Apart from not 'having all the time in the world', I'm normally not inclined to criticize someone else's design. But in this case I will make an exception, because YAP, in particular the OPS, is 'cribbed' form my NFB-OPS and I don't like to be associated with the flaws of this design anyhow. Therefore I will continue to point out what's wrong with YAP. This time the flaws that add to distortion will be commented.
Opposed to THD20=1.65ppm as reported by Mr P, my sim reveals 21ppm. Such gross difference is very strange, the more so as simulations normally underestimate THD figures. At least two things are responsible for this unnecessary high distortion: the clamp and input stage of the OPS. Both represent a non-linear load of current mirror Q7/Q8 and both are easily avoided by:
1. Using a “Flying Baker Clamp” (see also: http://www.diyaudio.com/forums/showthread.php?postid=1434496#post1434496 )
2. Using Q11/Q12 in a diamond configuration.
Starting with the latter, reconfiguring these trannies not only takes care of temp. compensation, but also provides extra gain and isolates the nonlinear Cob. With only this simple modification the THD20 drops to 10ppm. (BW=200kHz)
Using a flying Baker clamp results in a further decrease to 2.7ppm. But we are still not done. What are C8,C13, R6 and R13 for? Looks like TMC, but it certainly doesn't improve THD. Without these components, I finally get 1.6ppm, almost the same value as Mr P. has measured.
This exercise leaves us with two questions:
1. Why this gross discrepancy between simulation and measurement? Are the models that bad? I don't think so. Are different distortion mechanism (in the real amp) canceling each other out? Maybe. Or is the schematic as published by Mr P not correct. Perhaps, as there were also other errors.
2. Despite that both techniques have been discussed on this forum and are well known, why are these simple techniques not used in an amp that is targeted at ultra low distortion?
Anyhow, I'm curious about the answers.
PS: One remark about CCSs I4 and I5. If the OPS has also some voltage gain, then these CCSs may simply consist of a few resistors. For an example see R37...R40 in http://home.tiscali.nl/data.odyssey/MCP-FE.html
Regrettably, this simple solution is not an option for Mr P. as he don't like CFB plus voltage gain (too difficult to analyze), see: http://www.diyaudio.com/forums/showthread.php?postid=1595672#post1595672 2nd paragraph
and my analysis of H(s) (see: http://www.diyaudio.com/forums/showthread.php?postid=1374791#post1374791 ) wasn't good enough for him (I'm still waiting for his version).
Curiously, he appears to be unaware that even in the early sixties (before a LTP was commonly used in the IPS) the IPS of a many power amps were using CFB + V-gain and nobody was moaning about analyzing problems. The IPS of those amps comprises just one tranny. The base was the non-inverting input and the emitter, tied to a resistive divider, the inverting (read: CFB!) input.
But that's not all. The IPS of YAP contains an op-amp, which is configured as ......CFB input stage with voltage gain. Again analyzing problems Mr P? Of course not as in this case the well respected Mr Alexander has already solved that for him, instead of a ****ng PITA like me.
Opposed to THD20=1.65ppm as reported by Mr P, my sim reveals 21ppm. Such gross difference is very strange, the more so as simulations normally underestimate THD figures. At least two things are responsible for this unnecessary high distortion: the clamp and input stage of the OPS. Both represent a non-linear load of current mirror Q7/Q8 and both are easily avoided by:
1. Using a “Flying Baker Clamp” (see also: http://www.diyaudio.com/forums/showthread.php?postid=1434496#post1434496 )
2. Using Q11/Q12 in a diamond configuration.
Starting with the latter, reconfiguring these trannies not only takes care of temp. compensation, but also provides extra gain and isolates the nonlinear Cob. With only this simple modification the THD20 drops to 10ppm. (BW=200kHz)
Using a flying Baker clamp results in a further decrease to 2.7ppm. But we are still not done. What are C8,C13, R6 and R13 for? Looks like TMC, but it certainly doesn't improve THD. Without these components, I finally get 1.6ppm, almost the same value as Mr P. has measured.
This exercise leaves us with two questions:
1. Why this gross discrepancy between simulation and measurement? Are the models that bad? I don't think so. Are different distortion mechanism (in the real amp) canceling each other out? Maybe. Or is the schematic as published by Mr P not correct. Perhaps, as there were also other errors.
2. Despite that both techniques have been discussed on this forum and are well known, why are these simple techniques not used in an amp that is targeted at ultra low distortion?
Anyhow, I'm curious about the answers.
PS: One remark about CCSs I4 and I5. If the OPS has also some voltage gain, then these CCSs may simply consist of a few resistors. For an example see R37...R40 in http://home.tiscali.nl/data.odyssey/MCP-FE.html
Regrettably, this simple solution is not an option for Mr P. as he don't like CFB plus voltage gain (too difficult to analyze), see: http://www.diyaudio.com/forums/showthread.php?postid=1595672#post1595672 2nd paragraph
and my analysis of H(s) (see: http://www.diyaudio.com/forums/showthread.php?postid=1374791#post1374791 ) wasn't good enough for him (I'm still waiting for his version).
