What Components Ensure Precision in Sensitive Analog Circuit Design?

[Topher925]

Design of precision circuitry?

I'm currently working on a circuit that is so sensitive that if I even so much as breath on it for just a moment, it becomes completely useless as my signal will become completely saturated. While the circuit works very well (when I'm not breathing on it), even slight variations in room temperature completely destroy its precision. Currently I'm just using generic ceramic resistors and capacitors which is the #1 source of my temperature drift error.

My problem is, that I know hardly anything about precision analog circuit design and what components I should be looking for as far as capacitors and resistors go. For example, I need a very low drift (<100ppm?) 10pf cap but I don't know what kind I should get. I used to think polystyrene was the best choice for precision stuff but they don't make it anymore. Now searching online I see some sources say C0G rated ceramic caps are the best and others say polystyrene carbonate is the best, which is it? As for resistors, there's thinfilm, metal oxides, ceramics, etc and I have no idea what to look for. Can anyone help me or give me a source of what types of components I should be looking for? My circuit is going to be inside of an enclosure in a laboratory type environment so I just need precision and low drift, not durability.

 

[f95toli]

Exactly HOW accurate should it be? And what is your budget?
Also, do you need absolute accuracy or is low drift enough?

Moreover, I assume that there are other considerations; e.g.wire-wound resistors tend to be very good for these kinds of applications but they are also inductive meaning they can't be used at high frequencies (unless their inductance is part of the circuit).

That said, the best solution to your problem is probably to simply regulate the temperature of the circuit (i.e. heat to 35 degrees C or so and keep it there); this is how most low-drift circuits are made.
You don't need very expensive equipment to do this; I've seen the temperature of metrology grade equipment being regulated using a simple temperature controller and some heating blankets...

 

[negitron]

Depending on the circuit, you'll probably some sort of temperature compensation on either a current source or a voltage source. I have a barometric pressure sensor which, while it has some internal compensation, still needs a TC current source for the greatest accuracy.

And who says they don't make poly caps anymore?

 

[Topher925]

I'm still working on simulations so I don't have exact numbers but less than 0.5% drift in a 10C temperature swing is a good ballpark number for resistor drift. I think (-)100ppm/C will be sufficient for the capacitors. I'd preferably like to keep the cost of each resistor below $10 each and the caps below $5 each. I don't really need accuracy, what I really need is repeatability. In other words, if the lab is 5 degrees warmer today, are my resistors still going to be close to the values they were at yesterday.

My signals operate at about 50Hz so its very low frequency and I'm working with nano-amps at the sensor end so I'll look into wire-wound resistors. I'm not sure I like the heating blanket idea but I'll take it under advisement. My circuit doesn't need to be an extremely accurate metrology grade device, it just needs to give me repeatable results.

 

[Topher925]

Depending on the circuit, you'll probably some sort of temperature compensation on either a current source or a voltage source. I have a barometric pressure sensor which, while it has some internal compensation, still needs a TC current source for the greatest accuracy.

And who says they don't make poly caps anymore?

I'm using some pretty good lab grade power supplies and the "laboratory television" tells me that they are very consistent.

As for the poly caps, I can't find polystyrene caps any where and from what I have read online manufacturers have stopped making them. They are no where to be found a digikey's, mouser's, or allied's websites.

 

[Phrak]

When you mentioned breathing on it, surface leakage came to mind. You might want to rule out any variations in parameters due to surface leakage as a function of humidity.

 

[Bob S]

If your signals are about 50 Hz, then you could use a low-pass dc feedback for stabilization. Also, you could look at what Analog Devices has. I think they have op amps that are in the range of 1 nanovolt per root Hz. Their ADA4898 looks pretty good:
http://www.analog.com/en/amplifiers...fiers-op-amps/ada4898-1/products/product.html

 

[negitron]

As for the poly caps, I can't find polystyrene caps any where and from what I have read online manufacturers have stopped making them.

