Based on that, I've changed TR to 260n on AndyC's 2sa1837/c4793_A and also Harry Dymond's 2sc1930/5171_d models in my copy of Toni's mylibs.lib
Toni, the 'pure Cherry' version NEEDS R22 at 27R but compensates with even lower THD than the TMC/TPC versions. I'll post some THD sims in a bit.
Thanks & grovel grovel to Guru Zan ...
... you mean you have a new pure cherry version of the "2stageEF" ready for real life testing ...
Toni, I try to keep 2 'pure Cherry' versions current. One with minimal changes so it is easy for you to try. Another with more changes in line with my prejudices.... you mean you have a new pure cherry version of the "2stageEF" ready for real life testing ...
I didn't post anything cos I always check overload & stability when I make changes so wanted the driver models good first.
#457 Cherry version
Minimal changes to convert this to 'pure Cherry'. C1/10 470 replace the TMC network. R22=20+6.8 This VAS emitter resistor decoupled by C7 220p + R27 10
I've only done a quick stability & overload check but everything looks OK so far.
THD20k for the Cherry version is less than 3dB better than the TMC.
As these THD levels are some 10dB below what Toni measured in #471 even with the Distortion Magnifier, I don't expect measurable differences between the two.
Probably the only difference you will see is that the Cherry THD residual is less spiky - higher order products smaller
Thanks & grovel grovel to Guru Zan .. David, can I get up now? My knees hurt
Minimal changes to convert this to 'pure Cherry'. C1/10 470 replace the TMC network. R22=20+6.8 This VAS emitter resistor decoupled by C7 220p + R27 10
I've only done a quick stability & overload check but everything looks OK so far.
THD20k for the Cherry version is less than 3dB better than the TMC.
As these THD levels are some 10dB below what Toni measured in #471 even with the Distortion Magnifier, I don't expect measurable differences between the two.
Probably the only difference you will see is that the Cherry THD residual is less spiky - higher order products smaller
Thanks & grovel grovel to Guru Zan .. David, can I get up now? My knees hurt
Attachments
Looks nice. Should have commented before that the PSRR trick with C10 looks neat.
I don't recall I have seen it elsewhere, is it your own idea?
I would be tempted to keep R22 a bit lower and stop the rail stick with a Baker clamp diode or some such.
Never studied clip behaviour much, for the reason I already mentioned.
I can do a bit more work on this, can you post the ASC?
Best wishes
David
Last edited:
I used it circa 1990. Works well. No idea if its novel. Pretty obvious with hindsight.
Its in Scott Wurcer's AD797 datasheet in slightly different (?!) form.
The problem with clamp diodes (ala JE990) is that the evil modulated C of the diode introduces distortion at these sub 10ppm levels.
Yes. A clamp diode 'cures' the sticking at the expense of THD. But 'pure Cherry' needs the slightly bigger R22 anyway.
I'm playing with a 'Baxandall cascode' for the VAS which shows promise without needing extra bits.
I'm including an updated version of Toni's mylibs.lib with the enhanced driver models with Guru Zan's TR. It's called mylibs.txt
Last edited:
Most ideas are obvious, after someone has them, especially if it's someone else.
Toni confirmed that lower R22 reduced THD at the expense of worse rail stick, so the idea was to keep the low THD and fix the rail stick properly.
Not much use if the cure also introduces THD, of course.
Bob writes of ways to float the diodes to minimize the effect but I don't need them so I never paid much attention.
Some care with stability needed there I suspect.
You read Edmond's notes on stability of Baxandall in his SuperTIS thread?
Thanks. I will look tomorrow. Jeez, already a week since my last Friday rush. "Havin' fun is full time job".
Best wishes
David
Minimal changes to convert this to 'pure Cherry'. C1/10 470 replace the TMC network. R22=20+6.8 This VAS emitter resistor decoupled by C7 220p + R27 10
I've only done a quick stability & overload check but everything looks OK so far.
THD20k for the Cherry version is less than 3dB better than the TMC.
As these THD levels are some 10dB below what Toni measured in #471 even with the Distortion Magnifier, I don't expect measurable differences between the two.
Probably the only difference you will see is that the Cherry THD residual is less spiky - higher order products smaller
Thanks & grovel grovel to Guru Zan .. David, can I get up now? My knees hurt
Could you state the phase margins of your cherry version and the TMC version. I sometimes see amps with tiny phase margins designed so it can have better THD figures. The picture changes completly when same phase margins are applied. Neglecting stability for the sake of outright THD numbers is not a intelligent proposition.
