I tried in reality, with and without Zobel, and also at medium and high (but not clipping) levels:
First, without Zobel and high level:
Then medium level:
Zobel + high level:
Medium level:
There is always some imperfection present: low-level oscillation, edge chamfered.
Values could be refined to iron out the last wrinkles, but unfortunately simulation is of little help: it would have to be done in reality.
Without the Zobel, the waveforms are almost good: it is just the HF instability that ruins it. Maybe time to reintroduce the ferrite bead
This latest sim fixes all issues related to current spikes in Q1/Q2 and Q5/Q9. R18 added.
Good sim results, including square waves. Q11 changed to a higher Vce one.
Good sim results, including square waves. Q11 changed to a higher Vce one.
I have tested the relatively minor mods on the OP stage: that's the only part I test, and R18 is out of it. However, before closing the loop, it is essential to have the OP well-behaved (unless you apply some kind of active compensation, but that's not the case here).
The minor mods did change the behaviour though: I am not going to post additional pics for that, because there are different behaviours according to the level.
To summarize: up to 10Vpp, the square is nice, apart from a small positive overshoot, not really a problem.
After that, the positive edge becomes very messy, jagged, then the "chamfer" also appears, and finally just before clipping, instabilities appear on the positive plateau.
Nothing that a ferrite bead could really cure, thus I didn't make the test.
The minor mods did change the behaviour though: I am not going to post additional pics for that, because there are different behaviours according to the level.
To summarize: up to 10Vpp, the square is nice, apart from a small positive overshoot, not really a problem.
After that, the positive edge becomes very messy, jagged, then the "chamfer" also appears, and finally just before clipping, instabilities appear on the positive plateau.
Nothing that a ferrite bead could really cure, thus I didn't make the test.
This is the best version, including the latest bias adjustments:
The square response is absolutely perfect; however, when negative clipping begins, a "chamfer" also appears, at the end of the trailing edge, but this not really a problem.
With a triangle, the clipping looks the same as before: not spotlessly clean, but acceptable.
My tests are made at 16V; with 42V, the values of R13 and R12 should be adapted.
I have left D10 in place: it is inactive, but it does no harm.
With R10 increased, C3 can be made smaller
The square response is absolutely perfect; however, when negative clipping begins, a "chamfer" also appears, at the end of the trailing edge, but this not really a problem.
With a triangle, the clipping looks the same as before: not spotlessly clean, but acceptable.
My tests are made at 16V; with 42V, the values of R13 and R12 should be adapted.
I have left D10 in place: it is inactive, but it does no harm.
With R10 increased, C3 can be made smaller
Thanks Elvee! It looks good. Do you think it's ready (as is) for a build?
Anything you would change in the pre-amp stages?
Anything you would change in the pre-amp stages?
I think it is ready, but C4 makes me uneasy: it links Q7 and Q3 which are each referenced to opposite rails.
Any noise on the supplies will be injected into the signal path.
If you cannot dispense with it, try to find an equivalent effect/configuration that does not degrade the PSRR
Any noise on the supplies will be injected into the signal path.
If you cannot dispense with it, try to find an equivalent effect/configuration that does not degrade the PSRR
Looks like for the Tandem amp this C4 doesn't make any difference, so I'll remove it.
It was improving square waves sim for a different amp that I'm currently building...
Have to go back to the drawing board and get rid of it too, and find another way...
It was improving square waves sim for a different amp that I'm currently building...
Have to go back to the drawing board and get rid of it too, and find another way...
Earlier, I meant R15, not R12. Anyway, R13 and R12 should be selected for the actual semi types, supply voltage etc. on the first of the series.
Once it is done, it can be reproduced on all the subsequent copies, provided they use similar devices.
The selection should be done based on the THD at an intermediate level, with the GNFB disabled, of course, otherwise it would be near-impossible to detect.
A good method is also to look at the output residue, when it is driven from an auxiliary amplifier (also no GNFB and input grounded), from the nominal output impedance also at mid-level: you should aim for the lowest/least distorted residue (a tradeoff is necessary, because nullification of Zo doesn't occur at the same time as the THD minimum)
Once it is done, it can be reproduced on all the subsequent copies, provided they use similar devices.
The selection should be done based on the THD at an intermediate level, with the GNFB disabled, of course, otherwise it would be near-impossible to detect.
A good method is also to look at the output residue, when it is driven from an auxiliary amplifier (also no GNFB and input grounded), from the nominal output impedance also at mid-level: you should aim for the lowest/least distorted residue (a tradeoff is necessary, because nullification of Zo doesn't occur at the same time as the THD minimum)
I was testing today 1st build of another amp with the same input/vas as the one published here to be used with Tandem amp.
Unfortunately this input stage does not work as expected.
So I'll either make it work (still experimenting with it), or use different input stage from one of my previous amps,
that is proven to work fine (E.g this one).
Unfortunately this input stage does not work as expected.
So I'll either make it work (still experimenting with it), or use different input stage from one of my previous amps,
that is proven to work fine (E.g this one).
It kind works now, but there is one problem - something with the way ground is referenced with the op-amp.
When watching any wave on the oscilloscope, the whole wave is slightly moving up and down, by around 100mV.
This effect disappears if I connect dirty ground with signal ground.
Perhaps R6/C16 should be going to signal ground? On my PCB they go to dirty ground.
Everything except input network and C20 goes to dirty ground.
All waves look great except for these small movements/waving up-and-down....
So I'll leave testing this module for long winter evenings when I have nothing better to do,
and
in a meantime, I also tested another input/vas module, which works like a champ, so I'm going to adapt it to the Tandem amp, and go ahead with a PCB.
Last edited:
Yes, I tried (2 zeners referencing ground, just like caps). didn't help.
Also tried shorting R9/R11 and changing compensation slightly so it works with direct connection between op-amp rails and emitters.
It works, just like original schematic, but the movement of waves on the oscilloscope is still there.
So here is another front end proposal (already tested).
I'm planning to build Tandem amp just like this OS stage platform (on the 2nd picture) that can take doughterboards with vas/input.
I already have several different input/vas modules already built, so I just need to build the Tandem OS part with heatsink, connectors, etc...
OS platform with HexFets outputs:
I'm planning to build Tandem amp just like this OS stage platform (on the 2nd picture) that can take doughterboards with vas/input.
I already have several different input/vas modules already built, so I just need to build the Tandem OS part with heatsink, connectors, etc...
OS platform with HexFets outputs:
The slow movement can either be a low-level, low frequency oscillation, like a problematic DC servo, or a beat effect between the test frequency, mains frequency or sweep rate of the oscilloscope (even if it's digital)
I see. The thing is, that I'm testing it with within the same environment as the other 10 amps recently built.
The same OS platform, same PSU, same oscilloscope, same generator.
If I plug in (and I did try) another input/vas module, it shows rock solid waves on the screen.
The only difference is the that the input/vas module being tested is different. Hence my conclusion that this specific module
is more sensitive to all these things you mentioned - which basically means a design flaw...
If I had to pick one from your list - I would go for "low-level low frequency oscillation".
It's OK, I have another op-amp rail-current driven input like this (see here), but it's more complex (and it works perfectly);
I tried to simplify things in this one, and failed.
Here is the 'blameless' input/vas module I plan to couple with Tandem OS - completed both channels playing music as we speak (with hexfet OS) 🙂
Judging from the sims, Tandem OS seems to be more stable and less likely to oscillate, compared to multi-pair fet OS, so I hope it should be easy to couple these two.
The same OS platform, same PSU, same oscilloscope, same generator.
If I plug in (and I did try) another input/vas module, it shows rock solid waves on the screen.
The only difference is the that the input/vas module being tested is different. Hence my conclusion that this specific module
is more sensitive to all these things you mentioned - which basically means a design flaw...
If I had to pick one from your list - I would go for "low-level low frequency oscillation".
It's OK, I have another op-amp rail-current driven input like this (see here), but it's more complex (and it works perfectly);
I tried to simplify things in this one, and failed.
Here is the 'blameless' input/vas module I plan to couple with Tandem OS - completed both channels playing music as we speak (with hexfet OS) 🙂
Judging from the sims, Tandem OS seems to be more stable and less likely to oscillate, compared to multi-pair fet OS, so I hope it should be easy to couple these two.
Another possible cause of such an instability is thermal oscillation, but I don't see where it could be generated here. Try a cold spray on some components, who knows?
In the schematic of the final version, I have forgotten to remove C25:
https://www.diyaudio.com/community/threads/tandem-based-amplifiers.388400/post-7094306
I have made a comparison between a conventional EF output and the tandem:
https://www.diyaudio.com/community/threads/tandem-based-amplifiers.388400/post-7094306
I have made a comparison between a conventional EF output and the tandem:
Maybe, but in the real world it is the opposite, and in the frequency domain, C25 causes a significant peaking. Having the PCB ready for any eventuality is therefore a wise precaution.
To remove any remaining doubt, I made an additional test, this time in global feedback mode: as the phase response showed a quirk around 10MHz, it could explain the "chainsaw massacre" of the sim, with the front-end present.
This is the result:
Absolutely spotless too (with a NE5534)
This is the result:
Absolutely spotless too (with a NE5534)