They might be harder to find these days but they're still out there. These guys still make them. And this page list several others:

http://www.americancapacitor.com/
http://www.centechusa.com/welcome.htm
http://www.tecategroup.com/ti/ti2.htm
http://www.vitelelectronics.com/asc.htm
http://www.filmcapacitors.com/
http://www.richeycap.com/
http://www.southernelectronics.com/
http://www.wescocap.com/
http://www.crcfilm.com/
http://www.seacorinc.com/
http://www.lcr-inc.com/products/polystyrene.htm
http://www.eci-capacitors.com/
http://www.evox-rifa.com/
http://www.electrocube.com/
http://www.seasonshk.com/
http://www.am-21.com
http://www.susco.com
http://www.rgaco.com
http://www.rtie.com
http://www.tscgroup.com/
http://www.eurofarad.com
http://www.capacitors.com.hk

 

[Topher925]

If your signals are about 50 Hz, then you could use a low-pass dc feedback for stabilization. Also, you could look at what Analog Devices has. I think they have op amps that are in the range of 1 nanovolt per root Hz. Their ADA4898 looks pretty good:
http://www.analog.com/en/amplifiers...fiers-op-amps/ada4898-1/products/product.html

Bob, I'm a little confused about how the low-pass dc feedback works for my application? The main source of drift in my circuit from a AD549 setup in a current feedback mode using a shunted 10M resistor (generic carbon film) in which I have about 50 pA going through it. How can a 4898 reduce this drift?

Thanks for the list Negitron. I was checking out American Capacitor earlier and am going to order some samples from them on Monday. I couldn't find any prices or minimum order requirements on their website though which kind of worries me.

 

[negitron]

I couldn't find any prices or minimum order requirements on their website though which kind of worries me.

I find that often--not always, but often--you can get around minimum order requirements if you tell them you want to purchase some evaluation samples and give them the impression that a larger order could follow. Not that I would do such a thing, of course...

Also, I see them listed on eBay, so you might want to look there, too.

 

[f95toli]

I'm still working on simulations so I don't have exact numbers but less than 0.5% drift in a 10C temperature swing is a good ballpark number for resistor drift. I think (-)100ppm/C will be sufficient for the capacitors. I'd preferably like to keep the cost of each resistor below $10 each and the caps below $5 each. I don't really need accuracy, what I really need is repeatability. In other words, if the lab is 5 degrees warmer today, are my resistors still going to be close to the values they were at yesterday.

That shouldn't be very difficult to achieve (0.5% drift is a huge number in metrology).
E.g. Vishay makes some very nice foil resistors with a drift of something like 0.01 ppm/ degress C.
I think I've also seen someone using precision caps from Vishay in one of the labs where I work; unfortunately I don't remember the type (and they are big and probably expensive).

But; I wasn't kidding about simply heating your circuit a bit which is probably the easiest (and cheapest) solution. Most books about precision electronics will contain a description of a suitable circuit. All you need to do is to put your circuit in an insulated box and then build a simple regulator with e.g. a thermistor and a resistive heater; there are plenty of cheap controllers around (or you could build a PI regulator using a few op-amps).
A good regulator should be able to stabilize the temperature to within a few tens of mK regardless of the external temperature if done right.

If you don't want to build something you can always buy a suitable box. Where I work we even have 19" racks where the temperature is stabilized this way (even though our labs are suppose to have fairly good temperature stability).
It can make a huge difference to the long time stability of sensitive electronics.

Edit: Found this via Google; probably too sophisticated but it should give you some idea of what I mean (it is a Master's thesis)
http://epubl.luth.se/1402-1617/2006/273/LTU-EX-06273-SE.pdf

 

[dlgoff]

I agree with f95toli and Phrak. If you could find a small temperature and humidity controlled cabnet/chamber your problem is solved (hopefully).

Edit: Probably too expensive but here's some: http://www.labx.com/v2/newad.cfm?catId=100"

 

[Bob S]

Bob, I'm a little confused about how the low-pass dc feedback works for my application? The main source of drift in my circuit from a AD549 setup in a current feedback mode using a shunted 10M resistor (generic carbon film) in which I have about 50 pA going through it. How can a 4898 reduce this drift?.

Tell us a little more about the signals you are trying to amplify, and the circuit you are using. Is your signal source a current or voltage source? If your signals are ac, and you are dc coupled, that could be part of the problem. Also, a 1 meg resistor fed back thru a 1 uF shunt capacitor and then to the inverting input thru another 1 meg resistor would get rid of dc drift (e.g., temperature drift), and any sub 1-Hz ac signals.

