I'm using a straightforward instrumentation opamp arrangement (three discrete OPA134) in a floating audio buffer, AC coupled input, fairly high impedence (~1M). unity to high gain (1-1000x). The performance thusfar is excellent and slightly exceeding hopes.
Conventional wisdom indicates I want a simple common mode/differential mode RF filter on the front. However, the OPA134 only has a gain bandwidth product of 8MHz to begin with and following the generic 'rules of thumb' about the filter settings (DM filter corner at 10% of GBP, CM caps < 10% of the DM cap value), that already solidly cuts into the audio band.
Due to the fairly low GBP of this opamp... are the filter caps actually needed at all?
The application is actual test & measurement (an instrumentation op amp... being used in instrumentation!
so the input should not be assumed to be controlled or controllable. Dirty signals shall abound. I'm concerned only about the theoretical RFI filter in this post; there's considerable additional active input protection in the design and I'm assuming it to be completely orthogonal for the time being.
Thanks!
Conventional wisdom indicates I want a simple common mode/differential mode RF filter on the front. However, the OPA134 only has a gain bandwidth product of 8MHz to begin with and following the generic 'rules of thumb' about the filter settings (DM filter corner at 10% of GBP, CM caps < 10% of the DM cap value), that already solidly cuts into the audio band.
Due to the fairly low GBP of this opamp... are the filter caps actually needed at all?
The application is actual test & measurement (an instrumentation op amp... being used in instrumentation!
so the input should not be assumed to be controlled or controllable. Dirty signals shall abound. I'm concerned only about the theoretical RFI filter in this post; there's considerable additional active input protection in the design and I'm assuming it to be completely orthogonal for the time being.Thanks!
Yeah, you probably need RF filtering, and probably not just on the signal inputs.
"The overall input filter bandwidth should be at least 100 times the input signal bandwidth." See Chapter 7 of "Op Amp Applications Handbook" (Walt Jung, Editor Emeritus), which should still be on line, free, at analog.com. (They also describe how to actually test for RFI effects, using a signal generator, which should be able to tell you whether or not you need the input filter!) You'll probably also be interested in Chapter 2 (and some of the others no doubt!). Here's the link:
ADI - Analog Dialogue | Op Amp Applications Handbook
You'll probably also want an RCR "T" filter on the output, which is yet-another input for RF.
Use low-value metal film resistors, for lowest noise, preferably 10k or less (and well matched 1% or 0.1% ones, not to mention well-matched capacitors) for the input filter and something on the order of 100 Ohms for the RCR "T" output filter. Keep the filter components as close as possible to the input, output, or power pin they are supposed to protect, and, in the case of the CM/DM input filter, lay them out symmetrically, over a large ground plane if possible.
You might also want to add a small resistance (probably less than 100 Ohms; maybe 33 Ohms; or maybe a ferrite bead; you'll have to test it) just upstream from each set of power supply bypass caps, i.e. for each power supply pin, since chips' power pins are also RF inputs.
Depending on the application, you might also want to consider using coaxial cables for the input and output signals.
"The overall input filter bandwidth should be at least 100 times the input signal bandwidth." See Chapter 7 of "Op Amp Applications Handbook" (Walt Jung, Editor Emeritus), which should still be on line, free, at analog.com. (They also describe how to actually test for RFI effects, using a signal generator, which should be able to tell you whether or not you need the input filter!) You'll probably also be interested in Chapter 2 (and some of the others no doubt!). Here's the link:
ADI - Analog Dialogue | Op Amp Applications Handbook
You'll probably also want an RCR "T" filter on the output, which is yet-another input for RF.
Use low-value metal film resistors, for lowest noise, preferably 10k or less (and well matched 1% or 0.1% ones, not to mention well-matched capacitors) for the input filter and something on the order of 100 Ohms for the RCR "T" output filter. Keep the filter components as close as possible to the input, output, or power pin they are supposed to protect, and, in the case of the CM/DM input filter, lay them out symmetrically, over a large ground plane if possible.
You might also want to add a small resistance (probably less than 100 Ohms; maybe 33 Ohms; or maybe a ferrite bead; you'll have to test it) just upstream from each set of power supply bypass caps, i.e. for each power supply pin, since chips' power pins are also RF inputs.
Depending on the application, you might also want to consider using coaxial cables for the input and output signals.
Last edited:
Hi!
In fact, I'm building two buffer variants, one for input to an A/D, one for output from a D/A. The output from the D/A buffer circuit will in fact be exposed to the outside world via test leads, varying lengths of wire, old coathangers or perhaps even small bits of conductive rubbish.
I'm concentrating on the input buffer first tho, and in this case it is both battery powered and shielded (fully inside a metal box). The only thing exposed outside the shielding is the input leads. Batteries, converter, etc are all inside the metal box. Would I still need to worry about the power supply pins and buffer output in this case? I know I've set it up as something of a stupid question, but occasionally one can learn from asking stupid questions, and I'm not certain I know the answer.
Thanks!
In fact, I'm building two buffer variants, one for input to an A/D, one for output from a D/A. The output from the D/A buffer circuit will in fact be exposed to the outside world via test leads, varying lengths of wire, old coathangers or perhaps even small bits of conductive rubbish.
I'm concentrating on the input buffer first tho, and in this case it is both battery powered and shielded (fully inside a metal box). The only thing exposed outside the shielding is the input leads. Batteries, converter, etc are all inside the metal box. Would I still need to worry about the power supply pins and buffer output in this case? I know I've set it up as something of a stupid question, but occasionally one can learn from asking stupid questions, and I'm not certain I know the answer.
I can't think of any compelling reason, offhand, for RF filtering between things that are inside the box.
Thanks. I was wondering if, perhaps, opamp interactions within the box were somehow going to require RF decoupling on the power supply pins.
I'll be borrowing an RF signal generator to test out the circuit susceptibility to RF input.
Oh, I forgot to think about that. Sorry.
It depends, somewhat, on what all is inside the box. But I think that you will probably still want to have something like the standard 10uF electrolytic and 0.1uF ceramic very close to each opamp power pin.
Those aren't just for filtering. They're each sort-of like a point-of-load power supply, meeting surge- and pulse-type demands for current that the inductance of the PS wires might otherwise affect, which also helps to keep the power for the other chips smoother, because of things like V=L(di/dt) voltages that would get induced across the PS wires or traces.
Along those same lines, you will also want to design your ground return system carefully. That's also covered in the Jung-edited book that's downloadable at the link I gave previously, as well as in, among other places, the classic, must-read Application Note, AN-202, "An IC Amplifier User’s Guide to Decoupling, Grounding, and Making Things Go Right for a Change", which is available as a free PDF download, at Analog Devices, Inc. | Converters Amplifiers Processors MEMS A/D Converters Analog to Digital Video Converter Temperature Sensors Analog Device RF Amplifiers Differential Amplifiers Digital Signal Processing Thermal Management D to A Converters Micro.
Tom
Ah, part of the unstated design assumptions: this is effectively a micropower application with purposely bandwidth limited gain/buffer blocks and thus relatively low slew rates using OPA134 opamps and battery power. TI/BB's own application guidelines reflexively slap a 10nF decouple on all the power pins and its not clear if they mean it as a double-duty RFI/power decouple, or if it's merely an RFI decouple.
I'll be using some high-current output buffers elsewhere, for those, absolutely power decoupling (10uF/.1uF).
I was asking more along the lines 'are the decoupling caps useful for additional stability?' and 'do they help prevent cross coupling through the supply?'
I spend alot of time thinking about grounds, and I actually have a beef with the listed references: They don't go nearly far enough in giving practical examples. Most of the circuits omit any non-logical handling of grounding issues such that it's generally impossible to connect two of their own examples without ground loops all over the place! Ten rules of thumb are useless when these guides all end with 'but you'll have to try out a bunch of stuff and just use what works!! LOL!!!' Thanks guys for setting that straight for me
I'm well aware most equipment fudges the issue in horrifying ways, how about a few (10-15) concrete examples that don't? Sorry, this one always gets me ranting.In any case, the design in progress has to interface with other equipment at 10-15 points and so its easier to keep every block balanced/floating, even the interconnects between blocks on the same board, and star ground it all. The only single-ended signal paths are the external ones coming in/going out to consumer equipment, and even then I'm floating the send/return 'ground' as an unbalanced REF. Sometimes, brute force is the answer
Thanks for the time and answers!
Last edited:
Yup. Differential is often the best way to go, if you can. And star grounding used to be kind-of an obsession, for me.
I would probably still decouple the power pins, unless the opamps aren't doing anything (and probably even then).
The article at the link below has what appear to be some more-concrete examples for layouts. I haven't studied it and don't know if it has a lot that's applicable to what you're doing. But it looks like it might be helpful, for someone, some time, at least.
http://www.analog.com/library/analogDialogue/archives/39-09/layout.pdf
And you have probably already seen AN-539, from analog.com. But, if you haven't, it's about in-amp errors and error budgets (which all might also have been covered in the Jung-edited book that I linked to, earlier); incredibly boring stuff.
I would probably still decouple the power pins, unless the opamps aren't doing anything (and probably even then).
