i appreciate the efforts of PB2 to get this discussion underway and the others chiming in with results.
i think this is a good opportunity to really flush out how far off the models are and what you can really trust the simulators for, as opposed to critiquing the adcom design. not to say such critiquing wouldn't be educational or otherwise useful. the market has proven the real amplifier to be a fairly bullet-proof design when the correctly specified parts are used. in my mind, once the simulation accuracy and precision has been baselined, the improvements and tweaking can commence in a very logical fashion.
just my unsolicited opinion.
carry on chaps,
back to lurking now ...
mlloyd1
i think this is a good opportunity to really flush out how far off the models are and what you can really trust the simulators for, as opposed to critiquing the adcom design. not to say such critiquing wouldn't be educational or otherwise useful. the market has proven the real amplifier to be a fairly bullet-proof design when the correctly specified parts are used. in my mind, once the simulation accuracy and precision has been baselined, the improvements and tweaking can commence in a very logical fashion.
just my unsolicited opinion.
carry on chaps,
back to lurking now ...
mlloyd1
i appreciate the efforts of PB2 to get this discussion underway and the others chiming in with results.
i think this is a good opportunity to really flush out how far off the models are and what you can really trust the simulators for, as opposed to critiquing the adcom design. not to say such critiquing wouldn't be educational or otherwise useful. the market has proven the real amplifier to be a fairly bullet-proof design when the correctly specified parts are used. in my mind, once the simulation accuracy and precision has been baselined, the improvements and tweaking can commence in a very logical fashion.
just my unsolicited opinion.
carry on chaps,
back to lurking now ...
mlloyd1
Hi mlloyd1,
Have you done such comparisons? I've not noticed on the board if you have or not.
BTW, we don't even have models for many of the exact transistors - not that I
consider this to be a major issue.
hi pete:
no i haven't done such comparisons. not enough time or resources anymore. that's why i was excited to see this discussion start.
no doubt that a lack of good, quality models can be challenging. andy_c did a great job with the improved models he supplied.
one of the things i miss about my previous job was access to well maintained and calibrated curve tracers and the big $$ versions of models and modeling software. on the other hand, one of the things i don't miss is 15 hr days, 6 and 7 days a week ...
mlloyd1
no i haven't done such comparisons. not enough time or resources anymore. that's why i was excited to see this discussion start.
no doubt that a lack of good, quality models can be challenging. andy_c did a great job with the improved models he supplied.
one of the things i miss about my previous job was access to well maintained and calibrated curve tracers and the big $$ versions of models and modeling software. on the other hand, one of the things i don't miss is 15 hr days, 6 and 7 days a week ...
mlloyd1
I've done quite a bit of simulation over the years, yes
with high end tools. I have a client who will probably
be getting a curve tracer within the next six months.
A Stab at Stability Mods
See mods for improved stability in the attached file,
added R4, C7, and C12.
For those concerned about models, try changing the VAS
for example to an mje350, KSA1142, or 2N5401 look at the
loop gain again. Or try changing the diff pair or any other
small signal transistor.
Then try the large and slow mje15033 for the VAS, check the
loop gain again. This part is obviously not a good choice.
See mods for improved stability in the attached file,
added R4, C7, and C12.
For those concerned about models, try changing the VAS
for example to an mje350, KSA1142, or 2N5401 look at the
loop gain again. Or try changing the diff pair or any other
small signal transistor.
Then try the large and slow mje15033 for the VAS, check the
loop gain again. This part is obviously not a good choice.
Whatever the transistors, the compensation mods you are proposing
reduce the gain margin and phase margin by a good chunk, that is,
they reduce stability..
As you didn t remove the input capacitor of 300pF, all simulations
will be innacurate since what is mesured is not the real phase and gain respônse of the amp, but the serial response of two systems :
First system is : Low pass filter consisting of R1 and C1 which
form a first order system...
Second system is the amp s transfer function which is second order
in that case.
The input low pass filter will reduce the generator output level at high
frequency leading to the false conclusion that the amp is stable
while in fact it isnt..
I let other people make the relevant and blatant conclusions...
Do you even understand that the source at the input is not
used for this test, rather it is V4 in the feedback loop? The
input filter must stay there because it is part of the design
though it has little effect. You seem to jump into this
assuming that I know nothing of what I am doing - how
rude of you and indeed you demonstrate that you are the
one who is confused. You are confusing open loop gain with
the feedback loop gain.
I should also mention that the mods I show are not required
but they do improve the gain and phase margin somewhat.
