I never worked with those systems, but I have relevant experience in a closely related field.
Beginning in the late 1960's I was a gradate student working in molecular motion studied by microwave absorption. The lab then got a Far-Infrared Interferometer to reach into that frequency region "next door". You see, our highest-frequency microwave system was a 70 GHz, and later we got one for 150 GHz. If you do the conversion, 150 GHz has a free-space wavelength of 0.2 cm. In the jargon of Infrared spectroscopy, that is "5 wavenumbers", a "wavenumber" being simply the inverse of the wavelength in cm. The equipment we got over a few years covered the range from 5 to 250 wavenumbers. That's very far (lower frequency) from the typical infrared spectroscopy region of 4000 down to maybe 500 wavenumbers, so it's "Far-Infrared".
In that region the hardware has two significant limitations. There are few high-output sources of radiation in that region, and what was used then was a150 W mercury vapour lamp which, acting as a Plank black-body radiator, put out "light" throughout the visible region and lower, but not a big output in the low Far-IR range. Then came the detectors. The common one used was a Golay Cell. This basically was a tiny chamber with a quartz front window for light entry from the source and filled with an inert gas (I think Argon, not sure) that would absorb incoming energy and warm up to expand. In front of the quartz window was a black polyethylene film to filter out much of the source's light and allow the low-frequency IR radiation to get to the detector. The back of the chamber was a flexible mirrored membrane, and a small light source was focused on it and then reflected back to a common photodetector. The light from the source lamp was passed through a chopper wheel so that the Golay cell output was a sine wave at the wheel speed. This went through a narrow-band amplifier and a phase-locked amplifier stage (locked to a position sensor on the chopper wheel) to give a very high amplification for the weak signal. The final result was digitized (12-bit resolution) and punched out to paper tape.
Thus we had a system that could use a weak source and a modest detector system with very high signal amplification to get a signal. That signal was MUCH too weak to waste most of it through a typical grating or prism dispersion device to separate wavelengths, as a "normal" IR spectrometer of the day would do. Instead, the entire output of the source went through the beam-splitter and moving mirror system of an interferometer, and the combined output beam from that passed through the sample cell to the detector. This was a single-beam instrument, so one did an interferogram recording (on punched tape) with an empty sample cell, then did another run with a filled cell. The two paper tapes then were take to the university mainframe computing centre where the system read in the two tapes, ran a Fourier Transform process on each to generate two separate power curves (i.e., beam power versus frequency), and then ratio the two to generate an output of Absorbance versus frequency, the graph that most spectroscopists are used to.
Far-Infrared spectroscopy only became possible in the mid-1960's when mainframe systems became available with enough computing power to do these Fourier Transforms in acceptable times. The university mainframe we were using was an IBM System 360/50 with 512 KB or RAM, magnetic disk packs, and tape drives for large datasets. A typical "run" for our data of two interferograms processed into one spectrograph was about 10 to 15 minutes of processing time, but about 2 to 4 hours turn-around time for us to wait.
By about 1980 after I had finished grad school some newer systems were becoming available based on microcomputers and minicomputers for both the acquisition/storage of data at the interferometer and the subsequent processing by Fast Fourier Transform. The actual processing time for a set of data was slower than a mainframe could do, but the machine was located in the lab and dedicated to this one task, so the turn-around time for researchers was much improved. At about this time the processing power of such computers had improved so much (and their costs decreased enough) that the Fast Fourier Transform of Interferograms technique became a good alternative design to using classical dispersion grating optics in "standard" Infrared spectrosocopy. As you say, Nicolet and Perkin-Elmer were prominent in that field. Initially it was PDP11 minicomputers, but as the 8086 family of chips exploded into desktop computer systems in the mid-1980's, the computing hardware became desktop PC's, and then improved with the introduction of dedicated Numeric Co-processors like the 8087 and later. Compared to today's chips, the 8087 certainly was more limited. However, compared to previous hardware in minicomputers, those chips and desktop computers were a big improvement in speed, and a HUGE reduction in hardware costs.
A related "impact of technological advancement" tale. I worked in the 1980's as an Industrial Chemist in the paper industry. In the mid-80's I was at a corporate technical conference where one presentation was on the use of a commercial software package called MASSBAL which came out of University of Waterloo. It was a system for modelling the flow of materials (mass) and energy in a large chemical processing system. It had been used strictly on mainframes up to then, and the new change was that one could use remote terminals or desktop computers to access the mainframe that ran the software. One could use a front end to specify the necessary large set of input parameters and the process design, then submit it for analysis and retrieve the results later. The actual work was done on a time-sharing mainframe, with turn-around times typically of a day or two. It was based on floating-point matrix algebra processing, so it took a lot of processing power. Not two years later the new presentation on this topic was a real improvement. One could now buy a fast desktop machine (IBM PC-AT or a Compaq compatible but faster machine) with an 80286 CPU plus an 80287 Math Co-Processor, and buy the software package, and do ALL this work at the mill in one's own office, with turn-around times less than half an hour!
