How the Selectric's digital-to-analogue converter worked

This great video explainer shows the inner workings of a digital-to-analogue converter, using the IBM Selectric's early example of the form to illustrate the mechanism:

Using slow motion video Bill Hammack, the engineer guy, shows how IBM's revolutionary "golf ball" typewriter works. He describes the marvelous completely mechanical digital-to-analogue converter that translates the discrete impulse of the keys to the rotation of the type element.

IBM Selectric Typewriter & its digital to analogue converter (Thanks, Fipi Lele!)



  1. I could piss off Mrs. Lauffer, my typing teacher, by covertly yanking on the lines that controlled the spin of the “golf ball,” which would screw up its alignment and make it type all nonsense.

    What? I was a self-taught typist and could type 90 words a minute by 5th grade and she gave me a D because I wouldn’t return my fingers to home position or use the “proper” fingers for each key.

    After graduation I bought Will & Ariel Durant’s 11-volume The Story of Civilization from her at a yard sale for $20, so it’s all good.

    1. i’ve been in a similar position myself, but she might have just been trying to save you from carpal tunnel…

    2. Perhaps she was concerned that you’d cost her a lot of money.

      I recall that it cost about $70 to get IBM to fix the delicate steel ribbon that moves the golf ball after I broke one installing a printer mod in a shiny new Selectric II in 1978.

    1. The printed character is certainly discrete, yet it is actually achieved by adjusting the ball orientation in the analog (spatial) domain.  The discrete output is the result of the extreme accuracy of the conversion (far beyond the number of input bits used).  With arbitrarily more input bits, it could be designed to move to “in between” orientations that would print misaligned combinations of characters, but there would be no point to it.

      1. are you saying reality is analog? I am pretty sure that reality is not a representation at all, but you may know better than I.

          1. You say the spatial realm is “analog.” Do you know that means it is a representation of something else? What is it a representation of?

            edit; (it is a pet peeve of mine.)

        1. When you say ‘reality’, do you mean ‘material reality’? Because reality is de facto real, while material reality may be construed as a metaphorical representation of reality.

          1. Don’t be distracted by that tunnel of mirrors over there, we are talking about . . . what was it we were talking about?

  2. There were also Teletypes, which had a different way of doing this. The model 14 had a 5 bit to 32 output digital demultiplexer built from sliding metal plates.  The plates would permit only one of the type bars to fling up at the paper. The later model 33 worked more like the Selectric with a type drum, but it didn’t use those delicate ribbons. Teletypes had to run continuously, mostly unattended, for years.

  3. Matthew Miller is right: there’s nothing analogue here at all. A digital signal–which letter on the keyboard was struck (choose one from 88) is being converted into a pair of digital signals–the extent of the tugs on the two cables (one from four and one from 22)–which are recombined in the movement of a ball to make another digital signal–the letter of the alphabet being printed (choose one from 88). At no point is anything being represented by a continuously variable signal: each signal is allowed to take only one out of so many discrete positions. I love the Engineer Guy’s videos, but getting this CompSci 101 point wrong shouldn’t happen.

    1. I know where you are coming from. I thought long and hard on this and came down about where immovable object did.  Perhaps better to classify it as a kind of mechanical adding machine? Still …. a fascinating mechanism isn’t it?

    2. No, sorry, you’re wrong. The pull on the cables is a continuously variable value; the converter only emits certain values (its input is digital; it cannot possibly do otherwise) but that cable tension itself is an analogue value.

      As a good comparison, look at an actual electronic DAC. These take a digital input signal – let’s say 8 bits of data, for example – and output a voltage level based on that input. The 256 possible output voltage levels are discrete in exactly the same way as the 22 levels on the cable are, but the signal is still considered to be analogue, because intermediate levels are physically possible and would be interpreted as such – i.e. if you output a half-way level between two letters, you really will end up with a mess of two letters – it doesn’t get re-digitised back into one letter or the other.

    3. I think you have conflated the concept of quantization with being digital.

      While the cable displacement on the Selectric is quantized (limited to a specific set of values +/- some small error due to mechanical tolerance) , it just as fully “analog” as the similarly quantized electrical output of an electronic DAC (Digital-to-Analong Converter). 

      The fact that the chosen quantization levels ultimately map to specific characters on the type ball, doesn’t negate the inherently analog nature of  mechanism. 

  4. Thanks for posting this, Cory! 

    I’m always fascinated by how mechanical devices solve problems, and without using driver-PCB-and-stepper-motor, buggily coded, fail-after-five-years, and completely unrepaireable computerized equivalents.

  5. I used to work on these back in the 80’s. I still recall the major workings, lthough I cannot recall all 385 adjustments that could be made. They employed an ingenious use of spring clutches, pulleys, spring latches etc. Entirely mechanical accept for the electric drive motor, which could be replaced by a couple of staunch hamsters anyways for a totally off-the-grid solution.

    1. Yeah … that’s main point here and really interesting part: This is an “ELECTRO-mechanical” not an ELECTRONIC device!  As you say hamster, coal fired turbine, anything will drive it in principle. I was able to dig up the early papers from IBM on this: They make fascinating reading.

      1. Steam powered goodness :) Can u please point me to those papers you mentioned, I would dearly love to get hold of them. I lost my maintenance manuals many moons ago. 

  6. Back in the 60s, the IBM 1130 console used a Selectric typewriter for printing system messages. Unlike on more expensive IBM models, the software drivers had to do some of the work to drive the golf ball. In particular, it had to convert EBCDIC (the IBM character coding before ASCII) into PTT8 which was an eight bit pivot and turn code with some of the bits telling how far to pivot the ball and the rest how far to turn it. Nowadays, this kind of conversion is completely hidden.

  7. But using your logic, isn’t the keystroke a quantized analog value as well? Making this an analog-to-analog converter?

Comments are closed.