 

[vk6kro]

You could look for a 10pF ceramic capacitor with the marking NPO.
These are used for precision oscillators (like VFOs in radio equipment).

NPO means low temperature coefficient. Maybe it stands for negative-positive-zero but that is one way to remember it.

There are ways of controlling a temperature so that it doesn't vary much. The classic is the crystal oven where the temperature inside a box is controlled to about 45 degrees C by having a heater and a temperature control setup. It has to be higher than any temperature you might encounter in the room as it has no way of cooling if the room is hotter than it is.

A better way now is to use a Peltier device which can control temperature by heating or cooling. So, the device can be run near normal room temperatures. These are quite small (about 2 inches by 2 inches), so they are best suited for applications where the critical components are also quite small.

 

[Bob S]

Bob, I'm a little confused about how the low-pass dc feedback works for my application? The main source of drift in my circuit from a AD549 setup in a current feedback mode using a shunted 10M resistor (generic carbon film) in which I have about 50 pA going through it. How can a 4898 reduce this drift?.

Topher
Here is a transimpedance amplifier with a gain of 10 megohms (10 millivolts per nanoamp) (see thumbnail) that stabilizes the DC drift with feedback. I decided to use a second op amp to do the stabilizing. The passband is from about 10 Hz to 200 Hz. The passband limits can be adjusted by changing resistors or capacitors.
Bob S

 

[Bob S]

Here is another transimpedance amplifier similar to the one above, except that I have added a 1000:1 voltage amplifier on the end to give an overall gain of 10¹⁰ ohms (10 volts out for 1 nanoamp in). The bandpass is unchanged. There probably should be a pot (not shown) on the afterburner to adjust the dc quiescent voltage. As before, there is a feedback circuit on the first amplifier to zero the dc output. This is a LOT of gain, so your EM shielding has to be very good. You should consider bypassing your +/- 15 volts right where it comes through the case. Use feedthroughs if possible. You could also reduce the gain of the first amp by a factor of 10 to get rid of the 10 meg resistor.

 

[Carl Pugh]

0.5% drift for 10 deg temperature swing is nothing.
If there are no other requirements, this should be easy to do.

Use metal film resistors and NPO/COG capacitors.

Consider putting the circuit in a metal container to prevent esectrostatic coupling.

50 hertz is close to 60 hertz, if you can operate at higher or lower frequencies, it would be easier to filter the 60 hertz out.

BREATHING ON A CIRCUIT IS A GOOD TEST FOR CONTAMINATION.
Some solder flux is highly conductive and if you breathe on it the resistance drops dramatically.
Use new/clean water or solvent to clean your printed circuit board.

 

[Topher925]

Thanks for all the replies, they really help. I've been looking for some low TCR and instrument grade components but having some difficulty finding them in the higher values I require. However, I've found some precision resistors with a TCR of +/- 50ppm/C and tolerance of 1%. Would this be considered instrument grade? I also found some praylene caps on mouser as well that looks like they will be perfect. I just have to be cautious when soldering them to the PCB.

Bob S, thanks for the schematics. My circuit is similar to the second one you posted except for my second op-amp is in series and not cascaded. I couldn't get my pre-amp stable in that configuration. Also, I don't have the cap in series with my photodiode or the 10M resistor on the inverted input. The cap (or even a jumper) causes to much distortion in my signal and the 10M resistor adds to much noise.

One thing you had mentioned was including a pot to adjust the DC offset. I originally had a pot on my circuit for this purpose but switched it with fixed value resistors. I'm wondering if it would be better to have a pot that I would "zero-out" every time I use the circuit or if I should have fixed value resistors instead. The pot would probably be ideal and theoretically it shouldn't affect the calibration if its always zeroed out but I don't know how it will affect the repeatability of the circuit in practice.

 

[Bob S]

Bob S, thanks for the schematics. My circuit is similar to the second one you posted except for my second op-amp is in series and not cascaded. I couldn't get my pre-amp stable in that configuration. Also, I don't have the cap in series with my photodiode or the 10M resistor on the inverted input. The cap (or even a jumper) causes to much distortion in my signal and the 10M resistor adds to much noise. .

