Dedicated Circuits for Analog Computing

These links should help (covering op amp analog math circuits)

http://www.scribd.com/doc/10782864/Basic-Op-Amp-Applications

http://www.opamp-electronics.com/tutorials/computational_circuits_3_09_06.htm

Simple binary math can also be done using logic gate integrated circuits (IC's).

http://en.wikipedia.org/wiki/Logic_gate

Thanks, jackofalltrades,

It'll take a bit of time for me to go thru all the references.

Bob Clark

Presumably you are doing this just for fun?
It seems a tad masochistic, I'd think you'd rapidly run out of voltage if you start raising to powers without some careful scaling. The big prob is you are liable to end up with a circuit only suitable for solving one specific type of equation.
Setting up initial conditions and getting sensible I/O is also liable to be tricky.
Writing some clever assembler to do it may be more relevant?

For my first year applied physics degree course project I was set the task 'solve second order differential equations using op-amps' This was a ludicrous project for a course that had only included some basic electronics.... ..(so my attempt was pathetic)
For a long time this bugged me and gave me a chip on my shoulder, I was even tempted to do it now that I'm an electronics designer...until one day I came to be at peace with my inner demons* and accepted that a) I have nothing to prove b) It was stupid project.

Hmmm sorry for the rambling...and for the fact I havn't answered the question...baaaad kitty

Ah maybe you'd need to do it with log/antilog circuits, then you could use addition/subtraction?
Del

* Insofar as any of us are fully at one with our limitations.

Here they are:

http://www.analog.com/en/other/analog-multipliersdividers/products/index.html

I hope you know what you are asking though. Del is correct - this is a beastly realm to work in.

Thanks LG Dave,

Nice reference. I hadn't found this particular analog.com site. You're right, I don't know what I'm asking. That's why I asked! Cheese!

Thanks again.

Bob Clark

antilog(log(X*Y)*Z)

Thanks, rcapper,

I agree, alog(log(x*y)*z)=x*y^z. But maybe you could tell me how to do this elecronically, using analog calculations of voltages with the answer showing on a simple analog voltmeter. That's the hard part. For me anyway.

Bob Clark

you have already found multipliers, there are also log and antilog converters. try Analog Devices or if you can't find anything I will dig into some old circuit books.

I think that should be x*alog(log(y)*z) so 2 multipliers, 1 log circuit and 1 alog circuit.

Good point. Strict standard notation would concur. I made a different interpretation based on the "word" part of the problem as presented and if correct then the equality should be

alog(log(x*y)*z)=(x*y)^z

Although, it is possible that he really did have it written correctly, in which case your conversion is correct.

Hamish, what was your your original intent?

You did not mention all of the necessary parameters. While you have specified the range of X and Y, you didn't say what Z is and what accuracy you need? Contrary to many engineer's opinions, digital is not always the best or easiest method of doing something. Digital has a lot of overhead requirements that analog may not.

Del the Cat is correct, you will have to carefully choose your input and output voltage ranges in order to accomplish your calculations, particularly with an exponent output result. I suggest you go to National Semiconductor and check out the AN-222 application note and the data sheet for the LM194 as a starting point. These circuits operate off of +/- 15 volt supplies which should be well regulated and low noise. In order to maintain linearity, output should be kept under 12-13 volts, 10 volts is even better.

You may be able to use one of the voltage multiplier chips to do the X * (Y^Z) result but voltage multipliers are not good at exponentials. In this case, use the LM194 circuit to do the exponential function and then use a multiplier chip to multiply the result by X or you can just do the whole thing with LM194 circuitry, either method should work.

Inputs may need to be scaled down by a factor of 10 or more depending on your output voltage needs. Run your calculations with scaled down voltages and see if the output is in the correct range, if not, reduce it further. You can also convert your input voltages to currents which might even be easier (just requires a series resistor at each of the inputs). These circuits have good accuracy over about 6 decades of current/voltage input/output. Many math operations can be executed with these circuits and they are reasonably fast and real time.

All of this was accomplished back in the good old days with vacuum tubes as well and they were quite accurate to boot.

Electronic wiz,

I do believe, you just may be with the right name.

Thanks,

Bob Clark

Hey folks:

I wanted to post this to a few of you specifically, and simultaneously, but there really doesn't seem to be a way to do it.

I am working on some robot designs, nothing particularly fancy, and more for my learning, and that of some budding electronics students I work with (Middle and High, or Level II and Level III schools), and like the fact that analog circuits, particularly where used as motor control circuits, can be shown to "learn" or optimize, as time goes on, and I like the granularity, even if not always linearity, or the wide range of values analog can make decisions based on. I also like the fact that analog "neural networks" seem to be more "I think so, so I'll do this to some extent while I learn more" based, than the "I know so I'll go all out (or I don't know so I won't do anything)" basing of digital decision makers.

But I've run into a couple of problems for which I can't find simple solutions, and I'm wondering if you can point me in a profitable direction on them.

At the end of a "run" (whatever the goal might have been), if I haven't achieved the goal, and want to start again from where I left off, how do I get an analog circuit to remember what it has "learned" to that point, so that, at power up for the next attempt, I'm not restarting from scratch? This is especially important if I need the system to return to its start point to baseline a value there (sun position, for example), and then speed through a range of motion to get to where it remembered being the night before, and then fine-tune the system position based on both factors. Two factors are easy to work with in analog, but not if you can't remember one, and going to sample the other loses what knowledge the system has.

I know about, and have used (even built some of my own) A-D and D-A converters, and can see without difficulty how to implement a register, and store a value till tomorrow, but I'd prefer not having to interface digital and analog any more than necessary.

