Loop gain question

[Pingrs]

Hi Lumanauw,

“The most obvious one is with feedback amp with CFP output stage. I suspect this for kind of amp Zobel is "Forbidden"”(sic)

Does Doug Self’s Load Invariant design count as one of these. If so, the traditional 10R-100n Zobel works fine on mine.

Regards,

Brian.

[lumanauw]

Hi, Pingrs,

Good question I don't know much about theories, but it seems CFP works the best at classA or heavy biased.

I cannot find load invariant in my Doug Self book (mine is 1st edition). But I can find 2 design of his in my book.

The first one is "Blameless amp" where this is "optimally biased classB" or cold amp and the other one is "ClassA" amp.

Interesting that DougSelf uses 2 different output stages for classB and classA. For blameless he uses EF, for classA he uses CFP. Both has zobel 10R+100nF.

He don't use CFP for low (optimally) biased classB.

I forgot to wrote that my experiment is with cold amp (low biased classAB). In this state, CFP seems don't like Zobel.

As far as I know classA (whether EF or CFP) seems far more stable (in any angle) than low biased classAB. You can een omit zobel or inductor with good designed classA.

[Christer]

Originally posted by AndrewT
Hi,
I am not sure of the reason for the Miller comp method sounding less good.
I think it more likely it has to do with "other reasons" than open loop gain or 2pole correction.
My thoughts, maybe it has something to do with
overloading the input LTP collector?
or slew rate?
or wrapping very few stages (VAS=1) inside the bulk of the amp feedback?
or very little feedback left? to attenuate non linearities from the other stages particularly at higher audio frequencies.
or almost no feedback left to correct high frequency intermodulation descending into audio band?
or almost no feedback left to taper the gradual attenuation of high frequency harmonics that form the supersonic spectrum?

Thanks Andrew,

You raise many interesting points.

The slew rate problem should be handled by designing for sufficient slew rate, and then choose the input LP filter such that we never get a slewing problem. However, as you point out, the Miller cap still presents a heavy load for the diff pair at high frequencies. I have thought about that too, and maybe that will cause a rising distorsion at high frequencis that could be a problem?

I find the point about HF IM interesting. Of course, the problem won't go away if we do the compensation otherwise, but it will shift to higher frequencies, at least, which might be a good thing. However, I think two-pole compensation would do an equally good, or better job here? Or maybe not, since we get more OLG up to a certain frequency, but less OLG above it, due to the steeper slope?

Anyway, lots of interesting things to think about.

[Christer]

Originally posted by lumanauw

It can do that (Flat OLG)? If it can, I'm very-very interested.

Hi Luamanauw,

the idea of two-pole compensation is to use a CRC network instead of a single Miller cap. This means you can keep the OLG flat to a much higher frequency, since it will fall off twice as fast above that (12 dB/octave instead of 6 dB/octave). An example of two-pole compensation can be found here
http://www.diyaudio.com/forums/attac...amp=1037049279
although in practice it is usually better to to tie the resistor to the rail than to ground, to improve PSRR. Randy Slone discusses this technique in hit book, and I think Self does too.

[AndrewT]

Thanks Christer for thinking about it.
Better than many who just follow blindly, or worse repeat balderdash.

[Johan Potgieter]

Andrew and Christer,

Re Miller compensation: Andrew, just about all of your points apply depending on the circuit, as I found it. To sum up:

The LTP collector(s) could be punished in a case of large Cdom, though this is seldom the major problem.
Slew rate will suffer if VAS can't cope (remember it usually must go the complete output swing). Recall how JLH and others feed Cdom from the output rather than the VAS collector for extra drive capability.
Also, as feedback at h.f. falls off, most amps (even Douglas Self's one) are left with increased distortion above some 8 KHz. This can be especially bothersome at supersonic frequencies where audible intermodulation products can be generated.
I found (in Spice analyses) that high order harmonic generation is sensitive to Miller compensation probably through a combination of the above factors. As said, it depends heavily on circuit design, transistor parameters,etc. That also brings in another member's remark regarding high ft's chosen. With BJT spreads one would rather have compensation under control using a 5% capacitor, instead of relying on varying ft's.
In the quoted Ellis article the case is made for the more instantaneous effect of lead compensation (i.e. a capacitor over the feedback resistor). In conclusion, some well-designed circuits do work well with Cdom compensation; one does not condemn that outright. But in my experience and perhaps that of others, I feel more comfortable in keeping such compensation as far out of the audio spectrum as possible (I already mentioned that I succeeded in reaching a -3 dB point only at 30 KHz). I cannot confirm with personal listening tests - sadly my hearing goes no higher than 7 - 8 KHz at present, but others seem to hear an improvement in such a system. But then hearing is such an individual thing that I would not rely on that solely to characterise an amplifier as good or bad (and no offense intended; it is just simply a human characteristic, plus the psychology involved - but that is another story.)

Regards.