Back-to-back Capacitors

[Bob Scott]

Does anyone have any information about connecting electrolytic caps back to bck in order to use them as non-polarised caps? I've seen it before in amateur magazines years ago but I'd like a pro opinion.<p>I need some SMALL large value caps in an op-amp circuit in order to create a time constant of almost a minute. I bought a selection of mylars with values from 1 uF to 15uF this morning but they are huge. Digital circuitry and counters are out of the question. This is linear, and I want to keep resistor values down to a meg or less.<p>Thanks,
Bob

[russlk]

I assume you are using the op amp as an integrator, with the capacitor connected from output to input. If the output is always positive relative to the input, a polarized cap will work, but if the output goes negative relative to the input, the cap should be non-polar. The two polarized caps can be connected either way, + to + or - to -.

[Chris Foley]

Hi, Bob. Placing two polarized electrolytics back-to-back can lead to reliability problems. It's been done, but the caps generally don't like it, and your actual combined impedance starts to look like an electrolytic in series with a fairly mushy voltage- and temperature-dependent resistor. Not much precision there. Years ago I worked for a small industrial controls company that used back-to-back electrolytics for an op amp integrator in a potted control circuit. The circuit had some reliability issues, but the most annoying thing was that they all acted differently, and tended to change behavior once they were installed in the customer machine and warmed up.<p>Look at your problem again. With a smaller mylar cap (.1 uF), you've got a lot less charge there, but a whole lot less leakage current, and you have to be a little more careful keeping it where it belongs. You also have a greatly reduced "memory effect" on the cap, which is the tendency after quick discharge of creeping back toward its prior voltage. You're actually a lot better off with a mylar than you would have been with a larger non-polarized electrolytic as far as precision. <p>First, look for a good I(b) op amp for the junction. A plain op amp will steal a whole lot of charge to feed its input bias current, which will mess up your reading. If I've got bipolar supplies, I usually use a National LF411/412 for run-of-the-mill integrators -- inexpensive, low I(b) and low drift, especially for varying temps. <p>From your description of the problem, you'll have to bump up your resistance into the 10 to 50 meg range. Individual carbon films don't go higher than 22 Meg, and have a few problems in this resistance/current/frequency range, but mostly with leakage current across the body of the resistor. Metal films are better, but they don't go that high, and you'll need too many of them in series. Use several 1/2 or 1 watt (greater surface area to reduce leakage) carbon films of 10 Meg or less in series to get your total resistance, and keep them physically separated (don't just loop them back and forth in a zig-zag with the res bodies touching, and the solder joints right next to each other).<p>Once you've done that, make sure the board and the components on the board (at least around the op amp) are really clean. Use solvents to eliminate all flux from the board, and then use good ol' Palmolive green dishwashing detergent in water with an acid brush to clean. Shake off excess water, and follow by a gentle 130 - 140 degree F dry cycle in the oven overnight. Most gas home ovens have this temp with just the pilot light on. If you have any pots or other components that can be damaged by water, solder them in after you do this, and spot-clean separately. Wear gloves handling the board after cleaning, and you might want to put on a conformal coating or acryllic spray if the board is going to be in a dusty/humid environment. This process seems complicated, but it saves a lot of headaches.<p>All of the above may seem like overkill, but the techniques are effective for more severe problems than yours, too. You're just one order of magnitude into the "headache zone" of leakage current problems -- this part of the territory is the easiest. Feel free to email if you have any problems.<p>Good luck.
Chris<p>[ September 18, 2003: Message edited by: Chris Foley ]</p>

[Edd]

