CFA Topology Audio Amplifiers

[kgrlee]

Well, i got higher HD with them, slew rate reduced to 840V/µs, and a very strange behavior with square waves, that i don't have in real world.
I don't believe in those models neither.
At least, not more than i believe in simulations.

Christophe, NONE of the MOSFET models in the present Cordell-models.txt are of use for anything except stability & frequency response. Bob warns against their use especially for THD.

You need to wait until he finds his EKV models.

Dunno about slew & square waves.

 

[Esperado]

Christophe, NONE of the MOSFET models in the present Cordell-models.txt are of use for anything except stability & frequency response. Bob warns against their use especially for THD.

Richard, I don't believe neither we can get something consistent in stability and frequency responses, unless we provide all the board's and cabling inductances and parasitic capacitances in the schematic. Utopia.
I just believe in my soldering iron, oscilloscope and ears.
I discovered simulators lately. For me, it is just a nice calculator adapted for our purpose, witch help to faster the design. Witch is something like 20% of the real work, isn't ?
Real transistors calculate their own evils a lot faster in real world ;-)

 

[bimo]

Right now I obsess for making simple amplifier with two stages voltage gain but the distortion is as low as possible and the slew rate is as high as possible. I preferring less than 150 Watt for normal home using. Of course subjectively, it must sound like live .

It will be state of art in engineering

 

[Dave Zan]

Note that my comments about the ULGF apply to the global NF loop. For local loops you can go much higher...

Typically a "local loop" is faster because it does not include the slow output and driver transistors. For instance a local loop for a MIC will include the VAS and IPS only.
But this loop includes the VAS and the drivers and the outputs. All you have saved from the typical MC outer loop is the IPS which is usually fairly fast and has minimal impact.
How can the the elimination of just the IPS from the loop suddenly increase the reasonable ULGF from 3 MHz to 13 Mhz?
Bode never talks about "local loops" or "minor loops" because these are arbitrary (subjective!) terms
He doesn't even use "Loop Gain" because it is poorly defined when possible loops overlap or interact.
He uses Return Ratio, which is precisely defined at each node.

...
IMHO, the Tian probe on the IPS looks at the 'whole' amp and the probe at the VAS input within the Cherry or evil Miller loop looks at the local VAS + OPS loop...

The probe at the OPS output is also within the OPS + VAS, and is easier to use.
I think your latest post in Dado's thread means you have now realised this?
The probe at the IPS is very deceptive.
See JCX's excellent work HERE in the Cordell thread. Also check his next few posts.

This is the approach Cherry takes in his various NDFL & other feedback papers and in fact 'looking before the split' is one way to see how a zillion NDFLs amp increases the 'Return Ratio around the OPS'.

I'm just using the simplest possible NDFL .. one which my single brain cell can unnerstan en analiz

Yes, I also see the OIC as a simple NDFL.
But Cherry's paper has some heavy assumptions somewhat hidden in the maths.

There was a long discussion (I think in Cordell's thread) on whether unstable inner loops would be visible if you looked at outer loops including the Closed Loop response. I side with Bob in thinking this is the case.

I only skimmed that because some of the discussion wasn't too perceptive, especially compared to the JCX post linked above
But Bob is usually excellent so what did he think?

But the real 'proof of stability' is ringing the changes with different evil loads, levels, different devices, overload etc. of the complete amp. Doing this in 'real life' is as, if not more, important than emulating this as I'm doing in SPICEland.

Absolutely.
May I mention that the one time we have both analysed and made our respective recommendations and then had them checked on the same real amplifier that "Pure Cherry" came out as less stable and lower performance
On the other hand, I have had a look at your #500 and think it is very well done. You have tweaked it pretty close to the limits and I expect it will be hard to beat.
If a 13 MHz ULGF with stability just a little short of unconditional is acceptable, of course.
And that's not sarcasm, I really am not sure.

Best wishes
David

 

[Edmond Stuart]

Edmond, do you have, by chance, better models for lateral, I use those from Cordell?
BR Damir

Hi Damir,
No. I'm sorry.
Cheers, E.

 

[Dave Zan]

Cheers, E.

Hi Edmond, you have been a bit quiet.
I hoped you would comment on the OIC/NDFL ULGF issue.
You and Ovidiu probably have more simulation and real world measurements than anyone here, and possibly anywhere.

Best wishes
David

 

[Edmond Stuart]

EKV models

There's a good discussion in Cordell's book.

He points out that the 'usual' MOSFET models are good for stability & response but almost useless for THD. (That's what's in Cordell-Models.txt)

For THD, you need EKV models. Bob had some but seems to have misplaced them. He hopes to find & post them now that he's retired.

