LM3886 unity gain

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

I have found application note AN-257 from Analog Devices describing how to make almost any opamp amplifier unity gain stable. Article starts on the page 6 and it is named "Trick those op amps".

According the information here is my take on calculating the needed values, which are Rc and Cc. I am not sure if my calculations are right, thats why I post them.

I choose minimum stable gain for LM3886 to be 11 (according to the datasheet it is 10).
For this gain I have Rf=10k Ohm and Ri=1k Ohm. (But I think Ri doesn´t matter and Rf I can choose whatever I can? - If I understand the article correctly.)

Rc= Rf/10, so Rc=1k Ohm

Cc= 1/(2*Pi*Rc*(fc/10)), so I need fc now.

fc for gain 11 is = open loop BW/closed gain, so fc=8 MHz/11, fc=727,2 kHz

Cc= 2,2nF, Cc should be made 1000 times bigger to lower the overshoot,
so Cc= 2,2uF

Calculated results for unity gain stable LM3886 should be:
Rf= 10k Ohm
Rc= 1k Ohm
Cc= 2,2uF

Are these calculations right?



  • AN-257.pdf
    611.4 KB · Views: 332
  • lm3886t.pdf
    745.3 KB · Views: 264
from my reading of the 3886 info it appears the gain of 10 phase margin is only ~30degrees.
For a more useful phase margin of 60degrees the gain should be set to ~20 to 24times.
For 80degrees set the gain to ~30times.
If these higher gains were substituted for the 10times, where would your network values end up?
the article is a start

I don't think the 1000x factor on the C is really "right" - more goes into the calulation


I don't like this next presentation as well but more different veiwpoints help learning for most people


making the square wave overshoot ~ "the same" by eyeball on the 'scope isn't really a engineering critera - especially for a audio signal - especially with thier "single pole" recovery - at least look for ringing

for stability what you should want is a factor of 3-10 for the shelving noise gain to reach the min stable before the gain intercept - this makes it a function of the noise gain - not a constant 1000x

and you shouldn't calculate with 8 Mhz for LM3886 - completely "safe" would use the min GBW, not the typical spec

and I wouldn't use noise gain in a positive gain configuration without knowing that the source Z is low at frequencies near the gain intercept - not the case if you just hook up to a audio volume pot wiper

although if you can rely on CMRR not messing up apparantly a case can be made for the postive input R

Last edited:
I want to have simple unity-gain power buffer, which I want to drive with tube preamp. I have done it with OPA548/9 and it sounds very good to me, but looking for cheaper and more powerfull chip, which LM3886 fulfils perfectly except unity-gain stability. But it can be done unity-gain stable, just need some help with it.
I want tube preamp to the all the voltage gain (aikido preamp) and LM3886 just current gain. Look here: http://www.diyaudio.com/forums/tubes-valves/226405-simple-hybrid-amplifier-35w-8ohm.html

OK. Interesting.

I wonder if all the extra loop gain you get has to be thrown away in getting it stabilised. I haven't read the article but my guess (and it is a guess) is probably not. So that's good

Actually it comes to mind that there is likely a limit on the input voltage of the 3886 which is likely to be less than full supply. Have you checked that?

Very interesting.

No, I want current buffer, voltage follower. Tube preamp will make all voltage gain. Please read above posted link to the thread "Simple hybrid amplifier".

Dear kacernator, I'm very sorry but what you want to do is not possible with an LM3886.

I *have* read the article, but I have a few objections, pity it's a graphic image and can't cut an paste.
Objections which by the way have also been raised by other posters, both here and in http://www.diyaudio.com/forums/tubes-valves/226405-simple-hybrid-amplifier-35w-8ohm.html

1) the LM3886 is not unity gain stable, that's set internally, there's nothing you can do "outside" to change that.
Or, to be more precise, that can be done in Op Amps where you either use an external compensation cap, (in which case you would use the one needed for "real" unity gain stability) *or* by some design quirk you have access to some internal part of the circuit where you can add it, even if that's not the "official" way to do it.
None of these cases is possible with LM3886.

2) I've read the "trick" described, what he does is exactly what we all suggest: attenuate signal (5X in the example), then provide same 5X gain (what you are trying to avoid) to compensate.

All he does is, instead of having one resistive voltage divider (a.k.a. "attenuator") at the input and another, same ratio one as "gain setting" connected ad NFB, he does both functions with the same resistor set (Rf/Rc).
It does not "look" like conventional attenuating 5X, then having 5X gain, but that's exactly what he does.
He even does all the calculation to prove he has 5X gain , to meet stability criteria.
And he also takes great care to show that you do *not* have the bandwith which would be obtained with "real" unity gain (80 MHz in his example) but the 5X gain one (16 MHz).
Clear as water.