I am not sure what you mean by op amp in series and not cascaded. Having the cap in series with photodiode blocks all dc current. Maybe you want to amplify dc current also? Or is your signal 50 Hz or 60 Hz only? The 10M or similar resistor on inverting input is required to convert input current signal to a voltage output signal. The extra feedback circuit is there to block all low frequency signals (under 1 Hz). What is your circuit?

One thing you had mentioned was including a pot to adjust the DC offset. I originally had a pot on my circuit for this purpose but switched it with fixed value resistors. I'm wondering if it would be better to have a pot that I would "zero-out" every time I use the circuit or if I should have fixed value resistors instead. The pot would probably be ideal and theoretically it shouldn't affect the calibration if its always zeroed out but I don't know how it will affect the repeatability of the circuit in practice.

To limit noise, the whole circuit, including the pot on the afterburner, should be inside a shielded metal enclosure. Once you set it, the drift in the afterburner should be small. The active feedback loop on the main amplifier should be adequate to control dc drift on main amplifier.

Thanks for mentioning that your signal come from a photodiode. Is this a vacuum photodiode, or a solid state photodiode? solid state photodiodes are more likely to have leakage currents, and maybe Schottky (diode) noise currents also.
[Edit] I reran the Spice simulation with the series capacitor on the current source (photodiode) removed, and with a shunt 100 pF capacitor on the current source. No change on the simulation ac response shown on earlier post.

 

[Topher925]

Maybe you want to amplify dc current also? Or is your signal 50 Hz or 60 Hz only? The 10M or similar resistor on inverting input is required to convert input current signal to a voltage output signal. The extra feedback circuit is there to block all low frequency signals (under 1 Hz). What is your circuit?

Thanks for mentioning that your signal come from a photodiode. Is this a vacuum photodiode, or a solid state photodiode? solid state photodiodes are more likely to have leakage currents, and maybe Schottky (diode) noise currents also.
[Edit] I reran the Spice simulation with the series capacitor on the current source (photodiode) removed, and with a shunt 100 pF capacitor on the current source. No change on the simulation ac response shown on earlier post.

Sorry such a late reply, I've been rather distracted lately. I'm actually measuring a DC signal, not AC (sort of square wave, no negative voltage). So having that cap in series with the photodiode causes a bit of a problem. Also, being as I need to measure the current output of the photodiode, any slight change in capacitance or stray leakage of that cap would cause some repeatability issues I would think.

I'm using a solid state photodiode and it is my main source of noise. My circuit is about 40-50Hz, but any change in its attenuation of the signal is unacceptable. Currently, the circuit has no EMI shielding except the steel plate in which the breadboard is mounted on and the noise is at acceptable levels. If I include any more caps or resistors like a low pass filter the noise level increases, especially the high frequency (1KHz up).

I got my high tolerance and low TC components in and installed on the breadboard. This includes 10pf polypreylene caps and somewhat large metal film resistors. I've tested the circuit several times over the past few days and I have found there is still a very small amount of drift. Its difficult to tell what's its actually from the circuit, the power supplies, or the breadboard. The power supplies I'm using like to wonder in a +/-100mV range and I'm going to assume I'm going to have leakage in my breadboard since I'm dealing with high impedences and low currents. All of the photo-detector circuits that I have seen online that use solid state photodiodes say to mount the photodiode, feedback resistor, and cap on teflon terminals to minimize stray leakage. Are PTFE PCBs typically a lot more expensive than the usual FR-4 boards?

I've been taking apart all the instrumentation in the lab lately and studying it. Fortunetly some of it is rather old (10 yrs or so) so it all still mostly uses through hole design. Online, I've read a lot of people say that for best noise reduction you should make your ground traces cover the entire one side of the PCB. I have yet to see an actual designs that does this. Other than that, my circuit design has similar qualities to the instruments I have taken apart in that it uses high tolerance components, keeps analog and digital as separate as possible, and uses as wide of traces as possible. I just can't get a strait answer about what type of grounding is best, star or polygon?