From: jl@glen--canyon.com   
      
   On Wed, 11 Feb 2026 23:38:42 +1100, Bill Sloman    
   wrote:   
      
   >On 11/02/2026 6:39 am, john larkin wrote:   
   >> On Tue, 10 Feb 2026 07:49:22 -0900, Christopher Howard   
   >>  wrote:   
   >>   
   >>>> Standard apprach for low-accuracy approximation is diode-resistor   
   >>>> network.  But such network gets inpractically large if you need   
   >>>> good approximation.   
   >>>   
   >>> Thank you. Looking at all the suggestions put forward, it seems like —   
   >>> for my application — that just getting the AD633 might be best choice.   
   >>> My analog computer is short one multiplier for a simulation I wanted to   
   >>> play with, and I wonder if there was some easy trick to get a reasonably   
   >>> accurate square function without getting a decent multiplier. In a lot   
   >>> of basic physics simulations, one has to square velocity to get   
   >>> things like drag or lift forces.   
   >>>   
   >>> Line approximation with a resistor-diode has some appeal. Maybe it could   
   >>> be accurate enough for my little educational experiments? I think, to   
   >>> pull it off, I would need to use trim pots — two per segment, for the   
   >>> biasing and the attenuating.   
   >>>   
   >>> The schematic from the neurological paper seemed to be a line   
   >>> approximation solution with some of the diodes in the op amp feedback. I   
   >>> didn't try to get the whole research paper so I'm not sure how one would   
   >>> work out the correct resistor values.   
   >>>   
   >>> I downloaded the information on the MC1496-D and LM13700. I see the   
   >>> basic idea is that these chips also produce products of signals, but   
   >>> beyond that I can't make any intelligent comments at present. My analog   
   >>> computer is usually dealing with signals in the range of a few hundred   
   >>> Hz up to a few kHz. I am curious if maybe something could be done with   
   >>> the gain control pin on the LM386 chip (I have quite a few of those). It   
   >>> looks like, using a series RC network, that the gain can be set anywhere   
   >>>from 20 to 200.   
   >>   
   >> A simple opamp circuit with some resistors and diodes can have a   
   >> several-segment approximation to a square root. I did that in a   
   >> steamship throttle control once and it worked well enough.   
   >   
   >Didn't you have a DEC digital processor to play with for that gig?   
   >   
   >The DEC PDP-8 I got to play with for my Ph.D. project could be persuaded   
   >to do multiplication and division - I used it for signal averaging - and   
   >square root extraction shouldn't have been difficult.   
      
   Yes. I had a PDP-8 12-bit machine running FOCAL, and simulated the   
   steam valve, the turbines, the prop,  the hull, and the ocean. I   
   plotted the responses on a teletype machine. We showed it to the   
   shipyard and the owners and they liked it and we got tho sell them the   
   engine room consoles. That was for the 32,000 hp, LASH ships. I think   
   I was a sophmore at Tulane at the time.   
      
   Sea trials were fun. People treated this gawky kid like a god and I   
   got my own cabin.   
      
   The actual control loop was feed-forward from the throttle to the   
   steam valve position, with a diode nonlinear function generator, and   
   limited-influence RPM feedback. That limited the number of people who   
   might be killed if the feedback tach failed. Which one did, but that's   
   another story.   
      
   Ships are most all diesels now. Steam plants are efficient but complex   
   and it's hard to get crews who can run them.   
      
      
   John Larkin   
   Highland Tech Glen Canyon Design Center   
   Lunatic Fringe Electronics   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)