"Snubber" impedance graphs

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Scottw,

Having re-read, now I see what was your real question! And it's a good one and also important: the second peak is the resonance between the 100nF value cap's reminiscent ESL & the 2.4 nF. It's always very risky [~impossible] to paralell two film caps, and avoid getting them into resonance.. Or rather, to parallel anything, with a capacitance value smaller then, to a film cap..

That is, if we keep on not inserting small resistances in series. Carlos had tried it, correctly, but with a value not high enough.

Ciao, George
 
George,

I've just posted it elsewhere, but would like to report also here:
As Bruno explains here
, the necessary damping resistance depends on the impedance value of the original resonance peak. In case of the transformer and it's stray capacitance, this peak is higher. In these bypass networks the value is less, so .1 - 1 ohm values can be OK, but should be tested..

Ciao, George [the other]
 
hi, seahag

It's a good question and long story.. Such a resonant power supply impedance will cause sudden phase shifts in an amplifier's open loop response, at a frequency range where the inherent phase shift in itself is already high, so the applied negative feedback will become not so negative...
In a few word, a [high enough] speed amplifier under such bypass circumstances can / will become unstable, and oscillate.
That oscillation is still not audible, but saturates / shifts internal bias values, moreover gets mixed down into audible regions by the amp's nonlinear transfer curve. [gets mixed to the music signal]

How bad all this really is depends highly on the amplifier speed.
A really slow amp which does not show any signs of life at these frequencies, will continue happily singing.

Ciao, George
 
Hi George, thanks for your measurements.
Some coments...

What I see on these measurements is that, although there are two peaks, there is a lower impedance on the snubberized setup. And this is even more evident at low(er) frequencies.
The test setup is not exactly my complete PSU, and I recommend bypassing and snubberizing right across the caps' legs (usually under the PCB).

There are some remarks I would like to make, some of this is posted on my more recent thread about this last PSU:

- The increase of 100uf to 2,200uf directly on the chip's pins is purely based on listening tests.
The amp gains in dynamics, and bass is tighter. I reported this.
The amp seems to drive any speaker you throw at it, and certainly drives my Epos better than a Krell KAV-300iL I had at home for some time.
2,200uf at 50V rating is still a manageable size to be near the chip.
With this larger value, I recommend snubberizind on-board, and 0.1R+3.3nF gives very good results.
This on-chip arrangement gives very good results either with the unregulated or the regulated PSU.
Didn't try batteries yet, but I have 4x 12V/24A batteries (from big UPS systems) waiting for some time available...

Also, on page 15 of the LM3886 datasheet you can see this:
"If adequate bypassing is not provided the current in the supply leads which is a rectified component of the load current may be fed back into internal circuitry. This signal causes low distortion at high frequencies requiring that the supplies be bypassed at the package terminals with an electrolytic capacitor of 470 μF or more."

- I always used supply bypassing between the rails with op-amps, and here's a good reason to do so:
http://www.elecdesign.com/Files/29/1478/Figure_02.gif

The increase in value gave me better performance, mainly in the treble. Cleaner, more precise.

Well, this is audio and finally the end result (and what really matters) is: does the amp sound better?
Yes.
I also think George, that to be accurate, either measuring and/or listening, you would actually have to build the complete amp and PSU.
 
dhenryp,

Don't take me wrong!
It's only "really slow" as referred to the speeds involved here!
For audio, it still can be perfect! In fact, this is the general reason why a slower opamp, like NE5532 still works beautifully and gets suggested in places where others, like lm6172, would oscillate like crazy.

Ciao, George
 
Joseph K said:
dhenryp,

Don't take me wrong!
It's only "really slow" as referred to the speeds involved here!
For audio, it still can be perfect! In fact, this is the general reason why a slower opamp, like NE5532 still works beautifully and gets suggested in places where others, like lm6172, would oscillate like crazy.

Ciao, George
Hi All,

In the midst of such compelling information you must note that the effect of the snubber is constrained within the MHz region it is the precise reason why certain amps like Mauro's "My Ref", especially REV_C are intentionally compensated and constrained to within a carefully prescribed bandwidth. That is a mark of truly intelligent well thought out design, and the key reason a snubber arrangment is left out. The amp does not leave its healthy operation to the caps in the PSU. It manages its own state.

A very good study Joseph! Thanks for such a thorough treament so far. I will watch with an open mind.

It confirms my thoughts that a snubber has its place, as I have known for a long time.

Cheers!
Russ
 
Dear Carlos,

First of all: please feel free to criticize the test setup - this is the way to convergence! If necessary, I can modify and redo things. [though not tonight]
Second: All my remarks are strictly technical, [hope so], they should be, because I was not listening to the full setup [yet]. In fact, I am not listening to GC-s these days - they have too low Zout for me... :D
[this was just a remark about my recent fixations]

there is a lower impedance on the snubberized setup.

Yes, it's true, at places, for example below 1.5MHz, and above 25 MHz.