Stability
I won't post the 'linear' stability results for my latest 'pure Cherry' version as they are similar (but not identical) to those in #182. You can get your PM & GM from those if you wish. They exceed Bob Cordell's & Bode's recommendations even on the worst possible loads. I don't do single 'linear' (or other) stabiltiy trials but several; on different loads to try & provoke instability.
Instead, I'll show some for Toni's latest TMC version as I've not posted these before. The Closed Loop response is for the worst possible load .. which in this case is 1nF pure capacitance.
You'll recall Toni & I have discussed several times in this thread how to do 'real life' testing of stability. This is probably more important than simple 'linear' measures like PM & GM.
Some important (real life & sim) 'parasitic' issues have been addressed since those early posts. They do not show up in simple 'linear' stabiltiy measures like PM & GM.
I discuss stability in tpc-vs-tmc-vs-pure-cherry and also #2306 of discrete-opamp-open-design
Manso, some of my stability tests are described in posts #177-180, 182-183, 196 of this thread
I won't post the 'linear' stability results for my latest 'pure Cherry' version as they are similar (but not identical) to those in #182. You can get your PM & GM from those if you wish. They exceed Bob Cordell's & Bode's recommendations even on the worst possible loads. I don't do single 'linear' (or other) stabiltiy trials but several; on different loads to try & provoke instability.
Instead, I'll show some for Toni's latest TMC version as I've not posted these before. The Closed Loop response is for the worst possible load .. which in this case is 1nF pure capacitance.
You'll recall Toni & I have discussed several times in this thread how to do 'real life' testing of stability. This is probably more important than simple 'linear' measures like PM & GM.
Some important (real life & sim) 'parasitic' issues have been addressed since those early posts. They do not show up in simple 'linear' stabiltiy measures like PM & GM.
I discuss stability in tpc-vs-tmc-vs-pure-cherry and also #2306 of discrete-opamp-open-design
Attachments
Miller: simulated phase margin ~68 degree and gain margin ~21 dB
rock stable in real life.
TPC: simulated phase margin ~53 degree and gain margin ~20 dB
rock stable in real life.
TMC: simulated phase margin ~55 degree and gain margin ~14 dB
rock stable in real life.
Cherry: simulated phase margin ~30 degree and gain margin 16dB
not tested in real life - maybe unstable due to low phase margin.
BR, Toni
Real life results of cherry variant
Real life measurements of cherry compensation variant:
cherry compensation variant needs some rework to get the amplifier stable...
Real life measurements of cherry compensation variant:
THD21K@200W@8R
0.00188% (bw 80 kHz)
0.00226 (no bw limit)
Looking at square wave osci pictures we can see oscillation / instability signs. Also the overload situation using sinus is showing some oscillation.0.00188% (bw 80 kHz)
0.00226 (no bw limit)
cherry compensation variant needs some rework to get the amplifier stable...
Attachments
It looks like it oscillates when the drivers are recovering from the current surge.
I would work with the VAS degeneration decoupling first, and then change the miller cap if that doesn't work. Maybe a 20R trimmer?
It looks like the compensation is no longer the dominant distortion mechanism. You should double check your feedback and signal paths and the resistors used. Precision resistors may still distort at HF:
Bruno Putzeys r4 random rants, raves and ramblings
I would work with the VAS degeneration decoupling first, and then change the miller cap if that doesn't work. Maybe a 20R trimmer?
It looks like the compensation is no longer the dominant distortion mechanism. You should double check your feedback and signal paths and the resistors used. Precision resistors may still distort at HF:
Bruno Putzeys r4 random rants, raves and ramblings
Same hardware backported to TMC compensation is rock stable and has a THD figure of
THD21K@200W@8R
0.00145% (bw 80 kHz)
0.00185 (no bw limit)
So cherry compensation needs some rework to get more phase margin.
BR, Toni
No wonder the Cherry version has less THD in sims, Id love to see the numbers when its phase margin is more in line with the other two compensation methods.Maybe Im paranoid but I think even a margin of 55 is low, I design for at least 75 degrees. I think a fair comparison for linearity should only be made with comparable margins, like is the case with TMC and the TPC.
Im still doubtfull of a "cherry" amp success, the outputstage with its large capacitances will most likely make it impossible but lets see.....
Evil Cherry.
Last edited:
With compensation up to the limit, 100n caps at the rails will cause an amp to oscillate because their resonance changes global phase. 1u+1R snubbers across the rails help with this. These snubbers should go across the OPS rails, because the outputs are the main leakage point for rail resonances.
What is the frequency of the oscillation? This will help us understand the cause.
When you increase or decrease the driver bias, does the oscillation last for a longer or shorter time? This will tell us whether driver recovery is a factor in oscillation. If not, it could be TIS recovery.
