When the 250 ft Mk I radio-telescope was commissioned in 1957 at Jodrell Bank, it was controlled by an electro-mechanical analogue computer which, when given a radio source’s coordinates set up on the control desk, the sidereal time and the telescope’s azimuth and elevation positions, was able to servo its position so as to track the radio source across the sky.
In 1964, the second Jodrell Bank telescope, the Mk II, was completed and became the first telescope of any type to be controlled by a digital computer: the Ferranti Argus 100. Whilst small American computers at this time had word lengths of 12 or 16-bits, the Argus had a 24-bit word length. This was perfect for this application, as the encoders measuring the telescope’s position were typically 18-bits in precision; the precise computations required to determine the desired position to observe a radio source could be carried out using double length (48-bit) floating point calculations.
Building the hybrid computer
Two years later, for my PhD project, I was charged with designing and building a ‘hybrid’ analogue/digital computer to acquire the data from a lunar radar experiment. Not surprisingly, the digital computer employed the Argus instruction set and word length, but it also had several enhancements: a fast multiply, a division instruction and, most notably, a ‘correlate’ function to measure the similarity of two 24-bit words. This sped up the ‘inner loop’ of the data acquisition program by a significant amount.
Not long after, two Argus 400 computers were acquired. The first replaced the Mk II’s Argus 100, to both control the telescope and acquire its data, whilst the second was used to acquire the data from the Mk I telescope. The Argus 100 was moved into the Mk I’s control room to replace the electro-mechanical analogue computer. At this time, our offline computing was carried out first by the University of Manchester’s ATLAS computer, and then by its CDC 7600 computer.
Similar in size to the Mk II, the Mk III was built some 11km to the south of Jodrell Bank and was remotely controlled over a landline by commands stored on paper tapes, which had been generated by the Manchester computers. We also had a 25m radio telescope at Defford, 127km to the south. This was controlled on site, using an analogue computer to servo the telescope’s position as it tracked radio sources across the sky.
CIRCE (a computer equipped with a 48-bit, stack-based, floating point unit) was built at Jodrell Bank from Texas Instruments’ TTL circuitry and used for interactive analysis of the Mk II / Mk III interferometer data - as well as the ‘hydrogen line’ observations made by the extensively rebuilt Mk 1A radio telescope. It was renamed the Lovell Telescope on its 30th birthday in 1987.
Evolving into the 1970s
In the late 70s, the observatory was given a major grant to build an array of telescopes which became known as MERLIN (the multi-element-radio-linked-interferometer-network). This used the Mk II, Mk III and Defford telescopes, along with three new 25m telescopes in Cheshire and Shropshire. The array could synthesise results that would be obtained with a radio telescope 127km in diameter.
All these telescopes had to be remotely controlled: an obvious solution would have been to use Argus computers (so that all our specialised control software could be used). Sadly, however, Ferranti had stopped producing these machines in 1972, which gave us a real quandary as to how we might implement the MERLIN control system to coordinate the movements of all telescopes in the array.
A foray into new circuitry
Fortuitously, a solution arose: I had been using specialised circuits made by Advanced Micro Devices (AMD) to bring telescope positional data from the Mk I and Mk II telescopes to their control computers. I was contacted by Quarndon Electronics (AMD’s UK distributors) who asked me to study some data sheets that had just arrived from AMD and then go to Derby to explain their use to the Quarndon directors. I was thus, perhaps, the first person in the UK to become aware of the abilities of AMD 2900 microprocessor-bit-slice chip set.
The 2900 chip was a 4-bit device containing 16 registers (just like the Argus computer) and a parallel arithmetic unit. These could be cascaded to produce any chosen word length - such as 24-bits. Further devices allowed any desired micro-coded instruction set to be implemented. It soon became apparent that we could easily build state-of-the-art control computers that could run the Argus instruction set and so allow continued use of the software that had been developed over the previous 14 years.
Using outside contractors to carry out much of the construction, we commissioned 10 Micro-Circe computers, each with up to 16K of semiconductor memory. Six were employed for the on-site control of the Mk II and five remote telescopes; two provided a redundant system to replace the Argus 100 controlling the Mk 1A telescope; one supported a control desk at Jodrell Bank for each of the remote telescopes and communicated with them over dedicated phone lines; whilst the tenth was used to monitor the ‘L-Band Link System’. By using a network of radio links to each remote telescope, this system was able to make the whole array phase coherent - vital if radio images were to be made.
An unmanned mission
As the remote telescopes would be unmanned, it was decided that once the control software had been finalised, it would be loaded into Read Only Memory (ROM) for security. In case a computer glitched, a special code could be transmitted down the telephone line which would initiate a computer restart.
One final computer was required to run the overall MERLIN control system. To achieve the phase coherent requirements, many high precision calculations would be required, so the CIRCE-F was built by adding a 48-bit, stack-based, floating point unit to the 24-bit fixed-point unit used in the Micro-Circe computer.
Professor J. G. Davies, leader of the MERLIN project, had decided to use the Forth programming language which had been developed by Charles H. Moore and Elisabeth Rather to control a telescope at the National Radio Astronomy Observatory at Green Bank, West Virginia. As Forth was a stack-based language, the CIRCE-F was an ideal computer with which to implement the required code.
Go forth and prosper
Forth, an interpretive language, had one great advantage over higher-level, compiled languages: when using it, errors in the code could be corrected ‘on the fly’ without having to re-compile and restart the complete system. Having to restart the system and synchronise the array of telescopes would have taken far longer; without this facility, it is doubtful that commissioning the whole system could have been completed in a reasonable time.
Whilst developing the Forth code, it became apparent that CIRCE-F was not sufficiently powerful to run the system in real time, due to the overhead of the Forth interpreter reducing its inherent processing power. I realised that the computer could be ‘taught’ many of the fundamental Forth instructions, and so these were implemented in micro-code. The computer was able to switch from the interpretive mode into the direct execution of Forth instructions, saving considerable time and allowing real time operation of the system.
An identical CIRCE-F computer was initially used for the analysis of the MERLIN data, but in the late 1980s, we were able to purchase a VAX 11/780 with an AP120B Floating Point co-processor and then an Alliant FX8 mini-supercomputer.
The MERLIN system ran reliably for over 10 years until the early 90s, when it was enhanced with the addition of a 32m telescope at Cambridge. The overall size of the array increased to 217km, with a resolution comparable to that of the Hubble space telescope, enabling much collaborative research.
Gradually, the Micro-Circe computers were replaced by Micro-VAX computers and the CIRCE-F by a SUN workstation. But one Micro-Circe remained, monitoring the L-Band Link. Commissioned on April 11, 1978, this was upgraded with a faster AMD chip set and remained in use until 2017 - after 39 years of operation.
Forging into a new millennium
At the turn of the millennium, Jodrell Bank was given a major grant to upgrade the MERLIN array. Fibre optic links capable of carrying up to 30Gbps between each telescope and Jodrell Bank were installed, as well as a new correlator to combine the data, allowing MERLIN to become far more sensitive. It is now known as ‘e-MERLIN’, the largest fibre-linked array in the world and the second most sensitive. It continues to produce world-class science, along with that of the Lovell Telescope.
Now undergoing its third major reconstruction, it is pre-eminent in the observations of pulsars and a key element in the European VLBI Network, which can produce the highest resolution radio images of any instrument in the World.
Ian Morison is giving a talk on behalf of the Computer Conservation Society about Computing at Jodrell Bank on 21st November 2019. Find out more and register your interest
Image credit: Christopher Elison