I posted a sim of the Cherry "improver" single rush stage ndfl add-on to a MOSFET output amp:
http://www.diyaudio.com/forums/soli...terview-error-correction-124.html#post1324657
next page has another sim of the circuit comparing to Cordell's "HEC"
http://www.diyaudio.com/forums/soli...terview-error-correction-124.html#post1324657
next page has another sim of the circuit comparing to Cordell's "HEC"
Hi,
I sim'd the Cherry ETI 60 NDFL amp last night. Here is the distortion at 1 KHz and just over 40V p-p out.
Harmonic Frequency Fourier Normalized Phase Normalized
Number [Hz] Component Component [degree] Phase [deg]
1 1.000e+03 2.353e+01 1.000e+00 -0.81° 0.00°
2 2.000e+03 2.745e-04 1.167e-05 153.88° 154.69°
3 3.000e+03 4.621e-04 1.964e-05 62.65° 63.46°
4 4.000e+03 3.468e-05 1.474e-06 -11.76° -10.95°
5 5.000e+03 9.583e-05 4.073e-06 -56.24° -55.43°
6 6.000e+03 1.548e-05 6.578e-07 122.78° 123.59°
7 7.000e+03 8.422e-05 3.579e-06 -8.43° -7.62°
8 8.000e+03 1.650e-05 7.011e-07 78.66° 79.47°
9 9.000e+03 8.619e-05 3.663e-06 -17.77° -16.96°
Total Harmonic Distortion: 0.002383%
However I don't have the BD139/140 models on this PC, so I used Onsemi MJE340/350 which will be slower and lower gain I suspect.
The output pair are correct however and biassed around 60mA.
The sim does not fully tie up with measurement, for example the 3rd harmonic exceeds the 2nd in the sim, but otherwise its not a zillion miles off.
The sim is only a guide of course and I am not suggesting for a minute that 'my' amp would achieve 0.000028% THD if measured. I would be happy if it did 0.001%. At this level in the real world the slightest non optimal connection, earth, even mains noise can influence the result, not to mention other effects, induction etc etc and of course the Spice models are not always good..although I would say I trust On Semi and Philips to be representative.
Reading Cherry's ETI article, the layout and comments on the wiring and earthing are almost more interesting than the amplifier, because they are different from what has become the norm. I can only assume this was the config that gave the best results with his hardware. Cherry's star ground is effectively the casework, which is low inductance, but spread about a bit!
Thankyou JCX and FORR for offers of further articles. I have the 'improver' but would like to get hold of the first parts and subsequent parts of the ETI article, so I think those are March, April and June 1983. I now have private messengering turned on BTW...DOH!😕
I sim'd the Cherry ETI 60 NDFL amp last night. Here is the distortion at 1 KHz and just over 40V p-p out.
Harmonic Frequency Fourier Normalized Phase Normalized
Number [Hz] Component Component [degree] Phase [deg]
1 1.000e+03 2.353e+01 1.000e+00 -0.81° 0.00°
2 2.000e+03 2.745e-04 1.167e-05 153.88° 154.69°
3 3.000e+03 4.621e-04 1.964e-05 62.65° 63.46°
4 4.000e+03 3.468e-05 1.474e-06 -11.76° -10.95°
5 5.000e+03 9.583e-05 4.073e-06 -56.24° -55.43°
6 6.000e+03 1.548e-05 6.578e-07 122.78° 123.59°
7 7.000e+03 8.422e-05 3.579e-06 -8.43° -7.62°
8 8.000e+03 1.650e-05 7.011e-07 78.66° 79.47°
9 9.000e+03 8.619e-05 3.663e-06 -17.77° -16.96°
Total Harmonic Distortion: 0.002383%
However I don't have the BD139/140 models on this PC, so I used Onsemi MJE340/350 which will be slower and lower gain I suspect.
The output pair are correct however and biassed around 60mA.
The sim does not fully tie up with measurement, for example the 3rd harmonic exceeds the 2nd in the sim, but otherwise its not a zillion miles off.
The sim is only a guide of course and I am not suggesting for a minute that 'my' amp would achieve 0.000028% THD if measured. I would be happy if it did 0.001%. At this level in the real world the slightest non optimal connection, earth, even mains noise can influence the result, not to mention other effects, induction etc etc and of course the Spice models are not always good..although I would say I trust On Semi and Philips to be representative.
Reading Cherry's ETI article, the layout and comments on the wiring and earthing are almost more interesting than the amplifier, because they are different from what has become the norm. I can only assume this was the config that gave the best results with his hardware. Cherry's star ground is effectively the casework, which is low inductance, but spread about a bit!
Thankyou JCX and FORR for offers of further articles. I have the 'improver' but would like to get hold of the first parts and subsequent parts of the ETI article, so I think those are March, April and June 1983. I now have private messengering turned on BTW...DOH!😕
Attachments
Edward Cherry NDFL article 1983
Here's the article from ETI April 1983.
Zipped JPG per page again.
This is from the UK (British) edition, btw.
