operations is a formidable task, especially in a multi-sensor case. The trend has been for increasing the sensitivity of signal processing systems. The acoustic signal line sorting task that accompanies any increased sensitivity can be staggering. For instance, a 6 dB improvement in sensitivity, in a classified Navy sonar system, would result in increasing the target load by a factor of 16 and the computer processing load by a factor of 250 or more.

A digital sonar signal processing system, under development at the Naval Air Development Center (NADC), requires that subroutines operate on the target spectral lines (outputs from an FFT) and other input data to form outputs suitable for later use in classification algorithms. Since the system is a multi-sensor system, these subroutines must process a very large volume of data in real time. The content addressability feature of STARAN provides the potential for significant performance gains due to the requirement for many searches in these post-processing subroutines.

As a consequence of this potential improvement, NADC issued a contract to GAC to assess the comparative run times for the STARAN versus a large-scale conventional computer (the CDC 6600). NADC-developed algorithms for the most time consuming operations in the post-processor system were programmed on the STARAN computer. Real data was then processed on both the STARAN and, by NADC, on a CDC-6600.

The STARAN executed the programs, using the real data, 200 times faster than the CDC-6600.

A processing function used by several agencies for locating specific character strings (such as place names) in textual information, was developed for STARAN and tested on a sample data base. The same function was executed on a conventional computer (Sigma 5) for a timing comparison. The STARAN solution ran 100 times faster. This function is also applicable to nondefense applications such as patent,

Summary [Rudolph]

Although several manufacturers are developing associative processor equipment, the first version to be produced in a production configuration was introduced in May of 1972 by Goodyear Aerospace Corporation following FAA on-site tests in 1971 at Knoxville, Tennessee of a USAF-owned engineering model built and demonstrated by Goodyear in 1969.

The processor provides full content addressability and array arithmetic capability within "main frame" memory coupled with a unique capability for wide bandwidth (over 3000 megabits/sec for a 4-array STARAN) input-output data transfers to mass data stores. The associative programming language, APPLE, provides a flexible and convenient assembler for programming array arithmetic and search algorithms without the complex and costly indexing, nested loop and data manipulation constructions required in conventional computer programming.