CFA Topology Audio Amplifiers

godfray (4117)


in such a scheme works poorly (badly). That range of distortions in power and + -40 V input voltage of 3 V (peak)
9bf6a9799644.png
[/URL][/IMG]
http://radikal.ru/fp/2ad30e85f800484883459307a912424f] [/URL]
best regards
Petr
 
... if you can use 'tricks' shape a VFA's loop response . . . you can do the same with a CFA.

My interest was to see how the two circuits lend themselves to such tricks.
For instance VFA have the LTP emitter resistors to play with. CFA seems a bit more constrained by requirements to set the quiescent current.
Similarly the VFA VAS emitter resistors can be decoupled, as Ric Lee frequently points out. Does CFA have similar freedom I wonder?

. . . MIC, TMC, TPC, OIC all bring benefits to VFA's . . . you can use them with CFA as well

Haven't you already applied the CFA version of MIC?
But the others probably.
Want to check if there are any unsuspected tricks with a CFA.

I don't know if the thread has advanced but I personally have learnt.
Even the effort to debunk phoney claims is sometimes educational.;)

Best wishes
David
 
Hi Richard,

Do we know of any commercial CFAs that have been reviewed by John Atkinson in the last 10 years or so? It would be good to see how many, and what was said about them.

Cheers,
Bob

There are too many to mention. Also a price point should be chosen form 300 to 70 000 dollars. For cheaper unit try any Nad or better still Cyrus 6, multi award winner. Marantz also has good examples, PM6004, award winner as is successor PM6005 and PM9S2, best amp in the 5000 dollar range. Another heavy wheight is Krell and for exotics at 70 000 dollars there is Daniel Hertz, intrigingly Mark levison is the designer and owner of last example.
 
My interest was to see how the two circuits lend themselves to such tricks.

Similarly the VFA VAS emitter resistors can be decoupled, as Ric Lee frequently points out. Does CFA have similar freedom I wonder?

Best wishes
David

To some extent you can use decoupled emitter resistors in CFA. Don't ask me about the theoretical details but have used them in my prototypes. Have used up to 150pF (across 22R emmiter resistors - beta enhanced VAS) without problems. Too big and the amps oscillated.

The factors behind these oscillations are still a mystery to me... There may have been other instabilities in the amplifier as well that may have been an issue...
 
Last edited:
www.hifisonix.com
Joined 2003
Paid Member
My interest was to see how the two circuits lend themselves to such tricks.
For instance VFA have the LTP emitter resistors to play with. CFA seems a bit more constrained by requirements to set the quiescent current.
Similarly the VFA VAS emitter resistors can be decoupled, as Ric Lee frequently points out. Does CFA have similar freedom I wonder?



Haven't you already applied the CFA version of MIC?
But the others probably.
Want to check if there are any unsuspected tricks with a CFA.

I don't know if the thread has advanced but I personally have learnt.
Even the effort to debunk phoney claims is sometimes educational.;)

Best wishes
David

For instance VFA have the LTP emitter resistors to play with.

You can adjust the emitter degen resistors in the diamond output pair.


CFA seems a bit more constrained by requirements to set the quiescent current.

I assume here you refer to the diamond Iq. Easily done. You 'stand off' the bases of the diamond output devices using resistors between the input buffer emitter and the base of the diamond output devices. I used this in both the sx and nx amps. Another option is to use a diode instead of a resistor.

Similarly the VFA VAS emitter resistors can be decoupled, as Ric Lee frequently points out.

This simply raises the gain of the LTP at HF - not sure what the specific benefit of this is other than noise reduction; you still have to close the loop, so I expect Cdom would have to be increased to counter this, or, you'd have to be doing this on a MIC comp'd circuit where it might make sense. But, if you can already get to 2-4 ppm without it, whats the 'ear' benefit?

Haven't you already applied the CFA version of MIC?

Its what I've used in the sx and nx-Amps. See the Mark Alexander app note referenced in the back of the write up - a detailed treatement is given in the appendix wrt CFA topology.

Does CFA have similar freedom I wonder?

I think both topologies have their freedoms. It just depends on what kind of tradeoffs the designers want to make. For me, anything below 20ppm is adeqaute, and adding complexity to get it lower is wasted, with SR at least 1V per peak output volt. PSRR should be as high as possible, but I would not stop using CFA because its lower than VFA - I'd just do something to mitigate it.

Having built some of these things (VFA aand CFA), a lot of what is considered important actually makes little or no difference to the the listening perception. I get as much pleasure out of my 15W class A CFA sx-amp as I do the 180W VFA e-Amp, which produces >10x the output power.

:cool:
 
Last edited:
www.hifisonix.com
Joined 2003
Paid Member
To some extent you can use decoupled emitter resistors in CFA. Don't ask me about the theoretical details but have used them in my prototypes. Have used up to 150pF (across 22R emmiter resistors - beta enhanced VAS) without problems. Too big and the amps oscillated.

The factors behind these oscillations are still a mystery to me... There may have been other instabilities in the amplifier as well that may have been an issue...

If you have dcoupled the VAS emitter resistors, you have probably raised the loop gain at HF or introduced a zero in the response and reduced the PM. In my answer to Dave below, I was refering to decoupling the LTP degen resistors, which also raises the loop gain and the ULGF if large cap values are used (uF). The cure in this case it to increase CDOM to return to the original ULGF.

For low values of decoupling (10's to 100's of pF) I suspect you may just be inserting a zero the response - also a possible cause of instability.