The whole article was in one issue.
Cheers - Godfrey
Here's the article from ETI April 1983.
Zipped JPG per page again.
This is from the UK (British) edition, btw.
The whole article was in one issue.
Cheers - Godfrey
Thanks!
It's nice to see reasonable correlation between your sim and Cherry's measurements. Gives one a bit more confidence in the software, although I'll never let it "have the last word".
Cheers - Godfrey
not sure about you guys but I *really* miss the old ETI mag. I had to chuck out lots of mine when moving and kick myself still for that. Wish someone who had a full collection would make them available for a scan project for the web - the amount of collective learning in those pages was (is?) simply priceless.. btw sorry for the OT !
Hi Kasey,
Yes so do I...tried to get a job there once!
And thanks Godfrey, better post you some Ozzy beers...then again....8-((
Sorry that is a Pommie comment, as in beer should be flat and warm....
Yes so do I...tried to get a job there once!
And thanks Godfrey, better post you some Ozzy beers...then again....8-((
Sorry that is a Pommie comment, as in beer should be flat and warm....
Hi Godfrey,
Absolutely, the sim is only a guide. The problem is that I miss the AP Sys 1. You need real hardware, real measurements and
real ears. Problem is it is sooo easy to go forwards and then find out you have gone backwards.As you see I am across the purist/subjective camps (wait for the flames...).😱
The NDFL concept satisfies the engineer in me, predictable, repeatable gain, distortion and transient response and minimal circuit complication. When I make one that SOUNDS as good as an AKSA I will be a happy man...and probably 150years old!😀
Thanks again for all your efforts posting the ETI papers.
Thanks to Godfrey and the first part of the ETI article I now have a better idea of what is going on largely because the magazine article is written in a more digestable 'popular' form. This uncovers an interesting issue with 'VAS inclusive' feedback as applied in the Cherry amp.
If you read the Cherry paper he shows that it is possible to put a differentiating network around a stage which has a 2 pole forward path. The reason being that the single pole increase in loop gain (due to the differentiating feedback) cancels one of the poles in the forward path, so we have a single order cross at 0dB and 90 phase margin. In fact this is the crux of the whole concept, and is represented by the last two 'blocks' in the nest, the VAS transistor and the power stage.
Then you look at the actual circuit, the VAS and output stage form a gain block around which is placed a differentiating network, ie CDOM inclusive feedback. At the output of the amp is a passive RC pole, which only feeds both differentiating and resistive feedback to the front end and is also the Zobel network, neat! This passive pole is at ~200KHz....R=8ohms C=0.1uF
What I think happens here is that if you take the VAS/output stage block and its local differentiating loop in isolation then it is assumed that the gain will fall to unity BEFORE the slope goes to a 3 pole slope. I am guessing in the ETI amp this happens at around 2MHz, as the output stage is quite slow, MJ802 MJ4502. ie ~10X above the passive pole.
At 200KHz the output stage still has a single order slope, so combined with the passive pole, the two pole condition is still satisfied and this signal gets fed back to the front end of the amp, encompassing the pole/zero Rush stage and the LTP with differentiating and resistive feedback respectively.
But what this means is that if you build one of these, and the VAS/output stage is 'not as expected' in terms of its HF performance then the amp will be unstable. I know at least two people who have experienced this....me for one.
So my surmise is that for this technique to work the output stage HF performance must be absolutely defined, and consistant there after, and in these days of unavailable, substituted and faked transistors this might not be reliable. Plus of course you have lead inductances etc etc.
Cherry hints at this because of the use of inductors in the power rails, some series R and a warning to use bog standard electro caps and not bypass with HF types. Also despite the use of a EF type 2 output stage, there is no speed-up cap.
I'm not belittleing Cherry's efforts here, as mathematically the concept is sound and I am sure the ETI amp worked if built to the spec. But builders and tinkerers (ie me) should be wary. It is also a terrible warning over taking someone elses design and putting modern fast transistors everywhere in an attempt to 'improve' it.
What do you guys think...am I right?
If you read the Cherry paper he shows that it is possible to put a differentiating network around a stage which has a 2 pole forward path. The reason being that the single pole increase in loop gain (due to the differentiating feedback) cancels one of the poles in the forward path, so we have a single order cross at 0dB and 90 phase margin. In fact this is the crux of the whole concept, and is represented by the last two 'blocks' in the nest, the VAS transistor and the power stage.
Then you look at the actual circuit, the VAS and output stage form a gain block around which is placed a differentiating network, ie CDOM inclusive feedback. At the output of the amp is a passive RC pole, which only feeds both differentiating and resistive feedback to the front end and is also the Zobel network, neat! This passive pole is at ~200KHz....R=8ohms C=0.1uF
What I think happens here is that if you take the VAS/output stage block and its local differentiating loop in isolation then it is assumed that the gain will fall to unity BEFORE the slope goes to a 3 pole slope. I am guessing in the ETI amp this happens at around 2MHz, as the output stage is quite slow, MJ802 MJ4502. ie ~10X above the passive pole.