What is the frequency of the oscillation? This will help us understand the cause.
When you increase or decrease the driver bias, does the oscillation last for a longer or shorter time? This will tell us whether driver recovery is a factor in oscillation. If not, it could be TIS recovery.
Thanks for this Toni.
Is the THD20k residual with 80kHz BW? The slightly higher THD compared to TMC is likely to be connected to the slight instability.
As there is no sign of xover on your TMC version (even with the Distortion Magnifier) ... and my prediction (using evil sims) is that the Cherry version will only show less xover, there appears to be no reason to risk the dodgy stability of a circuit suggested by some beach bum in Oz.
Prof. Cherry was another scallywag from Oz too .. you can't trust dem!
______________
manso said:
sob sobManso, as I said, the only reason I'm playing with dis SPICE sh*t is that I hope to recreate some 'real life' effects I noticed in Jurassic Times in several experimental and at least 1 production amp. But I have never tried 'real life' Cherry on such a big amp.
The 'real life' stuff that Toni shows are almost exactly what I'm seeing when playing with a 'Baxandall pair' VAS (SPICE) version that attempts better overload recovery than Toni's TMC or the simple Cherry mod.
LTspice isn't telling the whole truth but is so...oo close.
I now need to force the SPICE model of the simple Cherry modified circuit to show these faults if this research is to go forward.
But as I said to Toni, from my sims, it is unlikely that a 'good' Cherry circuit will show any practical improvement on his latest TMC. I'm just hoping his curiosity will prompt him to experiment. I'm already in his debt for taking the time & trouble to humour me so far. Dunno about yus guys but I've learnt a lot from this exercise and expect to learn more.
I (and 'pure Cherry') am now on trial. Using a model that we already know is wonky (cos it doesn't reflect 'real life'), I have to dream up some simple mod that will cure 'real life' marginal stability in the simple Cherry mod to Toni's circuit. A big challenge for a beach bum
______________________
One of the things I found in my Jurassic 'real life' experience with sims (my own 'linear' circuit analysis programme) & amps is that simulating 'real life' Power Supply conditions had major effect on THD & stability.keantoken said:
I've yet to do this with any of my LTspice models.
But Toni's circuit is actually quite good in stuff to damp the very effects you mention. He has large electrolytics C10/14/44/45 + more caps decoupling the 'small' signal rails and his 10R series resistors R48/50/81/82 provide a lot of damping
One of the big factors that isn't seen in 'linear' measures like PM GM etc is the effect of the various 'speedup' caps. I mentioned earlier that 'pure Cherry' usually likes much smaller or no speedup caps across the bias spreader & driver emitter resistors.
I'm still playing but this seems to be a major factor ... at least in the 'Baxandall pair' version. It affects how much 'control' the VAS has on both +/- outputs under conditions of very high slew.
Adding 10n to the input to restrict BW to 100kHz eliminates these nasties but I still want to see if I can get rid of them by other means. The BW limit is vital in ANY 'real life' amp as it to avoid TID, PIM, other TLA induced evils.
Yes to all that. I was using the 10n just to limit BW to 100kHz for the sim.
In real life, the worse case condition I would allow for in a simple amp would be a 10k pot so 2k5 source R at -6dB setting.
I'm partial to even higher order i/p BW limiting like Edmond shows in his 'Super TIS' page. My ideal is a BBC/IBA 1980's spec which requires at least 3rd order Bessel to achieve. But this is all extra bits so you might as well have the balanced i/p stage too.
Last edited:
Back towards the start of this thread I noted that "Cherry" is actually a simplified, limit case of TMC so TMC could do at least as well and potentially better.
It is nice to have that confirmed but what is nicer is that it has finally occurred to me how to analyse the problem.
So how do we know that the optimisation of TMC will not result in Cherry?
In other words, that results will not continue to improve as the TMC resistor is reduced indefinitely and similarly the VAS output capacitor?
When an amp with specific IPS, VAS and OPS has its compensation altered then any reduction in distortion is the result of increased feedback.
So the problem becomes how to maximise the feedback within a certain stability limit.
This problem has been solved! By Bode of course.
The answer is NOT a simple function as produced by the simple one capacitor Cherry compensation.
It is actually not even a rational function, so can only be approximated in practice but it does look a lot more like the TMC return ratio.
OK, I have skipped over non-idealities like shunt compensation losses, and the analysis is only a "wave your hands" description rather than a proper proof but I really think it is the core of the issue.
Of course, "hand wavy" stuff from other people often sounds like rubbish.
Does this resonate with anyone?
LOL!Evil Cherry.
Best wishes
David
Last edited: