"Son, your sins are forgiven." Luke 5:20 😉
The discussion on TMC started about here: http://www.diyaudio.com/forums/soli...terview-negative-feedback-50.html#post1160022
Thanks ed...A lengthy thread, btw...
Well, my own acronym for such a compensation scheme is DMC...
Distributed Miller Compensation.
It work in sims, but i never implemented it, as i still lack
some necessary tools to check the thing in real world,
so , for the time, i have to stay on the catious side of
compensation solutions, i.e, the ultra common usual
miller cdom comp.
I guessed that you were referring to Fig 9b, but I haven't seen Cherry use that in a published design, so I question whether that is actually Cherry's preference. The designs I have seen are:
- A High Quality Audio Power Amplifier (Jan/Feb 1978)
- NDFL in Simple Audio Power Amplifiers (May 1982)
- 60W NDFL Amplifier (May 1983)
- also see Fig 2 in Electronics World Jan 1995 (where your snippet came from)
All use the network based on Fig 9a but with the feedback taken from the cap/resistor junction (Fig 10 from the same article as your snippet). Note that Cherry says that Fig 10 also has advantages over Fig 9.
Yes, but also the High Quality Amplifier article (see above).
If anyone can post the 1995 EW article that would be much appreciated, havn't read that one! Thanks, and thanks for all the contributions so far.
semantics?
Yes.
That's precisely my point, why he did it that way? (see below)
That's why. He opted for Fig 9a because he wanted to use that C/R junction as take-off point for feedback. However, this doesn't necessarily imply that from a Zobel point of view, he prefers Fig 9a over Fig 9b.
Yes, I've noticed that too. But I disagree, totally:
'the network does not introduce a 3dB loss at the cut-off frequency Wx'
True, it does roll-off at a slightly higher frequency: 250kHz vs 200kHz. Wow, what a big deal.
'radio-frequency interference picked up on the loudspeaker leads is isolated from the feedback point by a two-pole filter'
Nonsense. The 12pF cap across the FB resistor 'undoes' one of these poles. As a result, HF interference from output to inverting input is only suppressed by 6dB/octave (i.e. 1 pole).
Yes.
That's precisely my point, why he did it that way? (see below)
That's why. He opted for Fig 9a because he wanted to use that C/R junction as take-off point for feedback. However, this doesn't necessarily imply that from a Zobel point of view, he prefers Fig 9a over Fig 9b.
Yes, I've noticed that too. But I disagree, totally:
'the network does not introduce a 3dB loss at the cut-off frequency Wx'
True, it does roll-off at a slightly higher frequency: 250kHz vs 200kHz. Wow, what a big deal.
'radio-frequency interference picked up on the loudspeaker leads is isolated from the feedback point by a two-pole filter'
Nonsense. The 12pF cap across the FB resistor 'undoes' one of these poles. As a result, HF interference from output to inverting input is only suppressed by 6dB/octave (i.e. 1 pole).
Yes, I thought that was a bit of a wild claim...by allowing the adjustability in the 'removal' of this pole...or otherwise he allows the fine tuning of the transient response, OK fine, but I think the RF claims are a bit exaggerated!
I assume BTW the origin of the 'sq root 3-1' is a Bessel function???
Must have been talking to the marketing department...
In fact the whole issue of what goes on in the last stages are pretty glossed over. If you build this thing with perfect Spice models then the paper holds true, the theory itself appears correct. But the adoption of a real world output stage complicates things seriously.
In theory the technique is valid, in the real world the amp is stable, but I begin to suspect the benefits are not as great as one might suppose, over a correctly designed conventional 3 stage topology, and this might explain why this technique was not widely accepted.
Through the wonders of diyAudio I have been contacted by someone who actually has access to one of Cherry's original prototype. I don't think total dissection will be possible, but I am hoping to have a listen to it! Watch this space.
Hi audiopip, you should receive the EW 1995 article tommorow (by snail mail).
Have you got a copy of the IREE 1978 15 watt article yet? If so I'll stop looking for mine!! Ha ha.
Jonathan
Have you got a copy of the IREE 1978 15 watt article yet? If so I'll stop looking for mine!! Ha ha.
Jonathan
Re; output inductors etc. Just ever so slightly tangential to the main course of NDFL, if any of you do have RFI break through probs due to speaker leads, the original article that Ed Cherry referes to by Neville Thiele (Of bass reflex fame) elaborates on the simple version seen so far in this thread and even has a third order creation with the option of bifilar wound inductors giving interesting feedback take off points etc. Its in the JAES. (Can't remember which Vol. page no. details etc of the top of my head.)
Audiopip
---If anyone can post the 1995 EW article that would be much appreciated, havn't read that one! Thanks, and thanks for all the contributions so far.---
Cherry's EW&WW Jan 95 and EW July '97 articles are available from me by email.
---If anyone can post the 1995 EW article that would be much appreciated, havn't read that one! Thanks, and thanks for all the contributions so far.---
Cherry's EW&WW Jan 95 and EW July '97 articles are available from me by email.
Bessel function???
As for the 'sq root 3-1' factor, in the EW+WW article Cherry wrote:
'If the amplifier without feedback has just one dominant pole, and if the overall loop gain without network falls through unity at Wx, then the overall response is made phase-linear by chosen Rf2.Cf2 = (sqrt(3)-1)/Wx = 0.7/Wx' (Wx stands for omega sub x)
I have checked this by means of simulation of a 'normal' amp (without NDFL) and indeed, his statement is correct, that is, the step response is critical damped and shows no overshoot. Next, I simmed his NDFL amp, but then I got overshoot. Probably I'm overlooking something. The effect of slow trannies?
Actually, I don't care much for it, as this NDFL design is so outdated and based on obsolete trannies. Moreover, as Pete pointed out, this design suffers from bad recovery after clipping.
I'm not saying that NDFL is a waste of time, but a lot of work has to be done to get a really good amplifier, i.e. one that meets today standards.
PS: Regarding my remark about the cut-off frequency of the Zobel network, my figures were not entirely correct, as the ULGF was chosen a bit too high. Not that this really matters. The point is that the frequency response of his magic feedback loop is almost identical to the response of a traditional circuit.
As for the 'sq root 3-1' factor, in the EW+WW article Cherry wrote:
'If the amplifier without feedback has just one dominant pole, and if the overall loop gain without network falls through unity at Wx, then the overall response is made phase-linear by chosen Rf2.Cf2 = (sqrt(3)-1)/Wx = 0.7/Wx' (Wx stands for omega sub x)
I have checked this by means of simulation of a 'normal' amp (without NDFL) and indeed, his statement is correct, that is, the step response is critical damped and shows no overshoot. Next, I simmed his NDFL amp, but then I got overshoot. Probably I'm overlooking something. The effect of slow trannies?
Actually, I don't care much for it, as this NDFL design is so outdated and based on obsolete trannies. Moreover, as Pete pointed out, this design suffers from bad recovery after clipping.
I'm not saying that NDFL is a waste of time, but a lot of work has to be done to get a really good amplifier, i.e. one that meets today standards.
PS: Regarding my remark about the cut-off frequency of the Zobel network, my figures were not entirely correct, as the ULGF was chosen a bit too high. Not that this really matters. The point is that the frequency response of his magic feedback loop is almost identical to the response of a traditional circuit.
I noticed a bit of overshoot and ringing on the "5kHz square wave into 8R" scope trace in the April 1983 article, so I guess your sim matches his amp.
The theory's interesting, but his implementation wasn't great (by today's standards).
I saw that too, still I'm wondering whether that magic constant (0.732) is correct.
The theory's interesting, but his implementation wasn't great (by today's standards).
Not really surprising as the design dates from an age when cutting back on the number of transistors was common practice.
BTW, my latest NDFL design performs a little bit better: THD20=0.2ppm, though the voltage clamp (the toughest part) is still under development.
QUOTE]
With perfect spice models, any topology would yield perfect results,
so Cherry s implementation would be pointless.
Remember that it is supposed to give better results in the real
world, which is still not demonstrated.[/QUOTE]
Sorry misunderstanding, I was referring to the position of the poles and zeros and the slopes obtained. In particular the position of the first pole which appears in the loopgain response. This is typically placed at 20KHz to allow a large value of loop gain to be employed across the 20-20KHz band.
However, the adoption of the real world circuitry tends to push this back towards 1KHz and then IMO (as Edmond states) the level of feedback available is little better than a well designed amp using modern transistors.
With perfect spice models, any topology would yield perfect results,
so Cherry s implementation would be pointless.
Remember that it is supposed to give better results in the real
world, which is still not demonstrated.[/QUOTE]
Sorry misunderstanding, I was referring to the position of the poles and zeros and the slopes obtained. In particular the position of the first pole which appears in the loopgain response. This is typically placed at 20KHz to allow a large value of loop gain to be employed across the 20-20KHz band.
However, the adoption of the real world circuitry tends to push this back towards 1KHz and then IMO (as Edmond states) the level of feedback available is little better than a well designed amp using modern transistors.
Sorry misunderstanding, I was referring to the position of the poles and zeros and the slopes obtained. In particular the position of the first pole which appears in the loopgain response. This is typically placed at 20KHz to allow a large value of loop gain to be employed across the 20-20KHz band.
However, the adoption of the real world circuitry tends to push this back towards 1KHz and then IMO (as Edmond states) the level of feedback available is little better than a well designed amp using modern transistors.[/QUOTE]
As implemented in the schematic simmed, his amp has exactly the same
OLG , whether it use the NDFL or the classic miller compensation...
There s only one significant scheme in his amp, that s the output
stage included in the miller loop.
This loop is the dominant one that act as deep as cancelling the
other compensation loops..
http://www.diyaudio.com/forums/solid-state/148066-new-cherry-ndfl-amp-8.html#post2110484
Hi Jonathan,
Thank you very much I will look forward to it. I do not have the IREE paper as yet, but please don't go to too much trouble on my account!
Whilst the history, theory and implementation of the Cherry amp is of great interest I suspect there are other ways to do this today which yield better results.
At the time these papers were written they were a new slant on the subject.
My interest is that
1) Cherry is an Australian and lives reasonably locally to my adopted home.
2) He is respected in his field (audio was a sideline for him)
3) No one ever took up the design commercially
4) He was vilified by certain members of the audio community (which he probably didn't deserve)
5) He was kind enough to give me permission to 'play' with his ideas as a hobbist.
6) By forcing myself to understand his papers I am self teaching myself stuff which I would have been better placed to deal with if I had gone to Uni instead of messing about with bands and PA.....its an old bloke 'if I had my time again thing'!😀
I believe Cherry was drawn into the audio 'debate' after a meeting with Otala, when he was asked how more loop gain could be employed without compromising stability. NFDL was his response.
All in all that makes him a very interesting character!!
BTW I discovered the Cherry papers whilst researching the Quad current dumper design and the subsequent debacle as to how that was supposed to work. Although in some ways suffering a similar experience to Cherry, Peter Walker and Quad at least made some money out of the venture!
Thanks for your assistance.
Last edited:
Hi Arthur,
Yes I know Cambridge Audio, and Audiolab for that matter. I guess we should take Doug Self out of the equation. He has written some interesting stuff, fair play to him.
What I would say as a generalisation and IMO, is that amplifiers with a degenerated LTP input, enhanced VAS and a CFP output stage measure well but do not seem to do too well in the sound quality stakes.
The NDFL amplifier to my design falls into that catagory.
Yes I know Cambridge Audio, and Audiolab for that matter. I guess we should take Doug Self out of the equation. He has written some interesting stuff, fair play to him.
What I would say as a generalisation and IMO, is that amplifiers with a degenerated LTP input, enhanced VAS and a CFP output stage measure well but do not seem to do too well in the sound quality stakes.
The NDFL amplifier to my design falls into that catagory.