Hi.
This have probably been discussed many times, but somehow I'm not finding what I'm looking for.
On my recent line-level PCB's with multiple opamp's I have separated "power-gnd" to the opamps from the signal-GND with a 10Ohm resistor. Vcc and Vss to opamps are decoupled to Power-gnd with 2X47nF for each opamp.
It have worked fine and it's easyto implement, but I'm beginning to question whether its a waste of time or maybe even worse than a shared GND.
I'm NOT looking for 0.00001% distortions or 120db noisefloors.- just common sense with standard parts.
Any ideas?
TroelsM
This have probably been discussed many times, but somehow I'm not finding what I'm looking for.
On my recent line-level PCB's with multiple opamp's I have separated "power-gnd" to the opamps from the signal-GND with a 10Ohm resistor. Vcc and Vss to opamps are decoupled to Power-gnd with 2X47nF for each opamp.
It have worked fine and it's easyto implement, but I'm beginning to question whether its a waste of time or maybe even worse than a shared GND.
I'm NOT looking for 0.00001% distortions or 120db noisefloors.- just common sense with standard parts.
Any ideas?
TroelsM
It might be a good idea to post two or three example schematic diagrams, with big red circles around the "power-gnd" and big green circles around the "signal-GND".
None of the opamp based, line level circuits that I've ever worked on, had a "power-gnd" at all. Just +18V supply, -18V supply, and signal ground.
None of the opamp based, line level circuits that I've ever worked on, had a "power-gnd" at all. Just +18V supply, -18V supply, and signal ground.
I am not sure. I cannot measure down to those levels.
I cannot see any basic difference between a power amp and an amplifier using an opamp (this includes a power opamp).
They should both benefit from an improved topology.
I can't see why adopting separated Power Ground for a power amplifier and the lessons learned in doing that, can't be applied to the smaller signals and smaller currents in an opamp amplifier.
There may be a compelling reason for adopting a separation topology:
The opamp is capable of working at much higher frequencies where small capacitance/higher impedance effects could be more dominant.
I cannot see any basic difference between a power amp and an amplifier using an opamp (this includes a power opamp).
They should both benefit from an improved topology.
I can't see why adopting separated Power Ground for a power amplifier and the lessons learned in doing that, can't be applied to the smaller signals and smaller currents in an opamp amplifier.
There may be a compelling reason for adopting a separation topology:
The opamp is capable of working at much higher frequencies where small capacitance/higher impedance effects could be more dominant.
Last edited:
Basically, having dedicated signal and power ground returns is a good idea that avoids polluting signal ground with power supply noise, especially when return impedance cannot be kept particularly low. The returns then have to be kept separate until they reach an area of "low enough" shared impedance.
If you circuit is physically small, supplies are decently clean and ground impedance can be kept low, it's not going to make too much of a difference. For constructions with large single-sided PCBs and daughterboards (read: typical consumer audio) it ought to be pretty much mandatory IMHO.
If you circuit is physically small, supplies are decently clean and ground impedance can be kept low, it's not going to make too much of a difference. For constructions with large single-sided PCBs and daughterboards (read: typical consumer audio) it ought to be pretty much mandatory IMHO.
Input bias currents have to be returned to the "power ground" in order for the op amp to bias properly. But you don't want to connect the + input pin to a noisy power ground, even if it's connected through a resistor. So I return these bias resistors to the local signal ground. Remember, sometimes these bias resistors also set the input impedance of our circuit, so we don't want to inject noise here.
I connect the "power" ground to the signal ground at only one point, to avoid ground loops. This is typically right by the volume control ground or input resistor ground. In theory, the input bias currents introduce errors, but in practice you can work around it if it really matters.
This configuration is only for line level circuits, and has always worked for me.
Thanks for the input.
I think that the different grounding- techniques may also depend on what kind of PCB you use.
For a single-sided PCB it can be difficult to achieve low ground-impedance over a big area because there is no ground-plane. In this case I agree with SGrossklass.
Kind Regards TroelsM
I think that the different grounding- techniques may also depend on what kind of PCB you use.
For a single-sided PCB it can be difficult to achieve low ground-impedance over a big area because there is no ground-plane. In this case I agree with SGrossklass.
Kind Regards TroelsM
I can't see why one would want to 'isolate' power and signal ground. The point of having low impedance power is that the impedance of an amplifier's power supply, relative to its feedback network, needs to be low. Isolate those two, and you have unknown and probably too high of a power supply impedance relative to the feedback loop's ground reference. At best, the loop between the amplifier, its power supply and the ground reference has become immensely large, probably including the entire circuit wiring back to some magic 'star ground point'.
If you're using a bipolar power supply, the only 'dirty' part in a power supply rail (hopefully) is the rectified signal current components, split into the + and - supplies. Those need to be summed together ASAP, right next to the amplifier, and away from any sensitive ground reference nodes, like those needed by the feedback network.
You want to make it so that each half wave rectified supply current component cannot flow in separate ground foil paths between the amplifier and the load. If any of these currents separately induce voltages that contain this rectified signal current component, and these voltages are used as references to your circuit, you've injected distortion into your circuit.
My preference is to use a completely balanced internal topology, so that these currents are easy to cancel locally. With a balanced circuit, there's always an opposite polarity current nearby that can cancel any given current - everything is in opposite polarity pairs. This allows you to use a four layer board with a very large copper pour on each layer to create a very low impedance ground and power supply system. If you're not dumping trash into the ground foil, there's no reason to slice it up into high Z ribbons that attach to some mythical star ground point - leave it intact and enjoy the low impedance, which _will_ reduce any induced voltages that can result from less than perfect balance. Further, with a broadband low impedance ground, you can use high gain bandwidth amplifiers, have them remain stable, and enjoy the low distortion that they can give you.
If you're using a bipolar power supply, the only 'dirty' part in a power supply rail (hopefully) is the rectified signal current components, split into the + and - supplies. Those need to be summed together ASAP, right next to the amplifier, and away from any sensitive ground reference nodes, like those needed by the feedback network.
You want to make it so that each half wave rectified supply current component cannot flow in separate ground foil paths between the amplifier and the load. If any of these currents separately induce voltages that contain this rectified signal current component, and these voltages are used as references to your circuit, you've injected distortion into your circuit.
My preference is to use a completely balanced internal topology, so that these currents are easy to cancel locally. With a balanced circuit, there's always an opposite polarity current nearby that can cancel any given current - everything is in opposite polarity pairs. This allows you to use a four layer board with a very large copper pour on each layer to create a very low impedance ground and power supply system. If you're not dumping trash into the ground foil, there's no reason to slice it up into high Z ribbons that attach to some mythical star ground point - leave it intact and enjoy the low impedance, which _will_ reduce any induced voltages that can result from less than perfect balance. Further, with a broadband low impedance ground, you can use high gain bandwidth amplifiers, have them remain stable, and enjoy the low distortion that they can give you.
That's basically how people were thinking in the mid-'70s. Too bad voltage regulator noise does exist. Add some non-negligible return impedance shared with signal ground and sizable decoupling caps, and you may be in for a really nasty surprise when it comes to real-life PSRR.
For basic simulation, I picked Rout = 0.3 ohms, C = 20 µF, Rreturn = 0.1 ohms, which I hope we can agree would be the right ballpark - that gives me the expected highpass response with -38 dB at 1 kHz and -19 dB at 10 kHz, easily an order of magnitude worse than even a modest opamp. Let's further assume a noisy 78xx vreg at 200 µV over a 10 kHz BW, working out to roughly 2 µV/sqrt(Hz). That minus 40 dB still is 20 nV/sqrt(Hz), and things would be even worse here. All of this noise is now superimposed on the ground. Knot that grate, now is it?
Let's add 22 ohms of series resistance in the supply. Now attenuation is a minimum of 47 dB across the band, so we'd be looking at less than about 0.9 µV, or 9 nV/sqrt(Hz) of ground noise. That'll add less than 3 dB of noise to a circuit with a 4558 class opamp, which is not great but arguably acceptable for something using bottom-of-the-barrel parts. RC filtering (expectedly) also cleans up the supplies themselves in case circuit PSRR wasn't that great to begin with.
I saw an interesting related approach in the relatively well-known Bugle2 phonopre lately - in this one the supply comes in at the high-level output end and is successively RC filtered as it travels towards the lower-level sections along with the ground return. So the higher shared ground return impedance becomes, the more noise has already been taken out before, and undue ground pollution is avoided.
Parasitic inductance is another story for another day. It also has to be taken care of, of course. Here I just wanted to demonstrate why keeping SGND and PGND separate can be useful.
The amplifiers I like to use (LME49710, AD797) have very high PSRR and the regulators I like to use (ADP7142, ADP7182) are quiet. I can't see anything but the predicted Johnson noise from the resistors and amplifiers in these circuits' outputs. And, when using the LM317/337 instead of the fancy regulators, I see no difference in the circuits performance WRT output noise.
The PSRR of an LME49710 is around 100 or 110dB even at the upper end of the passband, so the regulator noise levels you quote, say 200µV, will get reduced to output referred levels around -170dB, far into the rest of the circuit's noise floor. I think this I why I have never noticed regulator noise problems, even when using the much noisier LM317/337.
When you quote the 200µV power supply noise, how does this magically become "superimposed on the ground"? Surely, ground is defined as zero volts, so it becomes 0V by definition. Do you mean that this power supply voltage induces ground currents related to the 200µV power supply noise signal? That's a different situation, and whether it's relevant or not again depends upon PSRR and where these currents flow within the ground foil.
You're correct that these noise currents can't be cancelled by a balanced topology, since they're uncorrelated signals - each noise signal is unique. However, because of that, they add in an RMS fashion, so even if there are multiple regulators (which I use), their currents don't grow that fast as the number of regulators is increased.
However, if these noise currents generate a voltage within the ground foil, the contribution of that IR developed voltage can be coupled equally into the positive and negative halves of a balanced circuit, as long as the layout is done to closely connect opposing connections to ground. Because of this symmetry, these circuit connections to ground have very low output referred gain, something worthwhile that I think is easier to accomplish in a PCB layout than a segmented ground structure.
Using RC filters to clean a supply rail can move these noise currents from one one place to another, but again, this does not make them disappear.
So, for my balanced circuits using high PSRR amplifiers, I don't see any penalty to having one better connected ground as opposed to a segmented system.
This may be a problem with a single ended circuit, especially using discrete amplifiers that can have sometimes near zero PSRR. Indeed, that tiered RC filter system works well in that situation, as it does in old guitar amp circuits that have near zero PSRR. But, if you're trying to get a 'clean' circuit, those sorts of amplifiers have many other impediments to that goal.
The PSRR of an LME49710 is around 100 or 110dB even at the upper end of the passband, so the regulator noise levels you quote, say 200µV, will get reduced to output referred levels around -170dB, far into the rest of the circuit's noise floor. I think this I why I have never noticed regulator noise problems, even when using the much noisier LM317/337.
When you quote the 200µV power supply noise, how does this magically become "superimposed on the ground"? Surely, ground is defined as zero volts, so it becomes 0V by definition. Do you mean that this power supply voltage induces ground currents related to the 200µV power supply noise signal? That's a different situation, and whether it's relevant or not again depends upon PSRR and where these currents flow within the ground foil.
You're correct that these noise currents can't be cancelled by a balanced topology, since they're uncorrelated signals - each noise signal is unique. However, because of that, they add in an RMS fashion, so even if there are multiple regulators (which I use), their currents don't grow that fast as the number of regulators is increased.
However, if these noise currents generate a voltage within the ground foil, the contribution of that IR developed voltage can be coupled equally into the positive and negative halves of a balanced circuit, as long as the layout is done to closely connect opposing connections to ground. Because of this symmetry, these circuit connections to ground have very low output referred gain, something worthwhile that I think is easier to accomplish in a PCB layout than a segmented ground structure.
Using RC filters to clean a supply rail can move these noise currents from one one place to another, but again, this does not make them disappear.
So, for my balanced circuits using high PSRR amplifiers, I don't see any penalty to having one better connected ground as opposed to a segmented system.
This may be a problem with a single ended circuit, especially using discrete amplifiers that can have sometimes near zero PSRR. Indeed, that tiered RC filter system works well in that situation, as it does in old guitar amp circuits that have near zero PSRR. But, if you're trying to get a 'clean' circuit, those sorts of amplifiers have many other impediments to that goal.
I presume you're not seeing anything because, well, your layouts don't suck. 😉
That's not necessarily the case for older consumer gear. This thread, for example, proved to be quite instructive - the stock circuit suffered from inadequate PSRR and a noisy regulator, and when trying to fix the PSRR issue by adding RC filtering, the decidedly non-star grounding reared its ugly head, resulting in hiss being basically reduced but rising even above original levels with the volume turned up. Ultimately grounding was reworked to something more starry, and a dedicated power ground return was added for the filtering, finally making the unit nice and quiet as it should be.
Fast-forward from 1976 to 1979, and what do you know, the manufacturer had clearly learned from their mistakes after a look at the schematics.
I also remember an episode involving a higher-end Yamaha tuner (maybe a T-80 or T-85, so we're talking early-mid-'80s) where the unit was recapped and star grounding added (not too many details disclosed), and afterwards measured a fair bit quieter. I was always wondering why that would be - well, it sure seems less mysterious these days. Hmm, can't find that right now, may have been in the FMtuners group. Anyway, we are no doubt talking large single-layer PCB with a fair bit of circuitry involved.
Circuits that don't work properly clearly have their benefits, there's plenty to be learned from 'em...
That's not necessarily the case for older consumer gear. This thread, for example, proved to be quite instructive - the stock circuit suffered from inadequate PSRR and a noisy regulator, and when trying to fix the PSRR issue by adding RC filtering, the decidedly non-star grounding reared its ugly head, resulting in hiss being basically reduced but rising even above original levels with the volume turned up. Ultimately grounding was reworked to something more starry, and a dedicated power ground return was added for the filtering, finally making the unit nice and quiet as it should be.
Fast-forward from 1976 to 1979, and what do you know, the manufacturer had clearly learned from their mistakes after a look at the schematics.
I also remember an episode involving a higher-end Yamaha tuner (maybe a T-80 or T-85, so we're talking early-mid-'80s) where the unit was recapped and star grounding added (not too many details disclosed), and afterwards measured a fair bit quieter. I was always wondering why that would be - well, it sure seems less mysterious these days. Hmm, can't find that right now, may have been in the FMtuners group. Anyway, we are no doubt talking large single-layer PCB with a fair bit of circuitry involved.
Circuits that don't work properly clearly have their benefits, there's plenty to be learned from 'em...
- Status
- Not open for further replies.