Why does Class A distortion increase with frequency?

[tcpip]

This is a basic question which I guess many of you will be able to help me with. I've been reading about distortion in classical power amp topologies (Lin 3-stage with complementary pairs at OPS, either EF or CF, ignoring QC for now), and I can't understand why a Class A power amp built in this topology will have its THD curve rising with frequency after being flat till a couple of KHz.

I'll tell you what I've understood. It seems that a well designed Class B amp using the classic Lin type topology (both single-ended and mirror-image) can drive down most other types of distortion almost to the noise floor, but the crossover distortion remains. GNFB linearises this crossover distortion. And since the compensation capacitor reduces global open-loop gain at HF, therefore the GNFB itself keeps reducing with rising frequency. Since the linearising influence of the GNFB reduces, the crossover distortion becomes more and more visible at the output with rising frequency.

And from what I understand, Class A has no crossover distortion.

In that case, why does a Class A amp show a rising THD curve with frequency?

Note that I'm not debating that the Class A amp's distortion may or may not be audible. I'm just asking, why should the distortion curve rise at all if there's no crossover distortion (and other distortions can be driven down into the noise floor anyway)?

I also know that everything I've read (just about two and a half books and lots of NS app-notes and forum threads) does not cover the non-Lin topologies e.g. the ones Nelson Pass designs, etc. So my comments may not apply to those topologies.

[woody]

This is a verry simplistic answer but it may be the capacitences
of the active devices. Take a typical output device with a HFE
of 100 passing 1A well that HFE is at DC. When we move from
DC to AC we have to start charging and discharging the base emiter capacitence and as the frequency rises this requires more current. Well 10ma may turn that device on and off at 2 or 3 hz but at 20khz it will take a lot more current which puts a bigger load on the driver ect.

I said this was a verry simplistic answer and I am sure there
are a lot of other factors. In a mosfet the capacitences are
different but still easier to drive at lower frequencys.

[Christer]

For all types of amplifiers using GNFB, you have the problem that above some frequency, the open loop gain of the amplifier starts to drop. Since the feedback strives to maintain the same closed loop gain the result is that the amount of feedback starts to drop when the open loop gain starts to drop. That also means that it becomes less effective in reducing distorsion. This is why many prefer amps with a flat open loop response up to at least 20 kHz, but the (most) important thing is to have enough open loop gain at 20 kHz so the feedback has something to work with.

Then there will also be various effects from capacitances etc. that affect the distorsion of the amp before applying feedback.

[janneman]

I agree with Christer. Dropping ol gain with F means less feedback available (which is OL gain - C gain) to reduce distortion with higher freq. And, it's not just for class A, it's with almost all amp types.

Jan Didden

[Mooly]

Hi,
Crossover distortion is only part of the problem.
Class A = no crossover distortion, but the input stage and voltage amplifier stages all contribute to the overall distortion figure.
The topology used, LTP or single ended input determines the "characteristic" of the distortion, for e.g. the rate of rise with increasing frequency and it's harmonic stucture.

[tcpip]

Originally posted by janneman
I agree with Christer. Dropping ol gain with F means less feedback available (which is OL gain - C gain) to reduce distortion with higher freq. And, it's not just for class A, it's with almost all amp types.

This far, I too have been able to understand. The Miller cap or compensation cap is responsible for providing local feedback to the VAS and this reduces the open-loop gain.

My point was: you need OL gain to generate GNFB so that you can linearise the distortion. But if, as Self says, a well-designed Lin 3-stage amp has pretty much no other distortion left other than crossover distortion, and I operate the amp in Class A, then there should be no distortion left to linearise. If I've understood Self right, then the dropping GNFB level should not result in increase of distortion, because there should be no distortion left to linearise.

That's really where my confusion started. What am I missing?

[tcpip]

Originally posted by Mooly
Class A = no crossover distortion, but the input stage and voltage amplifier stages all contribute to the overall distortion figure.
The topology used, LTP or single ended input determines the "characteristic" of the distortion, for e.g. the rate of rise with increasing frequency and it's harmonic stucture.

I don't have enough knowledge to have an educated opinion of my own, but going by what I've read in Self's book, the other sources of distortion can be brought down almost to the noise floor. The only one which can't is crossover distortion.

[Christer]

Originally posted by tcpip

This far, I too have been able to understand. The Miller cap or compensation cap is responsible for providing local feedback to the VAS and this reduces the open-loop gain.

My point was: you need OL gain to generate GNFB so that you can linearise the distortion. But if, as Self says, a well-designed Lin 3-stage amp has pretty much no other distortion left other than crossover distortion, and I operate the amp in Class A, then there should be no distortion left to linearise. If I've understood Self right, then the dropping GNFB level should not result in increase of distortion, because there should be no distortion left to linearise.

That's really where my confusion started. What am I missing?

I think Self means that after you have applied feedback, there isn't much distorsion left other than crossover distorsion. Crossover distorsion is harder to remove by GNFB. That means a class A amp should be very linear with GNFB since it has no crossover distorsion.

If Selfs amps were so free of distorsion already in open loop, why would he even bother with a lot of GNFB?

BTW, I have noted that Self is nowadays a member on this forum, so perhaps he reads this thread and can comment himSelf on what he means.

[tcpip]

Originally posted by Christer
I think Self means that after you have applied feedback, there isn't much distorsion left other than crossover distorsion.

Yes, this is what I'd understood.

Crossover distorsion is harder to remove by GNFB.

This is not how I'd read his text. I'd gotten the impression that he means crossover distortion could be removed if there was enough GNFB available, but there isn't enough at high frequencies.

That means a class A amp should be very linear with GNFB since it has no crossover distorsion.

Exactly. Then why does the distortion rise at HF?

If Selfs amps were so free of distorsion already in open loop, why would he even bother with a lot of GNFB?

In fact, I never got the impression that he is suggesting that his amps are low-distortion without GNFB. Well, he does talk about their open-loop distortion, and he does say that a good design should have open-loop distortion lower than a bad design (my words, not his), but he never says that his amps have acceptably low distortion in open-loop mode, IMHO.

BTW, I have noted that Self is nowadays a member on this forum, so perhaps he reads this thread and can comment himSelf on what he means.

Wow, not bad. Hope he or some of the others chip in.

I wonder whether Nelson Pass' super-symmetrical Class A designs also have THD rising with HF?

[tcpip]

Originally posted by Christer
BTW, I have noted that Self is nowadays a member on this forum, so perhaps he reads this thread and can comment himSelf on what he means.

He's done something like seven posts in a year. Not likely.