Hi Richard,

Bob must have them somewhere (for 2SJ201 and 2SK1530), as I sent them to him (dd 18-10-2007). I'm a little amazed they are not listed in his model file (Cordell-Models.txt)

Lateral MOSFETs are my favourite o/p devices. Their real cost seems to have gone down since I was playing with them in the early 90's. But the last time I burnt solder on amps, 'pure Cherry' seemed allow equivalent performance with VFETs.

These are not my favorite O/P devices. I prefer the verticals: 2SJ201 and 2SK1530.

When we get reliable EKV models, I'll abandon dem evil BJT OPS .. at least in SPICE world.

For those interested in verticals, the 2SJ201 and 2SK1530 models are based on real measurements of Id/Vgs in the weak and moderate inversion regions. See: http://www.diyaudio.com/forums/soli...-interview-bjt-vs-mosfet-252.html#post1295302
Andy_c has translated these data into EKV models, see: http://www.diyaudio.com/forums/software-tools/101810-spice-simulation-30.html#post1304004
and: http://www.diyaudio.com/forums/software-tools/101810-spice-simulation-33.html#post1315345
and: http://www.diyaudio.com/forums/software-tools/101810-spice-simulation-36.html#post1323417
And below are the models for LTSpice as well as MicroCap, which I've sent to Bob.
Please note they may differ a little to Andy's models.

Cheers, E.

 

[Edmond Stuart]

NDFL etc.

Hi Edmond, you have been a bit quiet.
I hoped you would comment on the OIC/NDFL ULGF issue.
You and Ovidiu probably have more simulation and real world measurements than anyone here, and possibly anywhere.

Best wishes
David

Hi David,
Indeed, I'm bit quiet and for several reasons:
- busy with other things (vintage camera repair for example)
- too much noise in this thread
- not interested in NDFL anymore
- the habit of the moderators who delete my posts without any notice, while leaving real BS like this one untouched.
But if you are interested in NDFL like circuits, please look at what Glen Kleinschmidt has done in this field: An experimental 4-th order linear audio power amplifier - Page 1

Cheers, E.

 

[Esperado]

please look at what Glen Kleinschmidt has done in this field: An experimental 4-th order linear audio power amplifier - Page 1.

Very interesting. Thanks for this, Edmond, and +1 for everything else.

 

[Bonsai]

I am headed in the opposite direction. Less complexity. The more I look at NP, the more I like his stuff.

However, I don't want to detract from GK's efforts. He is a pretty smart designer.

 

[manso]

If you are talking about normal loop gains (so 40-60dB at 20 kHz using advanced comp techniques, or 25 to 35 dB using MC) then you have to close your loop (ULGF) at 3 MHz - and this is the upper end on an EF2. Further, at ULGF = 3 MHz, you will need a Zobel and output inductor. If you want to close at higher ULGF . . . only solution is to drop the loop gain so that you ensure that the OPS pole falls well below 0 dB.

If you design a very fast CFA with WIDE bandwidth and keep the loop gain low out to MHz, I can see that you can make a ULGF up at 5 or 10 MHz work. Or, run open loop ZGNF

Earlier on in this thread I proposed that we stle on all distortion assessments at a fixed ULGF of 3 MHz because we saw some designs propositions with very low THD but with ULGF's that were not going to be stable in practice.

Not necessarily, if youre mentioning the design I showed which has ULGF of around 6 (simulated) you should build it and verify for yourself. That design has been in production for two well known and respected companies, one british, the other canadian for over 15 years and is perfectly stable. I doubt that the ULGF is as high as simulated anyway. Another example is the cyrus range of amps which display ULGF > 3MHZ.

 

[manso]

I am strangling with lateral MOSFET OPS and how to tame it. If I go for very low distortion, suffers slew rate, and vice versa. It’s quite easy to get very high SR, about 400 V/usec, but then distortion is more then 10 ppm at 20 kHz, and if I go for below 5 ppm I can’t get SR better then 150 V/usec. I tried modified TPC(with OPS included in one C branch) in combination with shunt compensation (it looks that whiteout it CFA is not very stable), I tried pure Cherry(TPC could lower distortion two to tree times compared to pure Cherry), always some nasty OPS speaks show its ugly head.
Now what distortion and SR level could be set as good compromise. I asked that Richard(RNMarsh) but no answer, and I realy don’t have experience enough to decide. We have a lot of audio experts here, and the books I’ve read said that more then 50 V/usec (not talking mastodon power amp) is not needed for audio.
It looks that we chase SR now as a low distortion chase in VFA amps.
BR Damir

ps. complicate schematic it's not a problem for me as small transistors are cheap, why to make simpler then it's posible(A. Einstein)

Damier Ill send a private mail with some details on how to increase slewrate. It seems you have a lot of time for experimentation. It works with a little different souped up diamond buffer as input but should also with the simple version. You can experiment and if you get better results show it here.