Sir Bob:
<I need some SMALL large value caps in an op-amp circuit in order to create a time constant of almost a minute. ><p>Since you seem to refer to a timing action , would it be a travesty to suggest the utilization of monolithic block ceramic capacitors (layered-small footprint- hi cap) in that function. Thereby pulling your R value back down into an acquirable value. Since you would be typically at the 25VDC range or lower ? With no Q factors involved as per RF utilization and probably around the ~20deg C operational range to be experienced.
I had fine results just using CONVENTIONAL ceramics in discrete xstr phase shift oscillators circuits for sub AF tone squelch injection/modulation into xmitters back in the 60’s. The keying into the sub osc threshold of a Twin –T circuit on the receive end then de activated the squelch of the receiver. And then along came Motorola… and you know the rest of the story.<p>73's de Edd
[email protected] ........(Interstellar~~~~Warp~~~Speed)
[email protected]........(Firewalled-Spam*Cookies*Crumbs)

[Chris Foley]

A bit of a math error on the last post -- a tau R*C time constant of 60 seconds with a 60 Meg resistor (6 X 10M, 1W in series) would give you a C of 1 uF, not 0.1 uF. You might want bump up the resistor to 110 M (11 ea. 10 Meg 1W carbon film in series) and halve the cap to 0.56 uF. Mouser 5989-100V.56, which is a Miniature Metallized Polyester Film Cap, has the following dimensions:<p>Width 0.472"
Height = 0.429"
Thickness = 0.217"<p>It's $0.43 USD in unit quantities.<p>Possibly we could give you a little more help if you told us what you're doing? I'm not sure (and I don't think Edd is, either )that this is the best way to get from point a to point b. I've occasionally had good luck with the "headache zone", mostly through reading app notes and Robert Pease articles for practical advice.<p>You might want to look at these Pease Porridge articles:<p>What's All This Tee Network Stuff, Anyhow?
This article shows you how, with a little math, you can replace the mongo 110 Meg resistor with some more reasonable values by using a Tee network (Thanks, Edd! You're right, of course. However, the business about not using high value resistors bothered me -- that attitude deserves attention, too, and there's no way to pass schematics on this BB. A 100 Meg resistor isn't wrong -- it's just an unusual way to get from point a to point b. A Tee network still requires all the precautions for a low I(b) op amp, leakage current, memory effect, and board cleanliness, because you still don't want to lose the charge on the cap.)<p>Understand Capacitor Soakage To Optimize Analog Systems
Why you _don't_ want an electrolytic here.<p>Good luck.
Chris<p>[ September 18, 2003: Message edited by: Chris Foley ]</p>

[rosborne]

Awesome links Chris. Thanks.
-Rick

[Bob Scott]

Hi Guys. Sorry about the delay. This is a new design for temperature regulation with PWM.<p>Yes it is an op-amp integrator. It is also "sort of" a timing circuit. The LM660 op-amp compares 2 voltages, the output of an LM35 temperature sensor pre-amplified 10X with a potentiometer voltage settting for the other voltage. The pot sets the temperature that you want. The integrator is used to generate an error signal that will be used to change the signal input of a PWM stage which controls a power MOSFET output. The FET controls the voltage to a Peltier device used for cooling. The Peltier device cool side has a metal block attached to it with the LM35 mounted on it.<p>The problem I can forsee is the time delay in change of temperature reaching the LM35 may cause the integrator output to overshoot or at least cause the integrator output to quickly rise (or drop) and clip at the rail. So there is indeed a "timing" aspect. I don't want the output to vary too much. I can see raising the resistor value or increasing the cap size in order to lower the gain of the stage.<p>I won't go over 5 megs period on the resistor. I've seen too many circuit boards develop internal leaking, causing problems with high impedance circuitry. Mind you they were cheap Japanese phenolic-paper boards.<p>Thanks for all your responses. I will be building and tweaking this circuit all week in Calgary.<p>Bob<p>[ September 21, 2003: Message edited by: Bob Scott ]</p>

[Chris Foley]

Hi, Bob. There are a number of good sources on the web about control loops, and particularly PID (proportional - integral - derivative) loops, which I think would probably help you. From what you're saying, it may be that having a very long time period may not help you achieve stable temperature, but will actually tend to cause oscillations. Just google "PID tutorial" for some resources. One I found with an animated Java-based do-it-yourself PID loop is<p>www.clabberhead.com/pidtutorial.html
PID Tutorial<p>Also, one of our contributors, Will, wrote a tutorial paper on PID loops he might be willing to share, if you e-mail him and ask. I've looked at it -- it's good.

