Home Brew Analog Computer System

[GK]

I just remembered another article, where the editor must have been a young person.
This was in EDN, maybe 10 years ago.  It was on the history of the vacuum-tube operational amplifier.
First, it described the design by Julie during WWII, which used a 6SL7 (octal) dual triode as the differential input stage.
Then, it described the postwar commercial design from Philbrick, a very similar circuit using a 12AX7 (9-pin miniature) dual triode.
Both devices were referred to as "high-micron" triodes in the article, as a misunderstanding of "high-{mu}".

Hi Tim,

The full manual for the EC-1 is here:
http://www.analogmuseum.org/library/heathkit_ec1_operation_manual.pdf

The whole "library" of Bernd Ulmann's "Analog computer museum website" is worth checking out:
http://www.analogmuseum.org/english/library.html

It was my initial discovery of the EC-1 manual and then Bernd's site that got this whole project started.

[T3sl4co1l]

Hmm, fascinating.  Nice test!

Also, that's not a bad 8/20us pulse, crudely speaking.  If that puts any kind of perspective on it, relative to conventional surge protection diodes.

The datasheet says 4A I_FSM at 1us, so you're definitely well beyond that, but for the range between "yeah, it'll do this under continuous service for decades" and "it only has to survive a few hundred surges", who knows.  Those tiny diodes used for USB that claim 15A surge (8/20) and have lower capacitance than this one, can't possibly be holding up too well after their tests... but they only have to withstand 10 of them for the entirety of the test, or whatever.

Tim

[GK]

Made a start on loading the Logarithmic amplifier PCB's, which arrived last week:



The thermistor mounted above the SSM2212 dual precision NPN transistor for ambient temperature compensation of the scale factor:



The original plan was to fix the Coke bottle cap down with Araldite and then fill from the bottom of the PCB. That's what the two (5mm dia.) holes are for - epoxy syringed into one and air out the other. However I'm now perhaps having second thoughts about filling with epoxy (a la Jim Williams) as that would make any future re-work (replacement of the SSM if necessary) a bit of a bother. Has anyone here ever looked into the long term conductivity/insulation properties of hot melt glue?


The calibrated log conformance and scale accuracy over the smidgen over 4 decades input (10mV to 100V) that the output is within its voltage swing limitations is within 0.1% by my preliminary measurements. Haven't got time to tabulate values and plot the error curve tonight so just a scope screen shot:

[SeanB]

Use beeswax, as then it will still be a good thermal conductor and yet will still be easy to remove for rework.

I have an embarrassing amount of Tektronix wax blocks though, all black ( black is 'free' when you order the colour wax blocks) if you want the finest quality ultra pure wax to use in that application. All in nice easy open single shot packs. If you want some shoot me a PM and your address (again) and I will pop them in the post for you on Tuesday ( Monday is a public holiday) if you want to wait the 2 weeks.

[GK]

Ooh... that reminds me, I still have a prezzie (a bit lame really) sitting on the shelf here to send in return. I don't know much about beeswax. Is it non hygroscopic and a very good insulator? What did Tektronix do with it? Google just brings up references to "Phasor" printers.
 

[SeanB]

Bingo, I have one Phasor printer still intact, and in perfect working order, you just need wax ink at $200 per pack of 3 blocks, which it will happily consume in a month even if you do not print a single page. If you turn it on it does a clean cycle, which uses a half block of each of the Cyan, Magenta, yellow and Black ( black is free BTW) wax blocks.

I used to print a lot of jobs to that printer with a black background and white text, just because it was so good at making a very good dense black print, and of course it actually does not cost any more, it will have dumped that black wax in the maintenance tray in any case.

There was one DTP graphics shop with a dozen of these printers, and a salon next door, which used the black wax blocks to darken the regular pale wax to do hair removal, sort of as a differentiator.

Beeswax is not hygroscopic, and is traditionally used as a protective coat in RF circuits, you will have come across it in older transistor radios as a coating applied to the FM RF and AM RF sections to keep them all from vibrating or moving if dropped. Melts in boiling water or a double boiler for you to pour into a mould, and is soft enough and flexible enough to enable rework without taking it all off.

[GK]

OK, I'll send you a PM.


