the computer once each frame. There are 15 32-bit storage registers at the telescope that are filled sequentially once each frame by a 32-bit shift register accepting serial command bits from the computer. The input to the computer arrives serially from the data link and is fed into a 22-bit shift register, which, when full, is dumped into a 22-bit storage register that can be parallel-accessed by the high-speed (24-microsecond) input/output buffer of the computer. Output from the computer is parallel through the high-speed input/output buffer to a 22-bit output storage register, parallel to a 22-bit output shift register, and serial into the data link. Because the computer has no "interrupt" capability, protection against loss of data-servicing sync is accomplished through "overflow" flip-flops, on both input and output, available as flags to the computer. The serial output-data rate from the computer to the mountain is 512 bits per second. The serial input-data rate to the computer from the mountain is 1024 bits per second. Each input and output frame requires one parity bit, 20 "phase select" bits, and an 11-bit "Barker Word" for frame sync. The input frame contains 714 data bits and 278 unused bits; the output frame contains 480 data bits (fig. 1). System Timing The system software must be compatible with hardware timing restrictions. The communications system is serial; therefore, the data can be distinguished only by its position in a frame. Because any given information bit can be received only once each frame, all realtime control must allow for one-second increments in both commands and monitors. During the frame, there are two shorter cycles that restrict the software. One of these short cycles is in the computer itself. The Raytheon 250 computer is a serial machine with magnetostrictive delay-line memory and registers. Its memory-line cycle time is 3 milliseconds; hence, interrupts are not feasible. The other short cycle is determined by the 22-bit input and output registers in the communication system at the computer terminal. They require servicing by the computer so that their serial stream of outgoing and incoming data remains unbroken. The 22-bit storage register plus the 22-bit shift register at 1024 bits per second allow time blocks, between servicing, of approximately 44 milliseconds. This time must be reduced by four computer machine cycles: two cycles to allow for lack of sync between the data cycle and the computer cycle and two cycles to allow the computer the time necessary to carry out all the housekeeping functions servicing the on-line input/output devices. Under these timing restrictions, the frame was broken up into 23 blocks of computer time, each block being 32 milliseconds long. Since the optimized "word time" of the computer is 24 microseconds, there were up to 1,500 operations possible during each block; and there were up to 34,500 operations possible for the on-line software during a one-second frame. (Only 6,000 on-line operations were needed to operate the system. Another 2,000 might have been added for on-line data reduction.) The two computer machine cycles involved in data sync were salvaged for on-line output to an asynchronous high-speed punch (fig. 2). Computer Interrupts The Raytheon 250 computer is a serial machine without interrupts; however, even if a computer with interrupt capability were used, there would be little advantage in servicing a serial data link. Experience in the system indicates that interrupts would be advantageous only with parallel data links and only when a well-defined hierarchy of priorities can be assigned. Computer Limitations Computer speed was not a limitation in this system. The primary computer limitation was memory capacity. The total of 9,000 22-bit words imposed a restriction against any major on-line data reduction. There are about 1,500 words occupied by essentially off-line debugging and hardware testing programs that could be put on tape outside the computer during on-line operation. If this were done, however, any on-line program filling those positions could not be software debugged; and any marginal system-hardware failure could not be checked without writing over part of the control program. This experience indicates that any modern computer used to control a system of this complexity should have a fast, multiheaded, peripheral disk memory to relieve its core of all programs and data not on immediate on-line call. One other important limitation of the Raytheon 250 computer was the requirement that all the software be written in machine language. This was necessary (1) because the system is primarily logical instead of arithmetic, (2) because the programs must be optimized for speed (a time factor of 128 is involved), and (3) because the memory capacity is small. This language limitation meant that the final, complete, software system required two years for an experienced programmer to design and write. A system of equivalent complexity might be written for a modern core machine with a good assembly language in only about 8 man-months of time. Data Link Limitations Although the one-second serial data frame requires some delay, received data or transmitted commands need not be delayed a full second. All information in this system can be more current if the data and the commands are scheduled properly in the frame so that receipt, computer response, and transmission follow sequentially. The layout of useful bits in this frame imposes the following minimum delays: 64-millisecond data generation/receipt, 44-millisecond computer response, 32-millisecond command transmission/actuation-resulting in minimum delay time of 140 milliseconds. Because the mechanical subsystems of the automated telescope are relatively slow, the communication delay is not a significant problem. A major problem, however, is posed by the fact that a full second must elapse between any given "data-response-command" sequence. The motor drives and the data acquisition are continuous mode devices, but they must be controlled through an incremental data link. In most cases, it is possible to stop a device short of the target or short of saturation. The solution for these cases is to read "on the fly," to extrapolate differences, and to make the "stop" decision that comes closest to the goal. The cases requiring precision target acquisition are solved by either deliberate sacrifice of time (as in the photoelectric finder) or by buffering through digital counter-registers that store a binary number of "steps" and cause a drive motor to tum until the number is counted down to zero by pulses from an incremental shaft encoder. Target Acquisition The computer can point the telescope to a desired star by commanding it to a position where the telescope coordinates, as measured by the 16-bit right ascension and declination shaft encoders, are equal to the tabulated coordinates of the star. However, the residual uncertainty in the position of the stellar image, due to telescope flexure, encodertelescope misalignment, least-significant-bit quantization error of the encoders, and sidereal clock error, is too large for most purposes. It is therefore necessary to provide a photoelectric star finder to enable the system to position the image more accurately. This instrument is called a "finder," rather than a "guider," because it does not control the position of the image during observation. Instead, the entire telescope beam is diverted first to the finder, which supplies the computer with image-position information for aiming the telescope. When this has been accomplished, the full telescope beam is allowed to enter the observing instrument. During the observation, the sidereal drive of the telescope maintains the image in its proper position. Since observations made with |