The test setup is not exactly my complete PSU, and I recommend bypassing and snubberizing right across the caps' legs (usually under the PCB)

Again true, not complete, as I have noted it also. Though I maintain that the critical parts from local bypass point of view are present.
Now just a notice: Don't be mislead by the seemingly untight construction on this test jig - as I have tried it explain somewhere,
and am too sleepy to repeat here precisely, the distances over the jig's ground plane & along the broken out transmission line are working in a different way. So here we have the 100nF directly accross the [virtual] supply pins, with very short legs; actually, the 2200 uF in the real life can not get so close like here; the only doubt can be about the snubber, but the loop area could be decreased only by decreasing the resistor size [the cap is small]. The prescribed size & value is 1W, as here. These are the main elements. The rest, as you could see, does not have a great influence. [But does not help as well, though for a moment I hoped that, this is why I tried]

The cap across the supply pins [3,3uF] is definitely a good idea, and also proven by many others. [also by Mauro] I have a suspicion that without damping causes an another resonance, but I will try to demonstrate that at a later instance.

So, what I wanted to point out, [and I would like to emphasise- technically], that this new approach of Yours brings new risks into the picture. Where I don't really understand the balance is the "snubber" values. With these values it does not snub, but creates impedance peaking. Sure, it also gives more attenuation, but in a very narrow band [between 25 and 40 Mhz, below much worse, above 2-3 db improvement].

Again, it's true, it's only a simulation, and the real values are different. But for me the tendencies are real, so the general picture should be similar in the real circuit, as well.
Unless some of you does not point out some major failure...

vow, it's really late, let's continue later!

Ciao, George
 
Russ White said:
The amp does not leave its healthy operation to the caps in the PSU. It manages its own state.

Russ, it's not possible to make an amp completely insensitive to the PSU.
Do you really believe that?
Slower devices are less demanding (and this is very true for op-amps), but the LM3886 is not that slow.
It's much faster and higher bandwidth than the OPA chips (OPA541/8/9).

Use a switching PSU, batteries, regulated, unregulated... they will all have an effect (more or less noticeable) in the amp's sound.
Obviously everything counts, and the PCB layout is very important, as I always said here too.
But you still can't escape that a large part of an amplifier's sound is the PSU.
 
carlosfm said:


Russ, it's not possible to make an amp completely insensitive to the PSU.
Do you really believe that?


Absolutely I do not believe that. See you have clearly missunderstood me.

What I said is a great amplifier PSU does not necessarily need snubbers, nor will they always make things "better". :)

That has been clearly demonstrated.

Snubbers have their place, but not in every LM3886 design.

Also, depending on the circuit, the LM3886 is not always operating at high bandwidth. It can be externally compensated much as any onther high speed opamp. Like the OPA637 for example.

Cheers!
Russ
 
Russ White said:
What I said is a great amplifier PSU does not necessarily need snubbers, nor will they always make things "better". :)

That has been clearly demonstrated.

Where has that been clearly demonstrated??? :confused:
How can you take the conclusion that, although with two peaks, there are no benefits from having lower PSU impedance?
You would have to listen to the complete setup (amp/PSU), right?

Taking conclusions on an isolated PSU is nonsense.
And even then, it's arguable what is the best: lower impedance or less peaking.
If it was the later, you would not even need to bypass. Just use some big caps on the chip and enjoy a nasty sound.
 
carlosfm said:


Where has that been clearly demonstrated??? :confused:


It has been demonstrated by the thousands of successful designs which have no snubber. And the many designers who, though certainly not ignorant of snubbing techniques choose not to use them.

It has also been born out in testing, both empirical measurments and listening tests. While there are often no explainations for what people "hear" I can tell you that I hear no improvement when a PS is snubbed as opposed to unsnubbed in "most" circumstances. You may not agree, and that is OK.

In support of the snubber I say this: If I were to design a LM3886 amp with low gain (say 10x) I would very likely use a snubber, because you are increasing the bandwidth of the Lm3886 significantly at that point.

So it is not that I think a snubber is useless, just that it is not always useful.

Cheers!
Russ
 
Russ White said:
It has been demonstrated by the thousands of successful designs which have no snubber. And the many designers who, though certainly not ignorant of snubbing techniques choose not to use them.

Sure... I have improved lots of 'successful designs' myself, and I see lots of badly designed stuff, no matter what price.
The last amp was a Pioneer A400.
You should listen to it before and after.
And lots of so-called high-end stuff, including Krells.
I'm not going to say anything related to other gear.
If snubbers generate so much irritation when applied to audio amplifier PSUs, just imagine on other stuff.


Russ White said:
It has also been born out in testing, both empirical measurments and listening tests.

I don't think so:

http://www.diyaudio.com/forums/showthread.php?postid=728161#post728161

Can you point me out a link for some of those (so well known???) tests?

Russ White said:
In support of the snubber I say this: If I were to design a LM3886 amp with low gain (say 10x) I would very likely use a snubber, because you are increasing the bandwidth of the Lm3886 significantly at that point.

Well, of course, as I've been saying for long, the LM chips sound better at lower gains (up to x20~22).
That's fine for use as a power amp, and that's what they were made to.
Of course, a very good pre is ESSENTIAL, and it provides the necessary gain to match the power amp.
 
carlosfm said:

You, of course, are free to think what you like. :)

Are you really trying to prove a point about the LM3886 buy referring to a nearly completely different class D arrangement?

It is not even all that similar to your snubber. So I am not sure the point. In any case, I am sure many designers would disagree that your adding a snubber made their design "better".

Lets be clear, its not that I don't think I would never use a snubber, its just that if I did there would be a precise reason why, and a clear purpose in the circuit. If I could not prove that it was effective (either by blind testing, or measurment) it would never make it to production. I have yet to see or hear that proof for "most" audio circuits involving chipamps.

I would need to see much more evidence to persuade me that it is "generally" desirable to include it.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.