The complexity and sophistication of the real-time data processing problems encountered in computerbased weapon systems severely tax all aspects ofadvanced data processing technology. Typically, these systems impose requirements in reliability, availability, cost, performance, and growth that stress today's data processing technology.I Furthermore, data processing solutions are required for a wide range of system concepts and operational environments.2 Although proven conventional techniques such as pipelining and cache memory have significantly improved computer performance and advances in circuit technology have increased processor capacity, current and projected needs still represent significant challenges. In real-time systems such as those for ballistic missile defense, the data processing problem is dominated by the necessity of meeting port-to-port response times while achieving total system throughput requirements. Response times may be as low as a few milliseconds with throughput requirements ranging up to tens ofmillions of instructions per second. Frequently, the data processing system must also remain dormant for months, activate on minutes' notice, and operate unattended with ultrareliability for periods ranging from a few hours to several days. Distributed computing promises to satisfy the requirements of these systems by utilizing moderately priced, contemporary hardware in networks. To achieve this promise, research and development is being pursued in all aspects of real-time distributed computing technology.3'4 Many techniques have been proposed for achieving reliability through redundancy, detecting and recovering from errors, distributing and managing shared data, communicating reliably between processes, allocating and scheduling resources in the presence of failures and overloads, and achieving high throughput through architectural innovation. These techniques must be proven experimentally before they can be used in a real-time system. Individually and collectively they impose overheads that create problems in satisfying real-time requirements. As a result, solutions to the problems of real-time distributed computing must be proven in a realistic environment for the specific real-time application. Two approaches can be considered for the development, evaluation, and demonstration of candidate solutions to the problems of real-time distributed computing. The first uses dedicated hardware and software resources for each application. This approach provides highfidelity results, but the diversity of applications and candidate architectural, hardware, and software solutions makes it prohibitively expensive and time-consuming. The second approach uses a highly flexible testbed that enables rapid, realistic investigation of a variety of candidate solutions for the specific application. This approach, although initially more expensive, is now costeffective because of the availability of (I) low-cost, 16-bit microcomputers that can be used to develop flexible multi-microcomputer systems and (2) well-supported minicomputers that act as centralized coordination facilities for software development, interactive access, and experiment control and monitoring. This article describes a flexible distributed testbed that is being developed to support the development, analysis, test, evaluation, and validation of research in distributed computing for real-time applications. The testbed not only provides the resources for experimentally obtaining quantitative results, but also serves as a focal point for the research, integrating related research activities and providing a mechanism for technology transfer to associated research efforts.

A flexible distributed testbed for real-time applications

A coherent optical processor is described which can be used to detect and locate objects in the sky thatmove against a background of fixed objects. A succession of sky images is recorded by multiple exposure onphotographic film with the imaging system stabilized on the stars. Stars appear as points in the compositeimage, and moving objects create streaks. Space invariant pattern recognition is performed on this compositeimage by generating its two -dimensional Fourier transform (spatial frequency spectrum) and scanning it witha rotating spatial filter matched to the streak spectrum. The filter angle at maximum energy transmissiondetermines the streak direction and a second Fourier transform provides coordinate information. The designof the optical system is described and some initial experimental results are shown.IntroductionAn effective midcourse ballistic missile defense (BMD) system requires the ability to launch an intercep-tor toward an incoming threat at the earliest possible time in order to find and destroy it. A hominginterceptor must be capable of searching a large sky sector and detecting the presence of a threat whilediscriminating among numerous other objects. Although simple in concept, the search task requires an enor-mous amount of computing power and data storage capacity. The trend in sensor design indicates that improvedsensor capabilities for BMD systems can be expected to increase the data rate by several orders of magnitudeover that of present sensors. The ability to handle these high data rates is important for advanced BMDsystems, especially when size, space, and weight are limited, as they are aboard a homing interceptor.This paper presents some initial results obtained in a program funded by the U. S. Army's BMD AdvancedTechnology Center (BMDATC). The program objective is to evaluate the capabilities of optical data process-ing to serve as an on -board preprocessor for the large volume data outputs from mosaic optical sensor arrays.Two processing phases are being considered:

J. Tindall

Papers

Optical Signal Processor For Moving Target Indication