I have asked you several times to post your .asc file so that
we can see how you are coming to your, so far, flawed
conclusions.
I suggest that you consider the tone of so many of your
replies, even if you were correct in your conclusions it is
uncalled for, but especially not when you are incorrect.
used for this test, rather it is V4 in the feedback loop? The
input filter must stay there because it is part of the design
though it has little effect. You seem to jump into this
assuming that I know nothing of what I am doing - how
rude of you and indeed you demonstrate that you are the
one who is confused. You are confusing open loop gain with
the feedback loop gain.
I should also mention that the mods I show are not required
but they do improve the gain and phase margin somewhat.
I have asked you several times to post your .asc file so that
we can see how you are coming to your, so far, flawed
conclusions.
I suggest that you consider the tone of so many of your
replies, even if you were correct in your conclusions it is
uncalled for, but especially not when you are incorrect.
Last edited:
Not at all, as everyone deserve the benefit of the doubt,
and i m eager to learn as well from anybody having
knowledge, including you of course..
I will mention, that i did the sim loop closed to check the
inluence of C4/R4 and C12.
I just want to bring to your attention that measurement
must be careful to not break compensation loops..
Isn't the target Phase Margin 55 +/- 10 degrees?
Pete looks like you are adding a couple of zeros to cancel poles at 2-3 MHz, Do we know the origin of those poles? In other words, how confident is the model vs actual.
I still think we should baseline all the sims we are planning on before mod changes so we know later if improving one parameter does not negatively affect another.
Yes, exactly, the big hit is the addition of R4 and C12 which provides a zero in the main gain stage and lead compensation. You can see that the result increases the phase margin to over 60 degrees all the way out to 18 MHz. The phase zero crossing is moved to 45 MHz where there is 50 dB of gain margin. I don't really think it matters if the model is accurate at 10s of MHz since there is so much more margin. Did you try the transistor subs that I mentioned - it is clear that as long as the VAS has reasonably low capacitance things don't move around much. This proves my earlier claim that as long as the comp caps are significantly larger than the parasitics stability is not highly dependent on having exact device models. I am reasonably confident with the results, however I'd like to see the effect of slower output devices. If we assume a much larger Cob for slower devices, say 1000pf more to be generous, then we could add 4000 pF from C to B in the output stage to model this.
The main pole is the "miller loop" made up of C4 around the VAS and diff amp, this is roughly at 100 Hz, and the additional resistor allows us to use the VAS gain to make up for other poles in the system. There is one around 1 MHz that is I think a combination of the output stage and the capacitive load on the VAS.
Bob Cordell mentions that by using C4 the diff pair - VAS slew rate is not tied into the dominant pole.
An analytical solution would be nice, but I did not plan to do anything regarding stability and I think this is a fairly good solution. I would proceed with caution if I ever implement it.
I divide up the feedback resistor and put a huge cap to ground on it to provide a DC path to balance the output stage but shunt the AC in order to look at the open loop response. This file is attached for the unmodified design and you can look at the main amp output as compared to the VAS output to get a better idea of where the poles are. Snip out the compensation caps to see what they do.
Don't recall if I posted this link about modeling:
http://www.intusoft.com/lit/WkwModels.pdf
http://www.intusoft.com/lit/WkwModels.pdf
Total lead compensation reduced to 8pF + 8pF.
Shunt compensation increased to 1nF + 1nF, that HAS
the same effect as increasing C2 to a value of about 47pF...
A little more effort, and we ll reach this stage :
http://www.diyaudio.com/forums/soli...simulation-adcom-gfa-555-a-9.html#post2134452
Not at all..
The two 1 nF cap from base to collectors of drivers
kill it much more...
Here a comparison of the slew rates with a positive
step of 1V at 10KHZ.
The caps references are relative to your asc file
in post 114.
Yes we need to get baseline Sims for all the important amplifier specs before mods are proposed. My mods were mostly from looking at the transfer function of the buffer stage. There is an undamped resonance near ft that affects PM, obviously brute force caps are not ideal. Perhaps that is why higher ft output devices give better stability?
You're comparing a .5A device to an 8A device, apples and oranges comparison.
Viewers can do what I said as substitutes go and see the reality, the transfer
function does not change much for similar devices, which suggests that exact
models are not needed. Know your tools, models and their limitations.
By the way guys I didn't start this thread to ask for help or other "opinions" LOL!
Just thought others might want to try out LTSpice or have a look at the 555.