Best thing to do with these experiments is to build a model in LTspice and look at the PM, GM and ULGF as you make the changes.
 
Irritatingly, about 12, depending on the details, the way I was doing it.

The circuit below illustrates the idea. in this version, VAS current gain = hFE*2/3 and VAS quiescent current = 4 * input stage quiescent current. One could obviously rationalise the mirrors to save a few transistors. I just drew it like this, with same-value components throughout to make the mirror ratios obvious at a glance.
[...]

Note that emitter resistors can't be used with the transistors in the yellow box as it relies on the exponential Ic/Vbe relationship to work properly.

Hi Godfrey,

Thank you for providing the details of your circuit. In the mean time, funny enough, I came to almost the same realization for a 'constant product TIS' and to the same 'irritating' number of extra transistors (see 1st figure, Q7, Q10...Q20). The 2nd figure shows the step response (blue) and the TIS collector currents (red and green curves). As expected, when Ic of the one TIS doubles, Ic of the other TIS stays far from zero. That's nice, but is it worth all the trouble? NO! Moreover, just as you did, I also observed an increase of the 2nd harmonic distortion. There are better and simpler ways.

The 3rd figure shows a CMCL (comprising 6 trannies instead of 12) that keeps the sum of the TIS collectors currents constant, thought with a twist. Please take a look at fig. 4. Here, the TIS that 'normally' should be switched off, keeps running at nearly the same collector current (see green curve). How is that possible? Very simple. C2, respectively C3 are responsible for that. During a positive going step, for example, Q18 is hard turned on and the output will rise. So far so good. However, this rise will also turn on (more or less) the opposite TIS via C3. That's the crux. More on this subject (and a funny quibble with Bob) can be found here.

Cheers, E.
 

Attachments

  • AMP26-FE-CPR.PNG
    AMP26-FE-CPR.PNG
    28.5 KB · Views: 202
  • AMP26-FE-CPR-STP.PNG
    AMP26-FE-CPR-STP.PNG
    2.6 KB · Views: 196
  • AMP26_FE.PNG
    AMP26_FE.PNG
    28.1 KB · Views: 207
  • AMP25-FE-STP.PNG
    AMP25-FE-STP.PNG
    3 KB · Views: 205
This simply raises the gain of the LTP at HF - not sure what the specific benefit of this is..

It allows you to improve the phase near ULGF if the zero is properly placed i.e. a Bode step.
Cherry discusses the use of this capacitor too in JAES but the maths is opaque despite several reads, more effort required.
Anyone understand this? I don't think he means a Bode step.

I reread my post in your reply and realised it was poorly written, could be read that I meant you when I wrote of phoney claims.
What I meant was that I had learned from and appreciated the thread and that, unlike JCX, I didn't even mind the dubious posts.
I don't think your contributions fall into that class, quite the contrary.

Best wishes
David
 
Last edited:
www.hifisonix.com
Joined 2003
Paid Member
It allows you to improve the phase near ULGF if the zero is properly placed i.e. a Bode step.
Cherry discusses the use of this capacitor too in JAES but the maths is opaque despite several reads, more effort required.
Anyone understand this? I don't think he means a Bode step.

I reread my post in your reply and realised it was poorly written, could be read that I meant you when I wrote of phoney claims.
What I meant was that I had learned from and appreciated the thread and that, unlike JCX, I didn't even mind the dubious posts.
I don't think your contributions fall into that class, quite the contrary.

Best wishes
David


Dave, I tried that technique when I was developing the e-Amp with low capacitance values (as I note in my reply, high values are another matter). It still does not stop the fact that you have to close the loop where the phase margin is adequate. I doubt in a high loop gain VFA it allows you to push the ULGF up significantly - but of course I am open to correction - I may need to revisit it. And, IMV we are now getting into advanced comp tecniques.
 
Last edited:
He also suggested a 3rd- order filter for the servo. Dang ! I found a filter from heaven (below 1). In according with "KISS" (keep it stupid simple), I am at 1 IC and 8 little passives.

What I have is a "cliff -like filter" ( 60db/decade) and good gain below 5Hz
to decimate any DC.
I suggest you check LF stability wih your zillion dB/decade servo .. or at least design your PCB so you can revert to a simple 6dB/8ve roll-off.
 
Set -3dB corner to below 1hz.

THx-RNMarsh

Yes , I will do ....

I have been sick (flu :(:( ) ... That got me thinking tempco.

As was mentioned (the blameless) is (almost) perfect , the two CCS's for
the LTP and VAS nearly negate themselves. It is not TOTALLY perfect , as
the 2 CCS's have different Ic and device types.
Still , it will be pretty accurate from a cold house to a hot summer day.

I sort of "lucked out" with my first CFA design , the hawksford's led cascode bias
negates the led IPS CCS's I used (real close). I found that a string of 2 led's for
the cascode could perfectly offset the use of a 2Q CCS at the input stage.

I wondered why my HK680 could be put outside (-10C) ... brought inside ,
and still measure within a mA of my set OPS bias. They "fine tuned" the
IPS (near perfect tempco) and matched it with an exceptional OPS Vbe setup .

It might be "anal", but I like that kind of performance . Most amps will never
see this extreme ambient swing or have their stages biased "hot" to override
external conditions.

I like cool :cool: .

PS - I thermally simulated 5 of my favorite VAS's in isolation , the hawksford
has 1/5th the Ic "drift" of the others.

OS