Curiously, he appears to be unaware that even in the early sixties (before a LTP was commonly used in the IPS) the IPS of a many power amps were using CFB + V-gain and nobody was moaning about analyzing problems. The IPS of those amps comprises just one tranny. The base was the non-inverting input and the emitter, tied to a resistive divider, the inverting (read: CFB!) input.
But that's not all. The IPS of YAP contains an op-amp, which is configured as ......CFB input stage with voltage gain. Again analyzing problems Mr P? Of course not as in this case the well respected Mr Alexander has already solved that for him, instead of a ****ng PITA like me.
Attachments
Edmond,
I'd be curious to see what your clamping diode model looks like. Back in the VAS clamping diodes thread, I ended up doing a model of the BAV21, a diode similar to the one used in YAP. One thing I found in fitting the capacitance parameters was a very surprisingly low value of the "M" parameter - around 0.08. Usually this is between 0.33 and 0.5. The very low value of M gives a weak dependence of capacitance vs. reverse voltage, which helps with distortion. My BAV21 model is here. See the discussion above that for my findings with regard to the M parameter.
Also, I'm not seeing how the circuit you posted constitutes a flying Baker clamp. My understanding of that technique from the link to Bob's post on the subject is that, using the positive side as an example, the cathode of the clamping diode should be bootstrapped to a level-shifted version of the output voltage. Just a rough look at the schematic you posted leads me to believe that may not be the case. In the absence of R112, the voltage at the cathode of D3 looks like it would be determined by a capacitive voltage divider. Yet R112 looks to me like its right side is connected to a voltage that's almost constant with time. Of course, I could be in error on this, but would appreciate an explanation of what I'm missing if so.
I'd be curious to see what your clamping diode model looks like. Back in the VAS clamping diodes thread, I ended up doing a model of the BAV21, a diode similar to the one used in YAP. One thing I found in fitting the capacitance parameters was a very surprisingly low value of the "M" parameter - around 0.08. Usually this is between 0.33 and 0.5. The very low value of M gives a weak dependence of capacitance vs. reverse voltage, which helps with distortion. My BAV21 model is here. See the discussion above that for my findings with regard to the M parameter.
Also, I'm not seeing how the circuit you posted constitutes a flying Baker clamp. My understanding of that technique from the link to Bob's post on the subject is that, using the positive side as an example, the cathode of the clamping diode should be bootstrapped to a level-shifted version of the output voltage. Just a rough look at the schematic you posted leads me to believe that may not be the case. In the absence of R112, the voltage at the cathode of D3 looks like it would be determined by a capacitive voltage divider. Yet R112 looks to me like its right side is connected to a voltage that's almost constant with time. Of course, I could be in error on this, but would appreciate an explanation of what I'm missing if so.
Hi Andy,
You are right that my version of the 'flying Baker clamp' differs from Bob's proposal, but the basic idea is the same:
Bob: "Normally a Baker Clamp diode sees the full p-p signal swing and its capacitance variation can cause some high-frequency nonlinearity at the output node of the VAS. What we would like is a clamp diode arrangement wherein under normal signal conditions both sides of the diode are moving equally with the signal, so that a high voltage diode is not needed and junction capacitance variation is minimized."
Also in my version, the voltage across D5 is held constant (about 0V) by means of R112, which also redirects leakage/reactive currents from D3 to a less sensitive node (R12-R115), instead of to the base of Q6 as in the original design.
I was using this model:
.MODEL 1N4148 D (BV=120 CJO=1.337037P IBV=100.0P IS=7.187424N M=0.316713486 N=1.94878 RL=3.521383G RS=439.577908M TT=8.952928N VJ=9.476621)
Using your BAV21 model (without flying hocus pocus) THD20=4.7ppm, indeed lower than 10ppm, as expected.
Cheers,
Edmond.
You are right that my version of the 'flying Baker clamp' differs from Bob's proposal, but the basic idea is the same:
Bob: "Normally a Baker Clamp diode sees the full p-p signal swing and its capacitance variation can cause some high-frequency nonlinearity at the output node of the VAS. What we would like is a clamp diode arrangement wherein under normal signal conditions both sides of the diode are moving equally with the signal, so that a high voltage diode is not needed and junction capacitance variation is minimized."
Also in my version, the voltage across D5 is held constant (about 0V) by means of R112, which also redirects leakage/reactive currents from D3 to a less sensitive node (R12-R115), instead of to the base of Q6 as in the original design.
