All computers should look like the AKAT-1 from 1950


#1

[Read the post]


#2

nice, but it would be even better in a different colour


#3

So… basically 98% of all partial differential equations people need to solve?


#4

It did just ODE’s, and only for initial value problems.

What used to be called “analog computers” are still in routine use today. Anyone who’s built an oscillator or active filter out of op-amps has used one. It’s just that we no longer think of it as computation.


#5

I’m a little surprised it’s 1950, actually, if it is transistorised. The transistor wasn’t even invented till 1948.
Yes - I’m right - it is 1959, which explains the Sputnik design ethos. The first really commercial transistors date from around 1954, and the “universal” 2N404 dates from 1958. It wasn’t possible to build a really stable operational amplifier until silicon planar transistors came along, as those old germanium transistors were extremely temperature sensitive. But by the time the silicon planar transistor arrived, digital computers were already consigning analog to the scrap heaps of the Western world.


#6

I had an answering machine about the same size back in the 70’s. I miss it.


#7

A thing of beauty is a joy forever.


#8

The real issue was offset compensation. The fact that the beta was highly temperature dependent wasn’t that major an issue because you could tame that with lots of feedback. It was really hard to get a pair with well-matched offset. There were a lot of designs that depended on close thermal coupling of a pair of transistors (usually tied to a common emitter resistor, the current mirror didn’t come along until later).

Vacuum tube op-amps were also common prior to that. Something like a 12AX7 made a nice differential input stage. Interstage coupling got weird, though.


#9

That’s a real sleek machine. Would be a sweet setup for a synth, audio board or DJ set up :smiley:


#10

Every Terry Gilliam fan is thinking, “That’s from Poland, huh? I woulda guessed it was from Brazil.”


#11

Also, it isn’t as if digital computers give you precise values for those either.


#12

That’s so much cooler than the one Danny and his friends used.


#13

They made up for it with ISIT though.


#14

I made myself unpopular with the demonstrators as a 1st year undergraduate because, faced with building an op amp using 2N3819 in the input stage, I asked if I could use 12AX7s instead (200VDC was available on the lab benches.) I also had a good 12AX7 and base handy.

Germanium transistors were basically made by diffuse and hope, and were individual dies. Once planar silicon arrived you could be reasonably sure of getting matched pairs so long as your lot all came off the same part of the same wafer. AD used to make extremely close single die matched pair for custom circuits and as far as I know they may still. The difficulty with lots of feedback is that it is fine if you want to, say, build a thermometer, but for an analog computer with any frequency response the current through the matched pair will vary, causing differential heating and a rapid change in Vbe, and bang goes your offset. I remember giving up on trying to help a research student in the early 70s - the issue was that the output went to a plotter several rooms away, and the drift was such that by the time the plotter was zeroed the circuit was already drifting. There was also no telephone link between the rooms. I could have replicated that experiment nowadays with a few op amps and a single PIC microprocessor connected via USB to a laptop, instead of it being spread all over a bench.


#15

I loved that book.


#16

I wonder if every undergrad electronics course at the time had a “build a differential amp out of crappy JFETs” as an assigment!

AD does indeed still sell matched BJT pairs. http://www.analog.com/media/en/technical-documentation/data-sheets/SSM2212.pdf is typical.

Yeah, as I said, offset compensation was the real issue, because the varying beta was manageable. (Except for the fact that you kill the bandwidth, but for a mechanical plotter who cared?) I can just picture you running madly between the rooms trying to tame that setup! (And I can recall pushing the cases of a couple of 2N109’s together with grease in between, in hopes of having good enough thermal coupling to keep them matched for at least a few seconds. I think I was maybe eleven at the time.)

When it comes to electronics, we’re living in the good old days.


#17

I have a sudden urge to buy a USB analog VU meter.


#18

Partial diffyq is an odd beast. I loved the hell out of the class but with things like boundary value problems they are good for a small set of initial data and get more wonky the farther away from your initial value. (this is why local weather forecasts longer than say 3 days are mostly guess work)


#19

That would make a totally sweet casemod.


#20

In fact, it isn’t as if any computer whatsoever can give you precise results for any real-world problem, for some meaning of precise.

I remember being told by a slightly excitable engineer that 12-bit A/Ds were now available with a 12us acquisition time which meant we could now digitise a 50kHz signal we were working with. He was not pleased to be told that 12 bit, 12us equates to a practical maximum full-resolution digitised signal around 12Hz.