[rshayes]

You may not need an integrator in the feedback loop at all. Usually a control loop has high gain at DC. It should cross unity gain at a rate inversely proportional to frequency to avoid excessive phase shift and oscillation. In between, it may fall at a higher rate.<p>One way, probably the simplest, of designing a feed back loop is to have one dominant pole that reduces the gain to unity before any other poles have a significant effect In the case of temperature control systems, there is usually a thermal time constant that causes a pole at a fairly low frequency. This time constant can be tens of seconds to possibly several minutes. Using the dominant pole approach with an additional integrator would require an integrator with a time constant of minutes to hours, and would result in a slow system.<p>The thermal time constant can be used as the dominant pole. In this case, the feedback amplifier will need either flat response or response that rises with frequency.<p>A rough idea of the thermal time constant can be obtained by applying power and heating the system up. Cut power and observe the temperature as the system cools down. This should follow a decaying exponential curve from which the time constant can be estimated.<p>If you don't need high accuracy, use moderate gain in the amplifier. This is the simplest to set up.<p>Note that Peltier devices tend to be nonlinear, so a fair amount of experimenting may be necessary. I would expect different behavior when the system is heating. The transient behavior on heating will porbably be different from that on cooling. A compromise will probably be necessary to get acceptable performance in both directions.

[Bob Scott]

All,<p>All your replies are appreciated. I used to work on VTR servos for Sony many years ago. In those days the error signal went through a low pass filter before it was amplified to the control circuitry that drove the motors. I believe technology had progressed, and engineers realized that although the LPF worked, what was really needed was an integrator. An integrator has no corner frequency but you can control its output by varying the gain.<p>I had given this design a lot of thought over the last couple of weeks. Stephen took less time to come up with some of the ideas that I did also. They include the existence of 2 thermal time constants and the need to keep gain levels low.<p>I can see 2 ways to do this project also. First way, just compare the desired temp with the LM35 output and use the resulting error signal to directly drive the MOSFET. This way there is no integrator and there will always be a slight error in temperature. An error signal in this sytem is necessary to be amplified into the drive signal.<p>Second way adds the integrator so that the error simply builds a charge on a cap to send to the output amp or PWM that controls the FET. No error, but if the temp changes, the servo's output may overshoot with sudden changes in error signal. I'm going ro play with this type tomorrow. The board is all assembled and gains are adjustable. I just have to recheck my wiring and go. <p>This subject brings up another third way to make the servo that I heard about from an ME who actually teaches this stuff. That is: Add a third integrator. The mind boggles but I think it meant that if the first integrator got rid of the error, then maybe the second gets rid of the overshoot leaving resultant products I don't understand any better than second order calculus. I mean I can understand velocity being an integral of acceleration, and I can understand position being an integral of velocity, but his knowlege fails to give me a grasp of the problem at hand.<p>I think I'll just stick with one for tomorrow.<p>Thanks again. This is interesting stuff.
Still posting from Calgary, Stampede City,<p>Bob <p>[ September 22, 2003: Message edited by: Bob Scott ]<p>[ September 22, 2003: Message edited by: Bob Scott ]</p>

[Will]