Also, I've used digital H-bridges, and PWM to control differential steering in a bot, but I'd like to do something similar in a purely analog design. Any ideas how to go about the polarity switching to allow DC motors to run in reverse, without fancy hardware switches? I'm playing with the idea of reed switches or steering diodes, but obviously, the more complex it becomes mechanically, the less likely it is to work reliably. I've also thought about letting a threshold signal drive the switch logically, and just employing H-bridge circuits there, but what I'd really like is for the system to work so that my students can actually see a reaction to a decision point working its way through the controls.

Micah

Sample and hold circuits?
Dunno how many 'values' you want to 'remember' for for how long.
There does come a point where digital is usefull, I must admit I still like the cunning little tricks one can pull off a hand full of Rs and Cs and a few CMOS gates...it's that nice area between digital and analogue..
I do love a nice logic 'half'
Del

Dunno how many 'values' you want to 'remember' for for how long.

Probably not more than two, position of this, and position of that relative to this. But of course, those are probably, sooner or later, going to involve some kind of coordinate coding in three dimensions (X, Y, Z). And storage from sundown to sunup sounds good, so figure, for the long winter nights where I live, plus a margin of an hour either way, 16 hours will work.

I like them simple, too, and really prefer being able to modify circuitry without having to play with timing signals and such like.

Micah

micahd02:

I want to look over your request then ask some questions and make some comments.

Two brief ones now; what accuracy are you talking about? Analog storage over a period of 8-12 hours is tough to do with accuracy with an analog sample-hold circuit.

Acquisition time, of necessity, needs to be short, hold time long, which creates a problem...a long hold time requires a very high time constant which, in turn limits speed of acquisition. I'll run some calculations but I think you will have to use an A/D/A storage medium to attain that many hours of signal holding.

What about using a non-volatile digital potentiometer?

I've used those for setting and storing analogue limits within a measuring instrument before now....

Electroman:

Yes, that could be a good possibility without too much mess.

Probably not THAT fast on the acquisition. I'm looking for "where am in sitting now?" just before it shuts down for the night, or to take off to go get recharged, or some other such housekeeping function.

And the precision is along the lines of "I was in this hallway (at home, no cavernous mazelike interiors to deal with. Maybe at the school, but again, nothing huge or mazelike) and didn't find it, so I'm going back there to look some more", for which I can design it to start looking at the entrance of the hallway, even if it was in the middle, having already cleared the first half yesterday. So precision for what I'm doing is not nearly as necessary, nor speed of sampling, as being able to remember what it "knew" the day before.

Micahd02:

The suggestion of non-volatile digital potentiometers is a good one I think. They will retain 'position' with the power off. I would suggest either 512 or 1024 position pots. Use + and - power supplies with center position (0 volts) as 'home', left or reverse (for example) can be negative voltage and right or forward positive voltage. Use a pulser to clock the pot every so often and to change directions, the pot's counter can increment or decrement with direction as necessary. Shouldn't take much circuitry.

Cheers.

Thank you. Sounds like half of an H-Bridge Circuit, with PWM could do the job. Of course, now I'm back into digital, and that would be for presetting the pot to a start point. On the other side, allowing a drag wheel (for example) to move the pot's shaft in synchronicity with directed motion could work as a "pace counter" for distance travelled from some known "here I start" point. I can see that working out. I'm assuming here that a digital pot HAS a shaft for input. If it is digital input only, then I could use a position sensor based on a code wheel keyed to driven shaft as the input.

So where do you find those? Are they implemented on a DIP chip, or how? And who makes them? Any part numbers available? If not, I can find them, but if you have a favored model I'd like to know what, and why, to help me make a better selection.

Micah

You guys have gotten me totally messed up. Is it ok to use the phrase, "screwed up"? I think I need to start all over now. I wanted to raise a voltage value to an exponent. Another way to represent this is to write: I want to do Y^Z. Now, when I have gotten this done, I'd like to multiply this quantity by X. Now, I could represent this by writing: X*Y^Z. Right? Now, after reading your suggestions, I think I might do this by multiplying the log of Y by Z, [Z*logY], take the alog of that quantity: [alog(Z*logY)] and multiply that by X. This should look like X*alog(Z*logY). Right? If I'm wrong, maybe you'd better not tell me. I think my brain might go Pthpthpth!

I really do appreciate the time and thought you have all contributed to answering this (simple?) question. It is nice to know that CR4 is the source of real and available brilliance that has been shown in this thread.

Bob

Hamish:

Yes, I can see why you are confused, too many suggestions with misunderstood directions, unintentional I think.

I understood your equation. Use the exponential circuit in AN-222 to do the Y^Z function first, then feed the result to a multiplier, either a chip multiplier or a multiplier in AN-222. Just observe the input/output range limitations. Scaling the input will not cause problems as long as you remember the scale factor all the way through the circuits to the end result.

Okay?

Ed

Hi Electronis Wiz,

Finally, It looks like you have given me the answer!! Thanks Thanks Thanks.

Bob Clark

Bob:

You are very welcome. Don't forget to 'call' if you need anything else.

Ed

Ed,

Just one more question: Please, where can I find AN-222?

Bob

Bob:

Check out page 9 of AN-222 for the math functions, see fig. 13 for multiplier circuit and fig. 15 for the exponential circuit. If you think the multiplier circuit is a bit much, opt for a multiplier chip, Analog Devices has some low cost chips and are easy to get.

Thanks, Ed.

You've gotten me to where I need to be.

Thanks again,

Bob