--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

I am currently finalizing the electrical design of a companion for my log amplifier module - the antilog module.

The log module detailed already has the transfer characteristic:

eout = k (ln * ein)

where ein is a unipolar intput ranging 10mV to 100V (4 decades).
k is a fixed scaling factor of 20 and eout clips at a +/-105V. A table of ein/eout looks like this:

Code: [Select]

ein        eout
10mV      -92.1V
100mV     -46.05V
1V         0V
10V        46.05V
100V       92.1V

The antilog amplifier the direct inverse transfer characteristic:

Code: [Select]

ein        eout
-92.1V     10mV
-46.05V    100mV
0V         1V
46.05V     10V
92.1V      100V

The equation is simply

eout = exp(ein/k)

where k again is a fixed scaling factor of 20.
So if you wire a log module in series with an antilog module (either way around) you end up with a linear transfer characteristic (disregarding errors).

My, perhaps stupid question is this: what is the most commonly accepted way of expressing a natural antilogarithmic equation?

exp (exponential function) as above, or:

eout = antilog_e(ein/k)

that's _e indicating base 2.71

or:

eout = e^(ein/k)

?

Here is the basic schematic:

[T3sl4co1l]

I don't know that there's any significance between "antilog" and "exponent".  I would go with "exponent".

Whereas there's a difference between antiderivative and integral (the integral is with the "...plus a constant" part, depending if it's definite or indefinite).  Of course, your integrator module is implicitly definite, so that's taken care of.

There are different types of logs, but this is an analytical rather than computational luxury (the complex logarithm has an infinite number of corkscrew layers to choose from; the Ln(z) function denotes the simplest "main branch" choice around zero).

Tim

[GK]

Antilogarithm appears to be an antiquated term for the exponential function. The Wikipedia "Exponential function" page cites a text dated 1911 for usage of the term. Funny that before I stumbled upon the Wikipedia page I couldn't find this explicitly and unambiguously stated anywhere.

So long as the superscript is practical I think I'll prefer e^x. That is more descriptive of the function (Euler's number raised to the power of x)  than exp(x).

[recjohnson]

Very interesting thread! I have just started reading and have a long way to go.. The sine function generator you posted in the first post reminded me of a function generator chip ICL8038 that has a similar topology inside.

While I go through rest of the thread, I just want to say, this is a damn impressive project! Thanks for posting in such detail.

[GK]

Hey, no worries. There are a few pages to get through now 

[GK]

Well, this evening I "wrote" and ran my very first program on the thus far completed hardware. I was about to strip the 1st iteration of the main board of the computers master control unit of it components as I have since improved the design and will be re-building it, but before doing that I patched it to one of my completed integrator chassis' and ran a simple physics problem. Here it (the master control unit card) is, patched to the integrator chassis:



You can't do a lot of simulation with just a panel of integrators, but one well known physics problem that can be solved is the falling body problem. It only required two integrators:



Here is a rather crappy video of the program being run. The simulation solves y and the computer is operating the the repetitive mode. The horizontal axis is t and the vertical axis is y. The control unit furnishes the re-trace blanking signal and the horizontal sweep of t (the independent variable) for the oscilloscope.

Note that this isn't a plot of the bodies trajectory but the distance that it has fallen during time t. To make the program a little more exciting I added a negative initial condition value for y, instead of starting at zero velocity. The I.C. value is negative, so the very start of the solution is a plot of distance climbed, rather than fallen.

I started recording the simulation with the constant for gravity (g) at zero, so the initial solution for y is just a linear ascent straight up into outer space. However as I manually wind the constant for gravity up from zero, the ascent is eventually overcome and the body begins to fall at an accelerating rate.



Incidentally, here is what the finished master control unit will look like when finished. I am working on this chassis now, with a degree of priority, because with the completed master control unit handy I'll be able to actually start running some serious programs as I slowly complete each successive chassis of analogue computational units. 

[GK]

...................Also got my front panel for the display unit from Front Panel Express. Things are slowly progressing.........

Can't believe that almost 3 years has passed. I've had the front panel taking up shelf space since and the CRT has been sitting on a towel on the floor just waiting to be accidentally tripped or knocked over. It's finally time to get this thing finished and mounted into the computer systems relay rack, out of the way.

