A few months ago, I was on a large boat (smaller than a small ship) that was going out to Santa Cruz Island off the coast of southern California. Shortly after we left port, I was taking a few photographs, and I noticed that our wake showed some zigging and zagging. Several minutes later, when the zigging and zagging was still observable, I went up near the wheelhouse. I watched the steersman turn the wheel to the right as the boat was veering to the left. She held the wheel steady for a while, but after a minute, the boat began to veer to the right and the steersman cut the wheel to the left. Aha. The boat was trying to go off in random directions, and the steersman was just trying to react and correct for the steering error. I asked the master of the boat if the rudder linkage was really that loose. He replied, yes. So, when you’re trying to servo a loop with a big dead zone somewhere in the loop, you have to be pretty fast and smart to keep ahead of the game.
I was reminded of this recently when I borrowed an old car, and I found that I had to steer alternately far to the left and then far to the right when the wind blew from one side to another. But after just a few miles, I got the hang of it, and I realized that I really did know how to handle this—just a big dead zone. When the car began to drift off to the left, I immediately took up all of the slack and turned the wheel right. Anybody who has driven old cars knows how they can have 2, 3, 4, or maybe even 5 inches of “play” or dead zone, as measured at the rim of the steering wheel. So I knew how to drive this old car with good accuracy.
As I considered this, I thought back to the boat. If you wanted to steer that boat in a straight line, it would be easy. It was a fairly big boat with two engines, so it would be easy to throttle back the left engine a trifle, thus setting the boat to veer to the left. Then if you cocked the rudder to the right, you would keep the rudder working against the offset of power, and the slack would be taken up consistently.
Oh heck, this reminds me of a story in the old Lightening Empiricists, published by George Philbrick Researches about 30 years ago. Henry Paynter, one of the most authoritative experts on Analog Computers at MIT, wrote a little story to go into that journal. His point was that hysteresis was the natural antagonist of noise or oscillation in a slow servo loop. For example, if you drove your car, you could turn the steering wheel back and forth at a relatively high frequency. As it bumped against the active regions, alternately left and right, it would impart the desired force to turn the steering wheel in the correct direction when the need existed. So a constant amplitude oscillation could be used to virtually CANCEL OUT a dead zone of similar size.
NOW, this is exactly the converse of a case where you have a signal with some noise or oscillation superimposed. To avoid severe noise on the output of your controller-amplifier, you can add in some hysteresis (an alternate name for a dead zone), which cancels out the oscillation. Only when the signal-plus-noise exceeds the hysteresis or dead band does a significant error signal go out to correct the loop.
If you have a loop with a dead band or dead zone, can you control this with a linear servo loop? Not very well; in fact, it will work lousy (that’s a technical term) and the precision of the servo loop will be poor. YET, if you ask the dead zone and hysteresis to cancel each other out, the loop’s controllability can be improved considerably. Hey, not bad for analysis in the 1960s, with no assist from any digital computer…
In the real world, you often find dead zones in mechanical systems. When you turn the knob of a trim pot, the electrical change doesn’t begin until after some dead zone. When you turn it back, electrical changes don’t occur until there’s a further rotation in the other direction. HEY, but we read that multi-turn pots are supposed to have infinite resolution. If you turn the pot to the left and to the right with infinite mechanical resolution, can you get to any value you want?
Let’s suppose we have a 10-kΩ pot with a 1.000-V dc supply across it. Your task is to set the pot’s wiper to any voltage you’re asked. Your buddy says to set it to 0.307 V, and you try. But if the pot is working badly, you get to 0.304 V, and then with a tiny increase, it lurches up to 0.309 V. You try to turn the knob back, and it lurches past 0.308 V to 0.301 V. Hey, this is one of these multi-turn pots that’s supposed to have “infinite” resolution. But it’s not working with “infinite resolution.”
Now, you finally got the trim pot’s output to 0.307 V. Just then an ant tiptoes past the pot and it shifts to 0.305 V. So, even though bad settability is bad, poor stability is worse. And this is one of those 22-turn pots that claims “infinite resolution.”
Well, what if you evaluate a single-turn pot? As a matter of fact, most single-turn pots are better than most multi-turn pots for settability or for stability under shock and vibration. You can check this out yourself.
Now, first of all, any pot with settability better than 0.1% is pretty good. If you can set a pot to 1-mV resolution out of a 1-V full scale, that’s a pretty good pot. The best ones can do this fairly consistently, but if you need to set a multi-turn pot to do much better than 0.3%, you’re going to waste a lot of time. And after you have actually set the pot to the desired voltage and you rap the pot lightly with a pencil, its output is likely to move. Again, the multi-turn pot more often than not will give worse stability.
Okay, go back and re-tweak the pot and re-tap the board it’s mounted on. You can probably get most pots to read any millivolt you name, but the multi-turn pots take more work, more time. A smooth guy with a single-turn pot can drop the voltage quickly anywhere he wants to. It’s true that he has to set the angle of the pot to a resolution of 270/1000, which is a very small amount of rotation—25x smaller than setting a 22-turn pot. But, yes, in practice it’s easier to do.
NOW, in any system, it’s usually bad practice to require a pot to be set with a resolution as small as0.1% of span. You shouldn’t try to trim over an infinite range and still demand good resolution. But at least you should be aware of the problem of decent resolution.
So, hysteresis, dead zone, settability, and stability are all related, and the problems are, in practice, all mushed together. A multi-turn pot isn’t necessarily closer to “infinite” stability, notwithstanding anything you read on its data sheet. And smart people, or smart systems, can learn to take up some dead zone automatically. Back in 1964, Henry Paynter had some pretty good ideas on this subject.
All for now./Comments invited! RAP/Robert A. Pease
Address:
Mail Stop D2597A
National Semiconductor
P.O. Box 58090
Santa Clara, CA 95052-8090