I was using this model:
.MODEL 1N4148 D (BV=120 CJO=1.337037P IBV=100.0P IS=7.187424N M=0.316713486 N=1.94878 RL=3.521383G RS=439.577908M TT=8.952928N VJ=9.476621)
Using your BAV21 model (without flying hocus pocus) THD20=4.7ppm, indeed lower than 10ppm, as expected.
Cheers,
Edmond.
Re: yap
Well, sort of, that is, a precision audio power amp with distortion figures that are far below any hearing level. And nobody is supposed to listen to such an amp, as it will probably be a disappointing experience. No coloration, no nice 2nd harmonics etc. In short, as dull as can be.
Don't worry, I'm every evening in a sarcastic mood
Cheers,
Edmond.
ferencz said:I'm wondering is your design going to be a precision audio power signal generator, or an audio amplifier for music-listening purpose?
Well, sort of, that is, a precision audio power amp with distortion figures that are far below any hearing level. And nobody is supposed to listen to such an amp, as it will probably be a disappointing experience. No coloration, no nice 2nd harmonics etc. In short, as dull as can be.
Sry, I'm a bit sarcastic this evening
Don't worry, I'm every evening in a sarcastic mood
Cheers,
Edmond.
Edmond Stuart said:
Ah, I see. So the idea is to prevent the local feedback via the nonlinear diode capacitance. Seems like this should give similar results to eliminating D5 and tying the cathode of D3 to a fixed DC voltage (like what Bob did in his amp). Of course, that requires more parts. But in that case, the nonlinear diode current would also be dumped into a less sensitive node, namely AC ground.
andy_c said:
Hi Andy,
If you omit D5 and tie D3 to a fixed DC voltage, then, as far as I understand, it's no longer a Baker clamp, rather a kind of brute force thing that limits the VAS output voltage. I don't like that as large Ib and Ic currents are involved.
BTW, did you have a look at a previous post mine?: http://www.diyaudio.com/forums/showthread.php?postid=1613482#post1613482
Cheers,
Edmond.
Edmond Stuart said:
OK, now I see what happened. Here was your original statement:
What I was missing was what you later posted here - specifically this:
The original post of that circuit was back in the HEC thread. Specifically, what I missed was that the non-traditional compensation approach of returning the RC network back to the input stage emitter instead of ground, when used in a unity-gain configuration, gives a closed-loop BW for the circuit that is approximately the same as that of the driver-output stage combo by itself without feedback. So yes, putting an additional stage inside the Miller loop should have negligible effect on the closed-loop BW of the output stage, I agree.
I suspect that had already been mentioned back in the HEC thread when it was originally posted. But when that thread developed into a holy war, I stopped reading and posting to it.
Bonsai said:
I have since confirmed the 300KHz bump being a resonance between C9 and the output coil, dampened by the load and R30. I would think this resonance can safely be ignored.
I guess all amps using Zoebel and an output coil would have the same, but the limited overall bandwidth may hide it from measurements.
Edmond Stuart said:
Hey, a little OT - I've got some news to cheer you up, since I know that you think high dissipation amplifiers are evil.
I just realised that my K800AB amplifier isn't class AB at all and I'm going to have to rename it. Although I run a little higher VRE for better bias current stability, it qualifies as an efficent D.Self ClassB.
Like him, I'm running 250mA Iq per output trannie.
Edmond Stuart said:
I'm suspicious of the op-amp model. These aren't intended to model distortion, so it makes me wonder what the model's influence on the simulated distortion is here. Also, according to Analog Devices AN-138, these models have a somewhat idealized output stage. This might affect the relationship between output current and supply current on each rail, maybe creating some kind of crossover distortion-like phenomenon, but with the rail currents.
It might be interesting to do sort of the opposite of what is normally done. That is, use an actual diamond buffer output stage (maybe with 0.5 mA DC current), with an idealized input stage and VAS, and a DC current source thrown in to get the correct quiescent current. If that results in simulated distortion less than measured, it would really point the finger at the op-amp model.
Hi Andy,
I've already done that, also an idealized diamond buffer. THD figures got even slightly worse. So it's not the op-amp. It's clearly the nonlinear load of the VAS. As you know, I've checked that too. With a modified Baker clamp and a better input stage of the OPS, the distortion is almost gone. It's a great mystery to me. Does Mr P have access to distortionless trannies and diodes or are the models exaggerating the nonlinearities by a factor of 10 or so?
Cheers,
E.
PS: Thank for the interesting app. note.
I've already done that, also an idealized diamond buffer. THD figures got even slightly worse. So it's not the op-amp. It's clearly the nonlinear load of the VAS. As you know, I've checked that too. With a modified Baker clamp and a better input stage of the OPS, the distortion is almost gone. It's a great mystery to me. Does Mr P have access to distortionless trannies and diodes or are the models exaggerating the nonlinearities by a factor of 10 or so?
Cheers,
E.
PS: Thank for the interesting app. note.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.