That "hidden in plain sight" attenuator/gain network carries other problems, that's why he writes, say, 80 lines explaining the "trick" (which ends up not being such a big deal after all) and then writes, say, 200 lines explaining "what's the catch?", showing all problems wit this configuration.

Which in your particular case *is* troublesome.

a) the main one: that resistor from your (+) input (high impedance) to the (-) input, which by itself is high impedance too, but is connected to a low impedance NFB network, sure wreaks havoc with your input impedance.
The third line , first column in page 13 says exactly that: "the input Resistor Ri (which would set input impedance) is in parallel with Rc (around 100 times smaller)" WTF? :confused:
Which was noticed by jcx in post #3 .
You simply can't connect this "compensated" LM3886 to a tube plate or to a sensible value volume pot.

By the way, if it happened only at "high frequencies", or it made to appear , say, 100pF across the input, it would be quite bearable, but since it really needs 1000X :eek: the capacitor value needed for compensation, you end having it "at ALL frequencies"; more precisely in your case (didn't check your Math but that's not the point anyway, I trust you) you are effectively placing 1K in series with 2.2uF across your input:eek:

b) since you are *really* having the "stable" gain anyway (11X minimum) you are still having all noise/distortion/whatever implicit .

In fact you are still having dreaded "SS gain", what you are desperately trying to avoid.

Am I dissing your design or goals?
Not at all, I find it a very interesting idea, only I see you won't achieve it with, of all chips in tne World, an LM3886 :(

By the way, what's wrong with the unity gain stable Power Op Amps mentioned in the original design?
Thank you very much for your explanation. I do not have enough knowledge to fully understand the article. That´s pity I can´t use LM3886.

There is nothing wrong with OPA548/549, except that they have max 35W/8 Ohm power, worse distortion than LM3886 and they are quite expensive. OPA549 sounds fine, no doubt, but it could have been better.

I am just looking for unity-gain stable power opamp for two reasons:
1) simplicity of build
2) the buffer has NFB around itself, I don´t want class AB open loop buffer

I know there was or maybe still is unity-gain stable power opamp - LM12, but I can´t find them in store anywhere.

Thanks for your help.
There may be a way you could do this. Mauro gets to use a 3886 in unity gain in his My_Ampli project which I think is also called Ref Freeman of similar and he did it by using an op amp (a humble 318 IIRC). I imagine it worked because it could use the Miller pole in the op amp and just had the 3886 as a follower, though I never bothered to really examine it, it seems to work well.

Do you have somewhere to pin the feedback in your pre-amp? Ie. so you could simply put the 3886 between the preamp output and the feedback loop. It may blow up both units for all I know but someone familiar with your preamp might know.

Or there is a more complicated solution in which you do the same as Mauro but feed the whole signal to the 3886, have the op amp correct it because it has been given a sample of the output and gets a sample of the input too.

Or, another or, you could bridge two of your OPA549s and get the voltage swing you want that way.
Yes, open loop measurement shows the 3886 model has 81 degrees phase margin and 32dB gain margin (details attached, though it's easy enough to do one's own). This is in obvious disagreement with the open loop figure on page 16 in the datasheet so I don't think the model is going to be very useful for this discussion.

One can, in principle, stabilize a 3886 at unity voltage gain via noise gain compensation. Refer, for example, to part 8 of Tim Green's series on op amp stability. The noise can be lower than the more popular approach of using a voltage gain of 10-20 but the required feedback capacitance ends up on the order of 10pF. This is quite difficult for the typical DIYer to hit, much less with the accuracy needed to maintain a 20dB/decade rate of closure over the part's GBP spread. Even with a good PCB fab and surface mount parts it's not something I'd try for; too easy to end up off by several pF due to stray capacitance on traces or the 3886 package's pins.

An easier solution is to operate the 3886 at 26dB local gain (or whatever) for stability within a composite amplifier with a global gain of unity. This is the same basic idea as the attenuation approach proposed earlier in this thread but with better error correction due to the global feedback---this fixes the 3886's poor small signal THD, delivers a noise gain of 2 instead of 21, and limits the output's DC offset to the controlling op amp's input offset.


  • 3886 Avol.gif
    3886 Avol.gif
    31.6 KB · Views: 388
  • 3886 Avol circuit.gif
    3886 Avol circuit.gif
    5.8 KB · Views: 414
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.