Special-purpose digital data-processing computers

Abstract

Motivated both by a renewed threat to national security and by challenging social and economic potentialities, the field of computation is currently expanding into hitherto unexploited regions. During the past decade attention has been focused on the development of large-scale computers and simulators whose main function has been computation per se. More recently, considerable energy has been diverted from these general-purpose machines and channeled toward the physical realization of special-purpose control and data-processing devices.A generalized real-time data-processing control system of the type to be discussed in this paper is shown in Figure 1. It is desired that a particular task to be accomplished be monitored and controlled to successful completion. Data Dl(t)…Dn(t), related to the progress or regress towards the desired goal is telemetered, perhaps from some distant point, to the processing computer, which, subject to instantaneous environmental conditions, Cl(t)….Cn(t), makes appropriate decisions and computations resulting in control responses Rl(t)….RP (t). Furthermore, the processor may be required to continuously survey some or all of its past responses so as to adjust its new decisions accordingly. The engineering problems presented by this type of system are perhaps as severe as any generally encountered in the computation field. Extremely stringent cost, weight and reliability requirements must be met. A variable process must be continually monitored and controlled to a high degree of over-all accuracy. In many applications the requirements are incompatible with present analog or digital techniques, and new techniques must be evolved. The general problem may be illustrated by recourse to a discussion of the engineering dilemmas arising from the instrumentation of a particular system. Our approach to their solutions will be offered. Figure 2 shows the control equations which must be simultaneously instrumented to perform the particular task. While these detailed relations are not of particular interest in themselves, it should be noted that numbers of integrations, multiplications, additions, subtractions, and function generations are required. The responses, R, are to assume the physical form of controlled shaft positions. The input telemetered data, D, are to be received as electrical signals, while the environmental conditions, C, are available as shaft positions. All signals may represent either positive or negative quantities. Note that R2(t) depends on the past history of R1(t). Because of the nature of the task to be accomplished, all input data and control responses must be continuous. The entire equipment must weigh less than 20 pounds and occupy 3/4 cubic foot. Assuming that the data, D, and shaft position, C1, have no intrinsic inaccuracies, while the conditions C2 and C3 are accurate to 1.0% but have maximum values considerably less than one, the processor must maintain an over-all accuracy of 0.1%. There are to be no calibration adjustments within the computer. These physical requirements, in total, are rather severe. The question arises as how best to meet them. The nature of the control relations might at first suggest that analog instrumentation be employed. Existing analog operational techniques might lend themselves nicely to the system. That is, a physical series of simulating components, interconnected in a straight- thru manner to achieve the continuous over-all function, might be employed. However, for obvious reasons, transmission of data in analog form is undesirable. While it is conceivable that at great cost and with painstaking care the required precision might be met, for extended service the analog type instrumentation is incompatible with the various physical requirements. On the other hand, if programmed digital machines are used and advantage is taken of the reliability of binary techniques, any desired degree of precision may be attained without recourse to calibration. Here, however, the price of more and more equipment must be paid. In general, such a machine would require large or at least heavy arithmetic, memory, and input-output organs. Instead of continuously operating on all variables simultaneously in an operational manner, the programmed-digital computer must successively sample, store, program, compute, sample in order to achieve the required functions. Indeed it is this very cyclic nature which negates the utility of the programmed digital computer in complex process control systems on which data cannot be appreciably delayed from input to response. The use of either existing analog or programmed-digital techniques having been precluded, our approach has been to develop a hybrid system. By taking advantage of operational-flow techniques characteristic of the analog approach, but by instrumenting the functions with binary digital circuitry, we have developed a third species capable of meeting the previously specified requirements. Such a hybrid system has been physically realized with novel developments related to information handling techniques, and new techniques of shaft-to-digital conversion, synchronization, multiplication, addition and timing, all making full use of magnetic, semiconducting and special electromechanical logical elements.