Bob,
What you are playing with certainly sounds interesting and if you are having fun with the way you are doing it then that may be the ultimate goal and reward. I can't understand why you would simply want to control the surface of a Peltier device as distinct from using the Peltier device to heat or cool some process load or something else but again, if you are having fun then perhaps that's all you need.
If, on the other hand, you are trying to develop a practical means of temperature control then you are more or less pushing at an open door because PID (Proportional, Integral, Derivative) controls are almost universally used to control anything and everything. The Proportional and Integral terms are relatively easy to understand and essential to almost all control applications. The Derivative is not quite so easy to understand or to apply and is mostly unnecessary.
I have been meaning for years to make up a PID controller from Op Amps but, although I have a good understanding of PID and controls generally, I am not very good at designing and building electronics. As a concept however it is fairly simple. You could build five modules (I'm certain that lots of readers will figure out how to do it in less than five) each using an Op Amp i.e. (1) To measure and buffer the control error 'E' (The difference between Measuired and Desired (Setpoint) Values MV and DV)
(2) The Proportional module - A device which accepts the buffered error signal as an input and outputs a gain adjustable value proportional to the input. For a universal controller the gain should be adjustable in the r ange about 0.1 to 50
(3) The Integral (An integrator) module whereby the error signal is integrated with time. This should have an output (Calibrated in minutes) such that, at any time setting 'T' It will take "T" minutes for the Integrator to increase(Decrease) the output by an amount equal to the magnitude of the gain adjusted error signal. The time "T" should be adjustable over the range about 0.01 to 30 minutes) (This term is nessary to prevent/negate the small error with which you said you would have to live.
(4) The Derivative module (A differentiator) - this would produce an output which is in direct (Adjustable) proportion to the Rate of Change of the Error (The first derivative dE/dt) The output would be such that would equal the amount by which the error changed in the set time 'T'. (Which should be adjustable in the range about 0.01 to 30 minutes.) Unless you are trying to build a uuniversal device then I don't believe you need this term for your application.
(5) A summing module which would add the outputs of (2), (3) and (4) to produce a single control output. In your case the calculated pulse width - in industry generally a 4 - 20 mA control signal.
It is possihle that Ed's stuff about poles may achieve he same in a different way - I don't know because, although I wish I did, I don't understand it (The general stuff about poles I mean)
If you connect two electrolytic capacitors back to back doesn't that mean that one of them is connected the wrong way round on DC, thus allowing an electrolyzing current to flow ? and both of them on AC ?
I will be happy to send you a copy of my paper on PID controls (As kindly mentioned by Chris Foley) if you will just post your E mail address

[Bob Scott]

Hi Will,<p>This project is fun but not just for fun. By looking at your post I can see that I have developed parts 1,2,3 of the sytem you describe including the gain adjustment part. I don't have the differentiator or the summing network.<p>I finished the construction of the prototype this morning, had the errors and omissions in wiring fixed by 3 and hooked up and tweaked by 6 PM. It works!<p>The Peltier device cools 200 Watt worth of resistor load and keeps the temperature of the cool side within +/-0.1 degree C of the desired setting once it settles down.<p>Yes now that I have something that works I can breath easier and I have time to peruse the technology that you describe. I only have until the end of the week to complete this project in Calgary. No, it is not an educational assignment or related to my day job. We're a bunch of geeks working on a deadline trying to develop a cooling system. I wish to gratefully accept your email. My account is so well screened, no spam ever gets in. I will add your email address to my list of email addresses that do not get screened out. Please send to [email protected]. I am assuming your address is the one with the "ev1" in it.<p>I mo onger need info on the back-to-back electrolytics although I am still interested in answers. The cap I am using is now a 0.1uF film. This value is sufficient. The idea of back-to-back electrolytics is that leakage is prevented by the partner cap which will not conduct the leakage current. When working on AC, the value of the point in the middle (where both - sides of the caps are connected together) pumps down to a low voltage due to leakage with the amplitude of the AC voltage and stays at that voltage. Then the caps work like a non-polar cap. At least that's what I understand. I was trolling for ideas about drawbacks to the method. I've seen amateurs use this method over the decades, but never industry.<p>Much Thanks,
Bob<p>[ September 23, 2003: Message edited by: Bob Scott ]</p>

[dyarker]

I just saw a sentence in a Mouser catalog that said a particular model of tantalum caps worked well for that kind of job. I wouldn't try it with aluminum caps.<p>[ September 24, 2003: Message edited by: Dale Y ]</p>