Lately I've done a lot of reading about how to decouple the power supply rails of opamps. But I'm still not sure, so now I want to know once for all, what is the right way to decouple the op-amps?
Let's assume I'm building a preamplifier circuit using a few NE5532 opamps (or similari). From that I learned so far the usual way to do this is to put two electrolytic capacitors at the the board, preferably close to there the PSU is connected to the board (to minimize that ripple current enters the signal ground). And then maybe also two smaller ceramic caps close to them as this;
Then, decouple each opamp power supply pins. This is what I'm unsure of how to do, but I be think there are three main options to choose from here. The first and simplest is to put a ceramic cap of 100nF between the two supply pins at each opamp. Like this;
The second option B is as suggested as almost all datasheet, one ceramic 100nF cap or so as close to each supply pins as possible, then to ground. The downside this this as that I be leave the signal and power supply ground should be separate at the board or very great care has to be taken with the layout, star grounding etc... So that ripple currents from the decoupling capacitor can flow through the signal ground.
The third C option I've seen is to put a capacitor as close to each supply pins as possible and the other rail, like this. This makes the distance from each power supply pin to decoupling capacitor shorter than in option A.
Option A and C has the advantages that no extra power ground is needed, simplification the board layout a lot. Question is if that is enough decoupling? Of course let's assume NE5532 (or similar BJT) op-amps and audio frequency use. And is C any improvement compared to A at all? Maybe overkill but ceramic caps are quite anyway.
Let's assume I'm building a preamplifier circuit using a few NE5532 opamps (or similari). From that I learned so far the usual way to do this is to put two electrolytic capacitors at the the board, preferably close to there the PSU is connected to the board (to minimize that ripple current enters the signal ground). And then maybe also two smaller ceramic caps close to them as this;
Then, decouple each opamp power supply pins. This is what I'm unsure of how to do, but I be think there are three main options to choose from here. The first and simplest is to put a ceramic cap of 100nF between the two supply pins at each opamp. Like this;
The second option B is as suggested as almost all datasheet, one ceramic 100nF cap or so as close to each supply pins as possible, then to ground. The downside this this as that I be leave the signal and power supply ground should be separate at the board or very great care has to be taken with the layout, star grounding etc... So that ripple currents from the decoupling capacitor can flow through the signal ground.
The third C option I've seen is to put a capacitor as close to each supply pins as possible and the other rail, like this. This makes the distance from each power supply pin to decoupling capacitor shorter than in option A.
Option A and C has the advantages that no extra power ground is needed, simplification the board layout a lot. Question is if that is enough decoupling? Of course let's assume NE5532 (or similar BJT) op-amps and audio frequency use. And is C any improvement compared to A at all? Maybe overkill but ceramic caps are quite anyway.
Last edited:
The key insight is that high speed decoupling capacitors need to have very little inductance in series with them, short, wide traces, little loop-area, otherwise they may not be effective. In many audio circuits without an RF groundplane this means as close to the chip as possible and direct traces. In RF circuitry you site the caps on each power pin (within 1mm if you can is good, 3mm is usually OK), and have a via straight through to the groundplane. High speed logic is similar. Unless we are using video opamps, opamp high speed decoupling is often less critical, but it varies, the NE55xx family exhibit degraded performance without good high speed decoupling, but nothing blatant, so its easy to miss this - Doug Self has done a lot of research over the years on these devices.
Be careful not to inject rail noise into a sensitive ground node with decoupling caps - that's why its a good idea for the 100nF to just be between the rails, and the bulk electrolytics to be away from the signal path - currents into these electrolytics are non-linear (half cycles), so best to route them to a star-ground point separate to the signal ground trace. The bulk electrolytics need to be to ground to prevent the supply rails providing feedback paths in the audio band (in high-gain amplifier setups at least), the high speed decoupling is a different animal really.
Be careful not to inject rail noise into a sensitive ground node with decoupling caps - that's why its a good idea for the 100nF to just be between the rails, and the bulk electrolytics to be away from the signal path - currents into these electrolytics are non-linear (half cycles), so best to route them to a star-ground point separate to the signal ground trace. The bulk electrolytics need to be to ground to prevent the supply rails providing feedback paths in the audio band (in high-gain amplifier setups at least), the high speed decoupling is a different animal really.
Correct, D.Self advocates for a single cap between PS pins, no ground, just like "A".
A lot of consumer audio equipment I have seen, and designs here on DIYaudio, use type B. They have one capacitor on each supply rail to ground as close to the ICs as physically possible, with the larger electrolytics far away from the ICs.
I believe it is better, at least academic, to have a single capacitor directly from one rail to the other (type A), rather than sharing the same ground plane as the audio signal ground.
I believe it is better, at least academic, to have a single capacitor directly from one rail to the other (type A), rather than sharing the same ground plane as the audio signal ground.
I suspect each chip is different and the absolute best approach might take some careful research - for instance some opamps have very different PSRR on each rail and that asymmetry may affect stability differently with the different decoupling schemes.