I ended up getting a little more adventurous with the design. I originally had a low-voltage, high current current-sense feedback amplifier for driving what was originally the horizontal coil of the CRTs yoke, and a high-voltage current-sense feedback amplifier for driving what was originally the vertical deflection coil.

The H coil has 2mH inductance and the V coil was comprised of two individual windings of 70mH each which were originally connected in series for 140mH. I had connected them in parallel for a much more manageable 35mH. Well it doesn't need to be said that the comparative bandwidth of the 35mH channel was crap. The small signal bandwidth was only 5kHz, while I had eventually gotten the 2mH axis close to 100kHz.

Well today I fixed that problem. Here is the disassembled yoke still retaining the 2mH (horizontal) deflection coil in the inside:




Here is the new vertical coil after I re-wound it. Each ferrite core half was originally wound with several hundred turns of thin copper wire to give the original 70mH per winding. Now there is 52 turns of much heavier enameled copper wire returning only 325uH per winding.



The yoke re-fitted to the CRT. I am still using the donor televisions old motherboard to supply the operating supply voltages for the CRT. Haven't progressed with this project to the extent of designing my own switched-mode supply yet.   



Here is the system up and running (my "Sprott Systems" analog computer [case M] providing the deflection signals). The current-sense feedback deflection coil driver amplifiers are those thingos sitting in front of the CRT. The originally high-voltage one has now been heavily modified to run on the same low-voltage power supply rails as the other amplifier. Performance is now awesome and more than adequate for repetitive analog computer stuff. Small-signal bandwidths for both X and Y deflection channels are currently in the order of 100 kHz, full-power (or rather full-scale deflection) bandwidths several kHz (current deliverable to the deflection coils is slew-rate limited by the maximum voltage that the linear amplifiers can swing).



This weekend I'll get the case fabrication completed.

[T3sl4co1l]

Curious: any interest/ideas about a constant-intensity display?  That is, intensity ~proportional to d(x, y)/dt?

Tim

[GK]

Don't think so.

[timofonic]

Would a similar design using SMD be possible and provide some advantages or the high current would do it unable to be made?

[rstofer]

Would a similar design using SMD be possible and provide some advantages or the high current would do it unable to be made?

The OP is building a 100V Analog Computer.  SMD parts may be hard to find.

Later generations used 10V as '1' so, yes, SMD parts definitely apply.  I used socketed 8 pin DIPs so that I could try various op amps (which I never have) but my next version will likely use SMD.  I just need to settle on some particular op amp.

Accurate capacitors will be a problem regardless of form factor and the capacitor is the key ingredient for the integrators.  There used to be some very accurate capacitors used in commercial analog computers.  I have yet to find a source much less in SMD.  0.1% would be acceptable.

Here is the machine I built.  It has 2 integrators, 3 summers and an inverter plus some other gadgets like a multiplier.  Two integrators is a serious limitation except that it will handle many problems in mechanics as it can handle acceleration, velocity and distance as this only requires two integrators.

http://www.analogmuseum.org/english/homebrew/vogel/

[GK]

Would a similar design using SMD be possible and provide some advantages or the high current would do it unable to be made?

Since an analog computer is an antiquated thing I decided to stick with traditional through-hole parts (mostly) throughout, though if you are only referring to the shown deflection drivers for the display unit, they could be mostly implemented in SMD except for the power output devices, which is more practical in through-hole due to the power dissipated. For example the originally vertical (now horizontal as the display is rotated 90 degrees) deflection channel with the re-wound yoke coil requires just shy of +/- 2A peak for full scale deflection (to deflect the dot from the center of the screen to either edge). The supply rails are +/-40V, so if the dot is (horizontally) static at the edge of the screen the deflection amplifier is dissipating ~80W.
 

[GK]

Here is the current supply arrangement to provide all of the operating potentials required of the CRT while I continue to prototype and test the (again) re-designed/improved deflection amplifiers.

I hack-sawed the direct off-line part of the PSU from the donor televisions motherboard and rebuilt the LOPT section dead-bug style on a rectangle of blank PCB, minus all of the superfluous functions.

The direct off-line part is a bit of a crappy discrete self-oscillating fly-back with voltage mode control - I have actually traced out the circuit in its entirety. A BU508 power transistor does the primary switching and a auxiliary secondary winding is used exclusively for feedback and regulation. The secondary winding for the supply rail outputs has a single tap to provide two supply rails (half-wave rectified) of 18V and 110V. The associated rectifier and filter caps are now located on the new LOPT board.

I'll design my own substitute off-line regulator utilizing the magnetics, but I'll do away with the crappy BU508 and all those high power resistors associated with the base-drive circuitry.

The 110V rail, originally serving both the deflection an LOPT circuitry, now only serves the LOPT. The 18V rail now powers an LM555 timer (via a 3V zener V-dropping diode) which substitutes for the horizontal timebase. The TVs original circuit used a 2SD1398 "flyback driver" transistor (with integral anti-parallel diode) BJT to switch the primary current of the LOPT, itself driven from a base-drive coupling transformer and open collector switching/driver transistor receiving the 15.626kHz, ~50% duty cycle horizontal signal from an LSI "horizontal/vertical combination" integrated circuit. I substituted all of that with the LM555 directly driving an STW4N150 MOSFET.

These LOPTs operate in a resonate mode with an external parallel capacitor (which I simply pulled off the TV motherboard). When operating correctly the primary voltage waveform is a clean high-voltage half-sinewave, as shown in that attached scope screenshot (alongside the MOSFET gate-drive waveform). The LOPT provides the final anode, focus grid, screen grid and heater supply voltages for the CRT.

 

[GK]

Long time no post. Got this little projects 19" frame assembled this evening. Now I have 84U to fill.......

[GK]

Just soldered the very last connection inside the display unit. I still have some lengths of 10x10mm square aluminum rod to cut, drill and tap to attach the top/lid before I can call this complete and bolt it into position in the 19" rack, but it's currently over 50 deg. C in the garage so I think I'll put off doing that for another day.

A graticule for the display will be electronically generated by a separate patch-able  unit, which will mount in the rack immediately below the display unit. This is what I am currently working on. The drawing of the graticule on the screen will work on a time-shared/alternate basis, synchronized/multiplexed by the computers control unit (the graticule being scribed in the computers repetitive mode whilst the integrators are being reset to initial conditions).

With the CRT having been idly sitting around lots operating equipment for some time in the meanwhile, it looks like the shadow mask has become non-uniformly magnetized a far bit. If several more power on/off cycles of the internal degausing circuit does not fix the colour impurity (it's a long time to wait for the PTC thermistor in series with the degausing coil to cool down again) I might have to home-brew a heavy-duty wand.

[GK]

Just for fun this evening I worked out a breadboard implementation of the "bouncing ball in a box" analog computer simulation detailed here: http://www.analogmuseum.org/english/examples/bouncing_ball/

It was a success:
 
EDIT: video deleted

The solution runs for ~20 seconds before automatically resetting/restarting. If you turn the volume right up you can hear the DPDT relays clicking. A more refined version would use solid-state analogue switches rather than relays, but I used what I had at hand. Here is my schematic:

EDIT: schematic deleted

EDIT:
Oops. I knocked this thing up on the breadboard in a bit of a rush "on the fly" and have just noticed that I got some signs mixed up in the circuit for solving x and the result is that (in the horizontal motion) rather than loosing speed with time the ball (defying the laws of physics!) gains speed instead. It's 1am now and I am going to bed. Will fix this tomorrow and up-load a replacement video and schematic.

[BrianHG]

I thought it was meant to do that.  It looked cool...

[GK]

I thought it was meant to do that.  It looked cool...

It did look cool, but unfortunately no, the ball gained energy with time. Anyhow, I fixed it and now it operates properly. IMO it looks even cooler now that the ball movement conforms to reality, slowing down with time and eventually almost coming to a halt. Here is a new video and the revised schematic:



 

[BrianHG]

Your right, that one's cooler...
Now all you need is 3 push button paddles, one for each border, to manually add energy back into the bounce only when struck exactly at the right time changing this into some sort of user game, or pushing just prior to the ball hitting it's border stealing it's energy slowing it down further.