At 200KHz the output stage still has a single order slope, so combined with the passive pole, the two pole condition is still satisfied and this signal gets fed back to the front end of the amp, encompassing the pole/zero Rush stage and the LTP with differentiating and resistive feedback respectively.
But what this means is that if you build one of these, and the VAS/output stage is 'not as expected' in terms of its HF performance then the amp will be unstable. I know at least two people who have experienced this....me for one.
So my surmise is that for this technique to work the output stage HF performance must be absolutely defined, and consistant there after, and in these days of unavailable, substituted and faked transistors this might not be reliable. Plus of course you have lead inductances etc etc.
Cherry hints at this because of the use of inductors in the power rails, some series R and a warning to use bog standard electro caps and not bypass with HF types. Also despite the use of a EF type 2 output stage, there is no speed-up cap.
I'm not belittleing Cherry's efforts here, as mathematically the concept is sound and I am sure the ETI amp worked if built to the spec. But builders and tinkerers (ie me) should be wary. It is also a terrible warning over taking someone elses design and putting modern fast transistors everywhere in an attempt to 'improve' it.
What do you guys think...am I right?
This is correct. Think what you want to do, you want to damp out noise, ripple and oscillations on the supply line. Damping can only work if you take away energy. If you use a very low ESR, very high quality HF film cap, there is no loss. What you do is move the noise etc to some other part of the circuit like ground lines. To clean supply lines you need lossy components and bog-standard elco's are ideal.
Exactly. The NDFL works by manipulating loop gain and phaseshifts in nested loops. Now if you start to modify these by substituting slower or faster transistors, chances are it does no longer work as intended. Only do this if you are absolutely sure what you are doing, otherwise you might end up with either a smoking amp or a nice AM-transmitter 😉
Edit: I made the individual pages in a single .pdf (7,5Mb), should be easier to read: http://www.linearaudio.nl/Miscellaneous/cherry ndfl.pdf
jd
Last edited:
Thanks Jan.
Perhaps you could do the same for the amplifier design?
It's posted here: http://www.diyaudio.com/forums/solid-state/148066-new-cherry-ndfl-amp-3.html#post2104685
Cheers - Godfrey
Hi Janneman,
Thanks thats a handier format! But hey jpgs are OK too!
Agreed, messing around is not good, my amp was not stable with the differentiation loop around the whole stage, only around the VAS as per normal and that seems to be most peoples experience. In the case of the Cherry amp it 'undoes' the concept somewhat.
What I was really thinking was of the implications for VAS inclusive feedback generally.
I think we can agree that his amp is stable if built correctly.
More simulations required I think.
Thanks thats a handier format! But hey jpgs are OK too!
Agreed, messing around is not good, my amp was not stable with the differentiation loop around the whole stage, only around the VAS as per normal and that seems to be most peoples experience. In the case of the Cherry amp it 'undoes' the concept somewhat.
What I was really thinking was of the implications for VAS inclusive feedback generally.
I think we can agree that his amp is stable if built correctly.
More simulations required I think.
Hmmm...reading the ETI article...the points where all responses converge and assume a first order slope Tx, is set to 800nS or 1.25MHz. That is a design objective.
The values for RY, CY around the Rush stage are set to Tx....so the values should be 1800ohm and ~68p. The choice of 1800 and 470p actually sets the point at 200KHz or there abouts, reducing the available gain for distortion correction quite considerably.
Needless to say what happens at the extreme HF is not effected, so the amp will work, but possibly not as well as intended.
Is this a genuine error or am I mising something??
The values for RY, CY around the Rush stage are set to Tx....so the values should be 1800ohm and ~68p. The choice of 1800 and 470p actually sets the point at 200KHz or there abouts, reducing the available gain for distortion correction quite considerably.
Needless to say what happens at the extreme HF is not effected, so the amp will work, but possibly not as well as intended.
Is this a genuine error or am I mising something??
this site don't always goes open - this often happens in generall with websites of the Dutch web service provider "tiscali"
Unfortunately, these pages are not archived about www.archive.org, so that - when the page is dead, even nothing more to be found there.
Because I had archived the weblink earlier at my own hard disc, I was able to generate a PDF (see attachement).
Last edited:
Hi,
Thanks for the ETI articles. I've read the AES articles before but did not understand why the feedback that was returned to the rush stage was filtered by the zobel network.
Now I think I understand what intention is but I have to read and think a bit more 🙂
A question to Mr Stuart:
You seems to thing that the idea of putting a pole in the demand gain is wrong ( or I misunderstand ). Could you explain a bit more, please?
/örjan
Thanks for the ETI articles. I've read the AES articles before but did not understand why the feedback that was returned to the rush stage was filtered by the zobel network.
Now I think I understand what intention is but I have to read and think a bit more 🙂
A question to Mr Stuart:
You seems to thing that the idea of putting a pole in the demand gain is wrong ( or I misunderstand ). Could you explain a bit more, please?
/örjan