The article at the link below has what appear to be some more-concrete examples for layouts. I haven't studied it and don't know if it has a lot that's applicable to what you're doing. But it looks like it might be helpful, for someone, some time, at least.
http://www.analog.com/library/analogDialogue/archives/39-09/layout.pdf
And you have probably already seen AN-539, from analog.com. But, if you haven't, it's about in-amp errors and error budgets (which all might also have been covered in the Jung-edited book that I linked to, earlier); incredibly boring stuff.
Last edited:
RF rectification: DC offset only, or full demodulation?
[having now spent lots of modelling and experiment time with opamps and strong RF inputs...]
One more 'stupid' question. The Analog design documents by Jung mention repeatedly that the input stage BJT/JFETs of an opamp have a tendency to rectify strong beyond-GBW signals. This has been recycled into multiple ADI documents, but the implications are mostly left to the reader to figure out.
The text states: "Depending on the device operating current, the interfering frequency and its relative amplitude, these differential pairs can behave as high-frequency detectors. As will be shown, the detection process produces spectral components at the harmonics of the interference, as well at dc! It is the detected dccomponent of the interference that shifts amplifier bias levels, leading to inaccuracies."
From that point on, the documents only ever talk about dc offsets introduced by RF rectification. However, if the input stage is behaving as a detector (as stated) wouldn't it be demodulating the interfering signal, not just producing a DC offset? Eg, a 60MHz AM carrier modulated with a 5kHz envelope would be demodulated by rectification, dumping a 5kHz signal into the input stage, correct?
I've tried the concept out experimentally, but unfortunately my signal generators are producing an AM signal that's too 'impure' to be of much direct use. Also, if rectification is happening, it looks like other distortion effects within the opamp have more to so with accidentally demodulating the signal.
Thoughts?
[having now spent lots of modelling and experiment time with opamps and strong RF inputs...]
One more 'stupid' question. The Analog design documents by Jung mention repeatedly that the input stage BJT/JFETs of an opamp have a tendency to rectify strong beyond-GBW signals. This has been recycled into multiple ADI documents, but the implications are mostly left to the reader to figure out.
The text states: "Depending on the device operating current, the interfering frequency and its relative amplitude, these differential pairs can behave as high-frequency detectors. As will be shown, the detection process produces spectral components at the harmonics of the interference, as well at dc! It is the detected dccomponent of the interference that shifts amplifier bias levels, leading to inaccuracies."
From that point on, the documents only ever talk about dc offsets introduced by RF rectification. However, if the input stage is behaving as a detector (as stated) wouldn't it be demodulating the interfering signal, not just producing a DC offset? Eg, a 60MHz AM carrier modulated with a 5kHz envelope would be demodulated by rectification, dumping a 5kHz signal into the input stage, correct?
I've tried the concept out experimentally, but unfortunately my signal generators are producing an AM signal that's too 'impure' to be of much direct use. Also, if rectification is happening, it looks like other distortion effects within the opamp have more to so with accidentally demodulating the signal.
Thoughts?
Unless we know what a chip's internal circuits look like, all we really know for sure is that "bad things can happen" if the wrong RF gets into the chip.
Basically, ALL p-n semiconductor junctions will rectify RF, to some extent at least, resulting in some DC, and probably other nasties, or even a demodulated AM radio signal.
Any stray DC in there could make the chip operate "incorrectly" (by changing internal DC bias points, etc), which, unless we know the internal circuit, could be "anything", except what we want.
Strong too-high-frequency RF will easily completely disable many types of sensitive electronic systems. So the magnitudes of the potential problems are probably significant. And I have read accounts, in the Chipamps forum, of radio stations' audio coming out of chipamp systems, at a listenable volume.
The basic solution is to keep the unwanted RF out of the system, using coax, shielding, proper grounding, and filtering. It can be difficult. I would certainly want to have RF input filters to try to keep out everything above my frequencies of interest.
Have you experienced a particular problem or are you just trying to make sure the design is solid?
Basically, ALL p-n semiconductor junctions will rectify RF, to some extent at least, resulting in some DC, and probably other nasties, or even a demodulated AM radio signal.
Any stray DC in there could make the chip operate "incorrectly" (by changing internal DC bias points, etc), which, unless we know the internal circuit, could be "anything", except what we want.
Strong too-high-frequency RF will easily completely disable many types of sensitive electronic systems. So the magnitudes of the potential problems are probably significant. And I have read accounts, in the Chipamps forum, of radio stations' audio coming out of chipamp systems, at a listenable volume.
The basic solution is to keep the unwanted RF out of the system, using coax, shielding, proper grounding, and filtering. It can be difficult. I would certainly want to have RF input filters to try to keep out everything above my frequencies of interest.
Have you experienced a particular problem or are you just trying to make sure the design is solid?
