Hi all,
could not resist and made a small exercise in analysing Cherry's example 60W NDFL amp from ETI 1983.
An important reslt is that the OPS may be stable in fact (phase margin 70 degrees under different loads with darlington both in OPS and VAS), if one accepts the high ULGF of 10 MHz. But this figure probably also can be reduced, in case one feels uncomfortable.
Furthermore, I found quite low ULGF especially in the global loop, i.e. low NFB around the IPS: only 20dB at 20kHz.
I think, if one carefully re-examines the whole circuit, including e.g. other distributions of the ULGFs, degenerated Rush stage, higher NFB around IPS etc., then one could end up with a very attractive circuit. Already now, THD20k-values in the ppm range (up to medium power levels) are possible with acceptable and stable stability margins in all loops, with single OPS transistors and OPS bias below 100mA.
In some of the examples in the text below (THD behaviour and investigation of the total NFB around OPS), I made a failure in chainging Cherry's choice of 470pF for the capacitor at the Rush stage input to 100pF. This seems to bring more total NFB around the OPS, but reduces stability margins in global loop and intermediate loop. Nevertheless, it is still in the text; it shows how easy one can run into difficulties when just "playing around". I assume that THD figures won't change much when going back to Cherry's value of 470p.
Cheers,
Matthias
View attachment ndfl.pdf
could not resist and made a small exercise in analysing Cherry's example 60W NDFL amp from ETI 1983.
An important reslt is that the OPS may be stable in fact (phase margin 70 degrees under different loads with darlington both in OPS and VAS), if one accepts the high ULGF of 10 MHz. But this figure probably also can be reduced, in case one feels uncomfortable.
Furthermore, I found quite low ULGF especially in the global loop, i.e. low NFB around the IPS: only 20dB at 20kHz.
I think, if one carefully re-examines the whole circuit, including e.g. other distributions of the ULGFs, degenerated Rush stage, higher NFB around IPS etc., then one could end up with a very attractive circuit. Already now, THD20k-values in the ppm range (up to medium power levels) are possible with acceptable and stable stability margins in all loops, with single OPS transistors and OPS bias below 100mA.
In some of the examples in the text below (THD behaviour and investigation of the total NFB around OPS), I made a failure in chainging Cherry's choice of 470pF for the capacitor at the Rush stage input to 100pF. This seems to bring more total NFB around the OPS, but reduces stability margins in global loop and intermediate loop. Nevertheless, it is still in the text; it shows how easy one can run into difficulties when just "playing around". I assume that THD figures won't change much when going back to Cherry's value of 470p.
Cheers,
Matthias
View attachment ndfl.pdf
I don't even have a photocopy of the Scott and Spears article. but will visit the library today.
Then at least I can tell you what's in it.
That's why I did one for Linear Audio.
You are very quick.
Your comments about the possibly excessive ULGF around the OPS are consistent with my posts to Richard about his similar Cherry style amp.
You can find them a while back, but the main point was that Richard's amp has ULGF almost 20MHz!
One recommendation is that it is safer to use Tian probe.
The OPS loop is very low impedance out so your technique is fine but some of the internal loops may be less predictable.
JCX had an example of this in an error correction amplifier that I still have not worked out.
Best wishes
David
Thank you very much in advance!
This will be interesting to read.
In this case, as the "right-hand-side" of the probe is driven by the 100nF "Zobel" capacitor, I think that also the internal loop evaluations will be rather reliable at high frequencies. But you are right, I should try the Tian probe, and be it only for exercise.
When thinking about the circuit, I aggree with Edmond who wonders about the combination of subsequent poles and zeros in the feedback networks. I think that these are result of the application of the rather general NDFL theorie, without further simplification. Maybe, a combination of the output-inclusive Miller compensated VAS/OPS (slowed down to a few MHz ULGF) with a (or two?) nested Miller compensated stage -- along the lines of my current amp, ULGF e.g. around 1 MHz, the same for the global loop -- is simpler and more sensible. The overall structure with intermediate Rush amplifier is the same, anyway.
Best regards,
Matthias
Last edited:
Matze, can you post your *.ASC model please.
______________
I second Guru Zan's recommendation to use a Tian probe. It's in
Program Files/LTspice4/examples/Educational/LoopGain2.asc
I show several examples of its use in this thread.
It's just as easy (in fact easier) to use as the cruder, more primitive probes and much less prone to error. Not having check that your driving & load Zs are appropriate is a great simplification.
See below, ndfl-darl.asc.
Additionally, I have tried a version with further reduced ULGF around the OPS and with my modified nested MC instead of NDFL. It performs quite well. Of course, the current sources would have to be included and all the clipping / short-circuit protection stuff is also missing. See nmc.pdf and nmc.asc
I apologize for still using the Middlebrook probe (I'm quite confident about the impedances) and the rather brute-force method to ensure good accuracy of transient simulation -- the isolated voltage source just spinning at 2 MHz (I do know the "correct way" ;-)
Matthias
View attachment ndfl-darl.asc
View attachment nmc.pdf
View attachment nmc.asc
Error with VAS transistors
In both above versions (NDFL and NMC), I swapped transistor types in the darlington VAS. The VAS was 2N5401, while the "pre-VAS" was BD140.
In the NMC version below, I corrected to VAS as BD140 and "pre-VAS" as BC560. This required a small change in the VAS degeneration (100R||47p --> 100R||33p) and the shunt compensation (10R+220p --> 0R+270p). Then, the ULGF of VAS/OPS alone again is around 4 MHz, and the rest may stay unchanged. Results are even marginally better.
Matthias
View attachment nmc.asc
In both above versions (NDFL and NMC), I swapped transistor types in the darlington VAS. The VAS was 2N5401, while the "pre-VAS" was BD140.
In the NMC version below, I corrected to VAS as BD140 and "pre-VAS" as BC560. This required a small change in the VAS degeneration (100R||47p --> 100R||33p) and the shunt compensation (10R+220p --> 0R+270p). Then, the ULGF of VAS/OPS alone again is around 4 MHz, and the rest may stay unchanged. Results are even marginally better.
Matthias
View attachment nmc.asc
Last edited:
I'm not sure who and on which thread posted a link to yet another Richard http://www.personal.reading.ac.uk/~shsmchlr/selected.htm
He isn't a mere electrician but a horny handed mechanical .. but the mechanicals usually have a better grasp of control theory than us electricians.
http://www.personal.reading.ac.uk/~shsmchlr/miscfile/asm2003.pdf
http://www.personal.reading.ac.uk/~shsmchlr/miscfile/scarpbode.pdf
http://www.personal.reading.ac.uk/~shsmchlr/miscfile/asm2004.pdf
http://www.personal.reading.ac.uk/~shsmchlr/miscfile/C2004RJMitchell.pdf
http://www.personal.reading.ac.uk/~shsmchlr/miscfile/ac2005RJMitchell.pdf
are all worth reading for those wanting a rigorous analysis of different loops.
For Guru Zan, he gives an practical implementation of Bode's irrational element. His lectures mostly show this important version of Bode's Max F/B design as in the attached pic.
I used to calculate something very similar when I was playing with TPC in da old days. You see this in the important Zan/Tian Probe (ZTP) at the output for properly designed 'pure Cherry', TMC and probably TPC and other advanced methods too.
His Bode Step is my Nyquist kink.
There is another useful 'Bode approximation' which has 6dB/8ve until the ULGF and then 12dB/8ve above that. You see that in the 'outer' loop ... but the ZTP will show Mitchell's behaviour.
http://www.personal.reading.ac.uk/~shsmchlr/miscfile/CS2007RJMitchell.pdfShows Cherry's Nested Differentiating Feedback Loops.
Attachments
Thanks for the comprehensive overview!
Will have a good time to go through the past of this thread and have o look at your references; I realise that I jumped into the discussion without having read enough all the many previous posts.
Probably, with the question to David, I did not express myself correctly enough.
The real question in my head was: Given Cherry's 60 W NDFL example amplifier from ETI 1983, has anybody done a detailed simulation analysis and discussion of the different loops and their interactions, using the concrete circuit, i.e. taking into account all the deviations and additional margin losses due to the actual active devices in the circuit?
This question came up, as it was unclear to which extend -- and in which direction -- the intermediate NDFL loop and the global loop with its chained pole and zero influence the effective margins in the critical innermost loop around the OPS.
The suspicion or hope was that this innermost loop might show improved stability margins with the additional loops in action, too. My simulations above seem to indicate that this is not the case, at least with the VAS/OPS implementation I used as example. Perhaps, your references may clarify why the hope was wrong.
(My hope was driven by the observation, that Cherry only suggested output-inclusive compensation in his NDFL examples, and not in his other papers, e.g on sensitivity analysis in an audio amplifier.)
Kind regards,
Matthias
Will have a good time to go through the past of this thread and have o look at your references; I realise that I jumped into the discussion without having read enough all the many previous posts.
Probably, with the question to David, I did not express myself correctly enough.
The real question in my head was: Given Cherry's 60 W NDFL example amplifier from ETI 1983, has anybody done a detailed simulation analysis and discussion of the different loops and their interactions, using the concrete circuit, i.e. taking into account all the deviations and additional margin losses due to the actual active devices in the circuit?
This question came up, as it was unclear to which extend -- and in which direction -- the intermediate NDFL loop and the global loop with its chained pole and zero influence the effective margins in the critical innermost loop around the OPS.
The suspicion or hope was that this innermost loop might show improved stability margins with the additional loops in action, too. My simulations above seem to indicate that this is not the case, at least with the VAS/OPS implementation I used as example. Perhaps, your references may clarify why the hope was wrong.
(My hope was driven by the observation, that Cherry only suggested output-inclusive compensation in his NDFL examples, and not in his other papers, e.g on sensitivity analysis in an audio amplifier.)
Kind regards,
Matthias
Last edited:
That was JCX, a while back, and he reposted it recently.
An excellent reference, as usual from him.
The interpretations in terms of mechanics made me think and provided a new perspective - so it's velocity feedback, hmm.
Best wishes
David
I have not done a simulation of Cherry's specific circuit.
It only recently became clear to me how to analyse the theory of this kind of nested loop structure.
On a practical level, as you know, I don't think the extra complexity is needed.
And I am in the middle of the work to see if there is any theoretical benefit.
Actually have the Scott and Spears paper about half read when I saw your post.
My observation is that the intermediate NDFL loop reduces the stability of the inner loop.
In a practical circuit you can be lucky and have the real transistor capacitances act as implicit compensation.
But that's a different issue.
In a follow up paper to the sensitivity analysis he does cover OIC.
The two papers were in the same issue of JAES.
Best wishes
David
Last edited:
Cherry's original 60 W NDFL amp
I have re-done the above two exercises for the circuit Cherry proposes in ETI 1983, i.e. with single-transistor VAS. It has very good stability margins in all loops, ULGF is around 5MHz. (see below)
But according to our current standards, THD performance is rather weak.
Kind regards,
Matthias
View attachment ndfl-orig.pdf
View attachment ndfl-orig.asc
I have re-done the above two exercises for the circuit Cherry proposes in ETI 1983, i.e. with single-transistor VAS. It has very good stability margins in all loops, ULGF is around 5MHz. (see below)
But according to our current standards, THD performance is rather weak.
Does Cherry use a darlington VAS in the OIC case? If yes, it would be really interesting to read/analyse that. This would be a strong support of KGR Lee's opinion which led him to start this thread.
OK, accepted! The whole question for me is/was whether NDFL may ease the application of OIC. Unfortunately, it does not seem to be the case.
Kind regards,
Matthias
View attachment ndfl-orig.pdf
View attachment ndfl-orig.asc
Last edited:
Yes, not a surprise, less ULGF and more THD.
Earlier in the thread I commented that Ric Lee's Cherry amp had excellent THD but that was mainly because the ULGF was ~20 MHz.
The THD to ULGF metric looked exceptional because he compared the THD to the much lower ULGF of the outer loop, which is appropriate in Miller Compensation but not OIC.
I haven't yet worked out if that is all the explanation.
Yes, see my earlier quote to Ric a few days back.
In what way?
OIC is NDFL. It is just the simplest case with one nested loop.
I read the rest of the Scott and Spears paper after my previous post.
It is explicitly intended to discuss NDFL, where there is an additional amp section, but applies equally to OIC.
Useful, but unfortunately it contains a fairly serious mistake.
The nature of this error is educational but makes the conclusions unreliable.
It is clear the extra loop does affect the stability of the inner loop.
This is the part they had incorrect, so it requires some more work to be clear to me.
In practice the effect is usually detrimental but very small.
As I wrote to you way back in our first discussion, there are conservation laws for various feedback quantities so most of the apparent improvements are illusions or errors.
Cherry commented on the Scott and Spears paper with a more or less similar conclusion.
Exactly what improvements can be made within the limits of conservation still escapes me.
I need to read Cherry's more detailed paper, not the JAES versions.
Best wishes
David
Last edited:
Thanks for these.
The darlington VAS appears in Feedback, Sensitivity and Stability of Audio Power Amplifiers but I'm not sure he tries that in his 'pure Cherry' NDFL amps. Probably not as it will have 4 roll-offs within his loop.Dave Zan said:
My $0.02 is that by judicious decoupling of the VAS emitter, it is possible to get away with zillion MHz ULGF around the loop that the Zian/Tian output probe looks at.
I have some 'real life' evidence of this from 2 decades ago.
I claim that in Jurassic times, wen i kud reed en rite, I modified his matrix method to investigate this.
I would really like to do more 'real life' work on this.
In SPICE world, looking at stability of the 'outer' loop, the Zan/Tian Probe loop AND Closed Loop gain shows an interesting divergence of stability between the 'outer' loop and the Zan/Tian stability. Making one loop more stable, makes the other loop less stable over an important range of Cherry cap values.
There is often a region where you see 2 or three peaks in Closed Loop corresponding to the 'stability' of the 'outer' loop, the ZTP loop ... and also the interaction of a capacitive load with the Zobel/Inductor.
But when the amp finally goes unstable, the 'outer' loop and the Zan/Tian probe do so at exactly the same point so their 'stability' converge again.
I don't think we can do more in SPICE world. In da 'real world' it is essential not to have VAS & OPS too far from each other if we want both in the same zillion MHz ULGF loop.
The benefit? 1ppm THD20kHz at 50W 8R from a very simple circuit using quite ordinary devices as in #4 If this is stable, I don't see the need for any further complexity. But we need to try it.
Dr. R Mitchell's stuff allows you to design for a particular PM & GM.
In the pic I posted, he shows the uncompensated gain. Anything you do has to lie within this ... and eventually match it at very high frequencies. One could load the 'uncompensated gain' with parasitics galore and this might give us a better idea of what is possible.
Many of the zeros that might be used to tweak the 'outer' loop have little effect on the ZTP loop. Cherry points out that ZTP stability is usually independent of the main feedback network. In fact the only 'zero' that is effective is that across the VAS emitter resistor.
Last edited:
Not as far as I can see either.
I remembered that he did it in the one article but I was incorrect in my recollection that it was in the Nested Loops too.
Thanks for the correction and sorry to misinform Matthias.
I am up for this.
Should I buy a .1 to 250 MHz oscillator yet?
The inner loop tells you all you need.
Found confirmation from Dr Cherry himself in his comment to the Scott and Spears article.
Inevitably.
Yes, JCX posted it in response to some claims of mine, so I read it very studiously.
This is perhaps an alternate expression of my point that the inner loop response is all you need to ensure stability.
Yes. it is hard to tune the inner loop.
This may be the usefulness of NDFL.
The theoretical limits are equivalent but the extra points to add zeros are helpful.
Also you can put poles in the feedback loop that are equivalent to zeros in the forward path, but easier to implement.
Best wishes
David.
The output stage/VAS local loop seems to be the most problematic
to the point that i wonder if it s not better to use a TMC loop at this
location, for the rest as much as 3 inner loops, as in Cherry s paper,
seems to work quite well with a resultant linearity that exceed notably
what can be done with more classic implementations a la blameless.
Hi David,
probably, most arguments are exchanged.
I'm really looking forward to your contributions in Linear Audio, as you appearently will provide mathematical treatment of the OIC (and further ;-) problems.
My experience from other research areas is that this way often can lead to more general insight and can result in guidellines for the 'optimum' solution within a given framework.
So again, good luck and 'efficiency' with your work!
Kind regards,
Matthias
Hi,
totally aggree with you! If you look at post #385 (small correction in #386), there is a method similar to NDFL that is slightly simpler: nested Miller compensation with additional zero in the intermediate Miller compenstaion loop(s) (see also the own thread on an amplifier with nested MC, where I have overdone the whole thing, mainly in order to test it).
Currenly, I'm not completely sure about the following issue. When taking the example in post #385/6, deleting the shunt compensation and applying instead TMC, evaluation of the overall loop gain around the OPS seems to show a problem: the stability margins are OK, but worse than in the case with OIC and shunt compensation (when dimensioning for the same ULGF). So I cannot not yet really decide between these two schemes. In principle, I would prefere TMC, as it uses the HF gain as NFB around VAS alone, instead of "waisting" it by shunt compensation. (Of course, shunt comp. is not a complete waste: it also reduces HF impedance at OPS input.)
Once that decision is made, and the overall circuit is finalized (e.g. protection stuff), I definitely will give the circuit a try in reality; really want to know, how it sounds.
Kind regards,
Matthias
Last edited:
Accepted. In simulation, the zillion MHz are OK, with the given device models. And the trick with capacitive shunt of VAS degeneration resistor works well (the lower the ULGF, the better it seems to work ;-).
Apart from the quality of the models:
Do you have an idea by how much e.g. ft, Cob, Cbe and so on paramaters fluctuate from spicemen to spicemen?
For static parameters like hfe we are used to have some info about the distribution, but for the dynamic ones ...?
Kind regards,
Matthias
Matze, I pontificate on this in #95 and subsequent posts to at least #113. Ignore the pseudo guru noise.
IME, the changes in stability due to wonky loads are more of a problem than device variations in a production amp. That is if you do full 'real life' stability trials as described.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.