Another thing to consider is whether its worth adding some RC filtering to the supply rails to each opamp, say be adding 10~50 ohms in series before the decoupling, further separating the opamp from its environment. I'm playing with the design of a switchable lab amplifier using opamps and switchable gain from +10dB to +80dB - that high maximum gain is asking for issues if the individual stages aren't well decoupled from each other - even if its stable the actual gain value might be influenced by tiny amounts of inadvertent feedback, perhaps in a frequency-sensitive manner leading to non-flat response. I'm trying RC filters between each stage on both rails, with incoming power at the output stage.
OK, that's a pretty extreme situation, but its another possible consideration.
Another thing to consider is whether its worth adding some RC filtering to the supply rails to each opamp, say be adding 10~50 ohms in series before the decoupling, further separating the opamp from its environment. I'm playing with the design of a switchable lab amplifier using opamps and switchable gain from +10dB to +80dB - that high maximum gain is asking for issues if the individual stages aren't well decoupled from each other - even if its stable the actual gain value might be influenced by tiny amounts of inadvertent feedback, perhaps in a frequency-sensitive manner leading to non-flat response. I'm trying RC filters between each stage on both rails, with incoming power at the output stage.
OK, that's a pretty extreme situation, but its another possible consideration.
The NE5532 has unusual frequency compensation that makes it sensitive to the voltage difference between the rails. That is why "A" is better.
Typical IC op-amps are sensitive to the negative rail. Those do better with "B".
BTW, a discrete transistor op-amp can be made insensitive to both rails. It has a ground connection.
Ed
Typical IC op-amps are sensitive to the negative rail. Those do better with "B".
BTW, a discrete transistor op-amp can be made insensitive to both rails. It has a ground connection.
Ed
Wow, never expected this many replies. Very interesting too, and based on what I read, I won't be completely wrong anyway. For a circuit using NE5532 opamp option A seems to be the best choice. It's simpler, won't inject noise to ground and shouldn't be any problem with the NE5532 at least. I know Douglas Self and Elliot designed some circuits like this too (Randy Slone suggest different though).
Wonder why the datasheet for the NE5532 suggests option B, but the datasheets doesn't always say everything. And this is probably like the general design rule suggested by the manufacturer for all opamps.
Interesting that no one commented option C. Guess the difference between A and C is negligible, at least for audio frequencies. I'm sure that I've seen this somewhere though I can't remember from where.
Wonder why the datasheet for the NE5532 suggests option B, but the datasheets doesn't always say everything. And this is probably like the general design rule suggested by the manufacturer for all opamps.
Interesting that no one commented option C. Guess the difference between A and C is negligible, at least for audio frequencies. I'm sure that I've seen this somewhere though I can't remember from where.
Ok, so for ex. LM4562, option B would be preferred instead? LM4562 has a gain bandwidth product of 55 Mhz, compared to NE5532 which has a unity gain bandwidth of 10 Mhz.
Electrically, A and C are identical except with C having capacitance of 200nf (it is simply parallel). I doubt the track distance would have any effect other than negligible.
I doubt there would be much noise 'injected to the ground' with B, as you are only taming stray high frequencies to keep the opamp stable. The bulk noise filtering is being taken care of close to your PSU. It can be a pain to add an extra ground, but it's my preference.
Let us know which version you choose and how it works out for you.
I doubt there would be much noise 'injected to the ground' with B, as you are only taming stray high frequencies to keep the opamp stable. The bulk noise filtering is being taken care of close to your PSU. It can be a pain to add an extra ground, but it's my preference.
Let us know which version you choose and how it works out for you.
Well I was thinking of keeping the ground for the pcb decoupling capacitors (c2 and c3 at the first picture) separate from the signal ground for this reason. Routing separate ground tracks for the decoupling caps at each opamp will be a lot more of work though. I plan to use maybe 10 chips or so...
Then I'm done with the design and the pcb is manufactured and assembled. I'll let you know of course!
Then I'm done with the design and the pcb is manufactured and assembled. I'll let you know of course!
High frequency decoupling capacitors must not have routed tracks to a remote ground location.
That makes them much less effective due to the added trace inductance and loop area.
The capacitor ground lead must go directly to the ground plane without appreciable trace length,
just like the capacitor's other lead must go directly to the IC supply pin.
That makes them much less effective due to the added trace inductance and loop area.
The capacitor ground lead must go directly to the ground plane without appreciable trace length,
just like the capacitor's other lead must go directly to the IC supply pin.
There was some discussion on bypass in another thread. A post by me followed by some from @KSTR may be of possible interest: https://www.diyaudio.com/community/...apacitor-simulation-model.382329/post-6944464
EDIT: Option B possibly doesn't always sound better, at least in cases where its not really needed.
EDIT: Option B possibly doesn't always sound better, at least in cases where its not really needed.
Last edited: