Here is a simulation comparison of different snubbing techniques. U1 is a film resistor model I made based on the measurements here:
Resistor's behavior at HF and VHF.
Note that it may be optimistic to assume a real .2R resistor will not have more inductance than film resistors usually do.
I know someone who spent a lot of time working on their power supply, testing different decoupling schemes, etc. They ended up using a 1uF high-Q cap with 70nH series inductance because that was what sounded best to them. They thought because this was a high-Q cap, that it would be good for RF decoupling. If it did improve the sound however, it was probably because the inductance was too large to resonate much if at all with the loop inductance through the reservoirs. As the film decoupling inductance approaches 5 times the loop inductance, the resonance disappears. So while this person thought they were improving the sound by improving RF decoupling, the reality was they were actually making RF decoupling so bad that it wouldn't resonate and get in the way. This is not necessarily a bad approach, because the high ESL film cap may still lower supply impedance significantly at it's SRF. The downside is high supply inductance.
This is what the series resistor damping solution seems to be to me, except it may actually be worse because the resistor doesn't have that much inductance, and you don't get the benefit of low impedance at the SRF. You can see in the chart that a small lytic will have less ESL than the 5mm MKT and series resistor combination. The snubber-lytic solution is special because it allows both low-ESR and low-ESL.
Resistor's behavior at HF and VHF.
Note that it may be optimistic to assume a real .2R resistor will not have more inductance than film resistors usually do.
I know someone who spent a lot of time working on their power supply, testing different decoupling schemes, etc. They ended up using a 1uF high-Q cap with 70nH series inductance because that was what sounded best to them. They thought because this was a high-Q cap, that it would be good for RF decoupling. If it did improve the sound however, it was probably because the inductance was too large to resonate much if at all with the loop inductance through the reservoirs. As the film decoupling inductance approaches 5 times the loop inductance, the resonance disappears. So while this person thought they were improving the sound by improving RF decoupling, the reality was they were actually making RF decoupling so bad that it wouldn't resonate and get in the way. This is not necessarily a bad approach, because the high ESL film cap may still lower supply impedance significantly at it's SRF. The downside is high supply inductance.
This is what the series resistor damping solution seems to be to me, except it may actually be worse because the resistor doesn't have that much inductance, and you don't get the benefit of low impedance at the SRF. You can see in the chart that a small lytic will have less ESL than the 5mm MKT and series resistor combination. The snubber-lytic solution is special because it allows both low-ESR and low-ESL.
Attachments
For this reason I think a more applicable test is to inject a signal from the output of a 50R square wave generator into the rails and watch how they respond.[/QUOTE]
Please give a practical schematic.
Keantoken,
I can very well accept your findings - they dovetail well with what Kendall Castor-Perry found. See at the bottom of this page here: Miscellaneous stuff .
What I wonder though is whether these resonances ever occur in the typical audio power supply.
Jan
Thank for this useful research, keantoken, more 'scientific' than all those aggressive comments from so calling 'objectivists'.
40 years ago, i was advised by a friend of mine, mixing desk manufacturer, to always use big lytics, each one paralleled by as many 1/10 values to can reach a MKT (or so) capacitance.
Ex: 1000, 100, 10, 1µF.
It seems it works well enough for audio with our imperfect components in an imperfect world... most of the time.
Last edited:
I was shocked a few years ago at how big the oscillations on my cascode circuit were. I would never have seen this without a wideband scope (180 MHz at c. 800~900 mV pk-pk)
So this is whay I am interested in Keantokens findings - of course its a different mechanism (PSU ringing), but I think some quite strange things can go on in an amplifier - and this is especially so with modern high ft devices.
So this is whay I am interested in Keantokens findings - of course its a different mechanism (PSU ringing), but I think some quite strange things can go on in an amplifier - and this is especially so with modern high ft devices.
Feedback shunt capacitor needs to be 'right'.
Keantokens findings very much need to be noted in this capacitor application.
Dan.
This point is interesting. For various reasons, (we don't know how to do it, we don't have good enough measuring instruments, we cannot see what happens etc) unexpected behaviors can happens in an amplifier.
Of course, when we can figure-out exactly where the problem is, we have to understand why, and cure-it the best we can.
But what the hell if an amplifier oscillate at Ghz and we cannot see-it ?
The main question is: "Did this affect the sound or not".
If yes, we have to try (more or less blind) to cure this invisible problem, a way or an other.
Yes, it would be much simpler if we knew everything all the time. And were able to just apply formulas. But, as all the 'make believe' disciplines (i thing about lens design in photography) there is a lot of art and feeling involved in the creative process of audio design.
And the best pure scientist will never be able to achive what can do an other engineer, more sensible, more modest about the area of his knowledge, using its sens and his feeling, in parallel with correct engineering (witch never can be bypassed).
Of course, when we can figure-out exactly where the problem is, we have to understand why, and cure-it the best we can.
But what the hell if an amplifier oscillate at Ghz and we cannot see-it ?
The main question is: "Did this affect the sound or not".
If yes, we have to try (more or less blind) to cure this invisible problem, a way or an other.
Yes, it would be much simpler if we knew everything all the time. And were able to just apply formulas. But, as all the 'make believe' disciplines (i thing about lens design in photography) there is a lot of art and feeling involved in the creative process of audio design.
And the best pure scientist will never be able to achive what can do an other engineer, more sensible, more modest about the area of his knowledge, using its sens and his feeling, in parallel with correct engineering (witch never can be bypassed).
In most cases it does affect the sound. In the early version of my CFA project, I had a slight (low level) oscillation @ 12.8 MHz. You definitely can't hear this oscillation itself, but what you can hear - the sound becomes "harder", less natural @ "highs". Such oscillation affects temperature regime of OPS, increases probability of shoot-through currents, etc.
Conclusion - any oscillation, whatever the frequency is, has to be addressed and eliminated.
I use 200MHz oscilloscope for watching the output until I am sure the circuit is rock stable...
Exactly. And, with experience, we are able to recognize such a behavior by this typical color of the sound.
It is like the perfume creators, they are able to list all the components of a perfume. Not because they have a better nose than us, but because they know the smell of all those individual components. Like we are able to recognize a friend in a crowd.
With age and experience, we use less and less often measuring instruments (and theory), as did those who taught us when we were young.
Thats a different issue.
Like I said earlier, you have to separate loop stability from parasitic oscillations.
You will chase your tail otherwise . . .
Comment; Generally, when we have wide band circuits as a side-effect of making a very, very linear amp, then we need to learn how to make physical layout and pcb and components suitable for stable wide-band operation.
I have never had such osc show up. Maybe its because of my RF background. I automatically do layout carefully as if for HF/RF stability. [I can measure anything up to 6GHz here in Marsh-Land. If it's there, i'll see it]
Throwing together a pcb layout for 20-20KHz and avoiding ground loops, probably isnt adequate for a circuit which can amplify signals far greater than the input signal source.
THx-RNMarsh
I have never had such osc show up. Maybe its because of my RF background. I automatically do layout carefully as if for HF/RF stability. [I can measure anything up to 6GHz here in Marsh-Land. If it's there, i'll see it]
Throwing together a pcb layout for 20-20KHz and avoiding ground loops, probably isnt adequate for a circuit which can amplify signals far greater than the input signal source.
THx-RNMarsh
Last edited:
The hundred MHz or GHz oscillations destroy good sound.
Attachments
I came across this web page on search for current drive amplifiers.
Amplifier topologies for current-drive | Current-Drive - The Natural Way of Loudspeaker Operation
It seems like a natural progression from current feedback. Why not a current amplifier from input to output. A large variable current source/sink. Operate the speaker in a series connection in a current loop rather than a voltage loop. After all a speaker is a current controlled device. This probably deserves a thread of its own.
Amplifier topologies for current-drive | Current-Drive - The Natural Way of Loudspeaker Operation
It seems like a natural progression from current feedback. Why not a current amplifier from input to output. A large variable current source/sink. Operate the speaker in a series connection in a current loop rather than a voltage loop. After all a speaker is a current controlled device. This probably deserves a thread of its own.
Last edited:
This is how I do it: Just take an amp you have sitting around with both film decoupling and large reservoirs on the board, drain the reservoirs, and inject a square wave from a signal generator across the rails. The circuit doesn't need to be on for you to observe the resonance. Just don't send DC into the input of your signal generator. Probe across the rails with your oscilloscope and watch the ringing. You may have to turn the signal generator voltage up to get a clear trace. I used 10V with a 50R output impedance which is 200mA. I also used the offset voltage function to check if the ringing changed with some bias voltage - maybe slightly.
I did this with a TDA7297 kit gifted to me by a friend. It came with a 5mm 100nF decoupler and a 2200u reservoir. The resonance was 3.3MHz. I damped that and sibilance decreased. I replaced the cap with a 5mm 1u decoupler on the same pad, so inductance was the same, and then damped that with a 100V/47u Nichicon VR (it was the junk cap that worked based on measurements). I have another chipamp I damped with 50V 4.7uF caps; it had 5mm 100nF film decouplers and long traces (inductors) between them and the 220uF reservoirs.
I did this with a TDA7297 kit gifted to me by a friend. It came with a 5mm 100nF decoupler and a 2200u reservoir. The resonance was 3.3MHz. I damped that and sibilance decreased. I replaced the cap with a 5mm 1u decoupler on the same pad, so inductance was the same, and then damped that with a 100V/47u Nichicon VR (it was the junk cap that worked based on measurements). I have another chipamp I damped with 50V 4.7uF caps; it had 5mm 100nF film decouplers and long traces (inductors) between them and the 220uF reservoirs.