I'm trying to make sure the design is solid.
My signal generators are old HP 3325As and only produce up to 60MHz; to that point, feeding strong (Vpp to the rails) RF into the OPA134 and OPA627 does not seem to produce any rectification effects. They're both still amplifying the signal, albeit with quite a bit of rolloff and increased distortion, and neither is producing a measurable DC shift despite a rated GBW product of 8MHz and 16MHz respectively. The decompensated version of the OPA627, the 637, has a rated GBW product of 80MHz-- perhaps the input stages are not yet in a zone where they're in trouble?
The 3325A can AM modulate, but the AM modulation is 'polluted' with a low level (about -70dB) of modulator feed-through. Again, when I feed high amplitude (10Vpp) AM modulated RF into the circuit, 100kHz through 60MHz, I do not see a demodulation product, at least I don't see any that's measurable given that the input signal is 'precontaminated' by the feed-through.
I'm not treating any of this as argument that I don't need an RF filter, I'm trying to understand what the RF filter needs to be. As my intended application is instrumentation (and so source impedence is not controllable), a passive RF filter will need alot of margin to be able to 'slump' without cutting into the audio band. ...or perhaps I'm missing a common technique for better source impedence decoupling (eg, input transformer).
Hi Lumba (have we met? I was in Stockholm recently.)
goatee: Oops, actually I misreported something.
When testing the opamp by pumping in RF with the opamp at unity gain and watching for a DC offset (Walt Jung's recommended test for RF susceptibility), I'm observing no shift *above* the GBW product. However, below the GBW product, I'm seeing a DC shift that peaks at about 1/2 the GBW and falls back off to zero right around the GBW. The DC shift is proportional to the square of the input amplitude, as Jung predicts a shift caused by rectification would behave.
That even seems vaguely intuitive given the large gate capacitance of a JFET input and source limiting resistor: you'd think it would behave as a HF shunt to ground.
...but that's also not what Jung's manuals strongly imply to be the case: That out beyond the GBW be monsters, and the filtering is done with that in mind. In my case, testing two JFET opamps, the monsters are within the rated BW. So I'm now mighty confuzzled.
goatee: Oops, actually I misreported something.
When testing the opamp by pumping in RF with the opamp at unity gain and watching for a DC offset (Walt Jung's recommended test for RF susceptibility), I'm observing no shift *above* the GBW product. However, below the GBW product, I'm seeing a DC shift that peaks at about 1/2 the GBW and falls back off to zero right around the GBW. The DC shift is proportional to the square of the input amplitude, as Jung predicts a shift caused by rectification would behave.
That even seems vaguely intuitive given the large gate capacitance of a JFET input and source limiting resistor: you'd think it would behave as a HF shunt to ground.
...but that's also not what Jung's manuals strongly imply to be the case: That out beyond the GBW be monsters, and the filtering is done with that in mind. In my case, testing two JFET opamps, the monsters are within the rated BW. So I'm now mighty confuzzled.
I would probably tend to try to base the input filter on what the desired input frequency range would be. If it's for audio, then cut off as much above that as you can without affecting your frequencies of interest "too much". Maybe you could even use a higher-order filter, if necessary.
I don't know if you need to worry too much about "high" levels of RF (although it is certainly interesting to look at that). I would probably worry more about what might get picked-up by cables and traces acting as antennas, or what might come from your input sources. I would try to figure out what a reasonable maximum RF input amplitude to expect might be and then see if you can filter the input well-enough and still have your overall accuracy etc be within your spec, without breaking some other spec.
By the way, have you tried connecting a regular FM or AM radio antenna (or even just a dangling wire) to the input, "just for fun"? You could probably also find some higher-frequency (GHz) RF sources to try (with some appropriate short length of wire for an antenna on your input); perhaps a cellular telephone, or a cordless telephone basestation or handset, or maybe even a handheld calculator. (I remember listening to calculators through portable radios, decades ago.)
I don't know if you need to worry too much about "high" levels of RF (although it is certainly interesting to look at that). I would probably worry more about what might get picked-up by cables and traces acting as antennas, or what might come from your input sources. I would try to figure out what a reasonable maximum RF input amplitude to expect might be and then see if you can filter the input well-enough and still have your overall accuracy etc be within your spec, without breaking some other spec.
By the way, have you tried connecting a regular FM or AM radio antenna (or even just a dangling wire) to the input, "just for fun"? You could probably also find some higher-frequency (GHz) RF sources to try (with some appropriate short length of wire for an antenna on your input); perhaps a cellular telephone, or a cordless telephone basestation or handset, or maybe even a handheld calculator. (I remember listening to calculators through portable radios, decades ago.)
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.