 

[Esperado]

I am headed in the opposite direction. Less complexity.

I believe all of us ?

The more I look at NP, the more I like his stuff.

?? NP= Nelson ?

However, I don't want to detract from GK's efforts. He is a pretty smart designer.

It is a study. Always something new to discover witch can even bring us in the opposite direction...... I just regret the lack of his listening impression comments, so i don't know what to keep of this, unless experimenting it ourself...
It is like what you call AFEC. Tried-it long time ago. The practical problems with phase turns and added complexity VS listening improvements, the need for precise adjustments had made-me abandon-it for industrial purpose.
Reason why i'm waiting with impatience and interest your first real experiments...
Well, at this time OPAs where not fast enough...
Now, they are, but our amps are faster too, so i believe the problem remain the same

About compensations, the only practical way i know is to design, build and experiment real values to make you amp stable enough. An obvious advantage with the simple Miller, isn't it ?
Did some of us had experimented other compensation schemes (Cherry, etc..) for real and found real listening differences ?

 

[Bonsai]

Not necessarily, if youre mentioning the design I showed which has ULGF of around 6 (simulated) you should build it and verify for yourself. That design has been in production for two well known and respected companies, one british, the other canadian for over 15 years and is perfectly stable. I doubt that the ULGF is as high as simulated anyway. Another example is the cyrus range of amps which display ULGF > 3MHZ.

It's very easy to modify one of MHz amps so that the loop closes at 6 or 10 MHz. No need to build a new amp.

I just have to change the comp network.

However, I do know from practical experience that to do so leads to instability problems.

 

[Edmond Stuart]

I am headed in the opposite direction. Less complexity. The more I look at NP, the more I like his stuff.

However, I don't want to detract from GK's efforts. He is a pretty smart designer.

Hi Andrew,

I think he built that thing just to demonstrate that such an amp with a pretty high-order compensation scheme, does work in real life. IOW, just a proof of concept.

BTW, the more simple NDFL amps tend also to be complex, because (most of the time) additional circuitry is needed in order to cope gracefully with overload conditions.

Cheers, E.

 

[Bigun]

Keep the discussion to the topic and don't let it get religious!!

we know that audio folk simply love a good argument, it's invaded all recesses of this hobby. tubes vs SS, AC vs DC heating, FETs vs BJT, feedback vs no feedbck, MC vs MM, Vinyl vs digital, full range vs multi-way ....

I figure this CFA vs VFA thread is a the perfect decoy, it's a completely nonsense topic but by keeping all the argumentative energy in one thread the rest of the forum can enjoy some peace and quiet

 

[wahab]

demonstrate[/I] that such an amp with a pretty high-order compensation scheme, does work in real life. IOW, just a proof of concept.

Hi Edmond ,

This was already implemented in a commercial mass market dedicated
design from Pioneer a long time ago , four amplifying stages with three
nested FB loops , IIRC the user manual even give explanations about
Cherry s NDFL concept....

 

[wahab]

demonstrate[/I] that such an amp with a pretty high-order compensation scheme, does work in real life. IOW, just a proof of concept.

Hi Edmond ,

This was already implemented in a commercial mass market dedicated
design from Pioneer a long time ago , four amplifying stages with three
nested FB loops , IIRC the user manual even give explanations about
Cherry s NDFL concept....

 

[Edmond Stuart]

Hi Wahab,

You might also tell this to Glen. Just subscribe to the other forum. But he did more than just implementing an Nth degree compensation. He also managed to tame clipping issues, one of weakest and troublesome points of NDFL and siblings.

Cheers, E.

 

[Esperado]

I figure this CFA vs VFA thread is a the perfect decoy,

No, Bigun, i believe this thread is about CFA design, according to the thread title.

Not about CFA vs VFA.

It is obvious that everybody interested in CFA design have some personal reason to prefer Current feedback amps... or open minded curiosity.
Nothing wrong as VFA fans are free to talk on specific VFA designs in any other thread.

There is nothing irreligious neither original in CFA. CFA exists till the early beginning of the audio electronic, tube's times, germanium, silicon, FETs, mass market OPAs, produced in hundreds of commercial amps, some highly regarded. This topology have, like all, its own specificities, advantages and inconveniences.

This said, comparing and listing those advantages and inconveniences, and try to find good solutions to get rid of the specific CFA inconveniences (PSRR as an example) , like we did, is nothing else than correct engineering.

More than this, as the both topologies are similar in most of their parts (VAS, OPS) some results of this studies are useful for everybody.

On my side, i find this thread one of the most useful in this forum, despite the noise and ... (read my lips:-).
It is a collaborative study, as long as we can... exchange... in a free and gentle atmosphere.