Dadod,
The peak current out of the IPS into the TIS will be related to the error signal. The faster the amp, the smaller the error signal.
So, feed a square wave into the amp with very fast rise times ( I used 10 ns although this of course will never happen in the real world. Make sure you have no input filter connected and you are driving the amp into its large signal range (try 40 V pk)
Monitor the output current of the diamond. What you should see is a square wave with overshoot. If you have you comp correct, the peak overshoot current should still leave the diamond current in class A -I.e no switch off of 1 half.
If you are going into class B, the either raise the feedback resistor, or, apply some front end filtering. In my case, I applied front end filtering ( -3 dB 1.5 MHZ) and this kept the front end in class A.
The peak current out of the IPS into the TIS will be related to the error signal. The faster the amp, the smaller the error signal.
So, feed a square wave into the amp with very fast rise times ( I used 10 ns although this of course will never happen in the real world. Make sure you have no input filter connected and you are driving the amp into its large signal range (try 40 V pk)
Monitor the output current of the diamond. What you should see is a square wave with overshoot. If you have you comp correct, the peak overshoot current should still leave the diamond current in class A -I.e no switch off of 1 half.
If you are going into class B, the either raise the feedback resistor, or, apply some front end filtering. In my case, I applied front end filtering ( -3 dB 1.5 MHZ) and this kept the front end in class A.
Distortion level
There is a good reason why CD had been established at 16 bits = 97db of signal/noise ratio.
It is because, at normal listening levels and environments , noise will never be audible. (24 bits are just for better little signals definition).
And, here, the noise can been not masked by signal during silences, while HD is.
Considering:
1- That the first harmonics will yet be present in any acoustic source.
2- The mask effect.
3- The natural distortion inside our ears.
4- The distortion generated by any electro-acoustic device that we use to transform signals in acoustic pressure.
4- The short duration of highest peak signals.
Do you really think that any harmonic distortion below -100dB (0.001%) has any chance to be perceived or make *any difference* ?
Music is made of transients. Don't you think any overshoots (or rounded corners) that we can see in square waves reproduction will have lot more influence on the energy transmitted to the speakers, means real relative levels of those transients ? Hence, the importance of slew rates.
There is a good reason why CD had been established at 16 bits = 97db of signal/noise ratio.
It is because, at normal listening levels and environments , noise will never be audible. (24 bits are just for better little signals definition).
And, here, the noise can been not masked by signal during silences, while HD is.
Considering:
1- That the first harmonics will yet be present in any acoustic source.
2- The mask effect.
3- The natural distortion inside our ears.
4- The distortion generated by any electro-acoustic device that we use to transform signals in acoustic pressure.
4- The short duration of highest peak signals.
Do you really think that any harmonic distortion below -100dB (0.001%) has any chance to be perceived or make *any difference* ?
Music is made of transients. Don't you think any overshoots (or rounded corners) that we can see in square waves reproduction will have lot more influence on the energy transmitted to the speakers, means real relative levels of those transients ? Hence, the importance of slew rates.
Richard,
Interactions between the different stages are a fact of life. IIRC the problem we had was with the current mirror TIS. Thinking back about it now, it actually was not the best idea for a power amp. In a conventional MC TIS, the output impedance at HF goes down as F rises due to the action of Cdom. So, it's actually a great way to drive an EF2 or triple. With a CM TIS you don't have that local feedback, so the non- linear loading of a big output stage is very problematic.
So, I think there's a solid engineering reason for what we observed. And, if we consider MIC, the same results as MC can be predicted - also better than a CM TIS.
The OPS is still the biggest contributor to overall distortion in a class AB. Easy to prove in sim as I noted in an earlier post today.
Interactions between the different stages are a fact of life. IIRC the problem we had was with the current mirror TIS. Thinking back about it now, it actually was not the best idea for a power amp. In a conventional MC TIS, the output impedance at HF goes down as F rises due to the action of Cdom. So, it's actually a great way to drive an EF2 or triple. With a CM TIS you don't have that local feedback, so the non- linear loading of a big output stage is very problematic.
So, I think there's a solid engineering reason for what we observed. And, if we consider MIC, the same results as MC can be predicted - also better than a CM TIS.
The OPS is still the biggest contributor to overall distortion in a class AB. Easy to prove in sim as I noted in an earlier post today.
Last edited:
Absolutely correct. In virtually ANY topology with a well-designed front-end, the OPS distortion dominates. This is usually true for both static and dynamic (HF) distortions. It is even more true for high-order distortion.
Cheers,
Bob
Two questions.
- How can-we measure the distortion of a VAS when there is no other feedback applied to it than the global ?
- Why to concentrate more on a stage than an other when the goal is to reduce overall distortion ?
My amp use two pairs of FETs in OPS. 150mA each.
Where it matters, in musical reproduction, it is class A. I mean signals long enough for any HD distortion can be discriminated, it is class A. There, the OPS distortion is not so high ?
Cherry, “Estimates of Nonlinear Distortion in Feedback Amplifiers” JAES V48#4 2000
https://secure.aes.org/forum/pubs/journal/?elib=12068
don't think its online for free anywhere, I have (poor quality) paper photo copy only from the library from the month it was published
but it really looks like a great approach to the question of how much each device's nonlinearities are contributing
of course you need sensitivity numbers from the full admittance matrix linearized amp representation for the math
with work you could extract the sensitives in sim
Cherry shows hadware meaurements agreeing with his method, changes bias - correctly predicts results on the hardware
https://secure.aes.org/forum/pubs/journal/?elib=12068
don't think its online for free anywhere, I have (poor quality) paper photo copy only from the library from the month it was published
but it really looks like a great approach to the question of how much each device's nonlinearities are contributing
of course you need sensitivity numbers from the full admittance matrix linearized amp representation for the math
with work you could extract the sensitives in sim
Cherry shows hadware meaurements agreeing with his method, changes bias - correctly predicts results on the hardware
Last edited:
I must beg to differ.
IMHO, the effect of Miller, TPC, MIC which tend to make the VAS output LoZ, is purely evil. 😱
LoZ is the worst kind of drive for a Class B OPS. The evils are enumerated in Self. You are also slugging the amp at the point when it needs to be able to slew fastest to get past the yucky xover bit.
Holy 'pure Cherry' compensation (and to a lesser extent TMC) keeps that point as unencumbered as possible so maintains HiZ drive. By current driving, you exchange the high order xover type distortions for low order beta droop type stuff which is much more amenable to the loadsa extra LG that current drive also gives you.
For me it was a 'Ah Ha!' moment in PA design.
Of course you have to deal with slight problems with stability but these aren't insurmountable with VFAs (see tpc-vs-tmc-vs-pure-cherry) 😀
But I dunno how to deal with these properly in CFAs at the moment. 😡
OK. I confess to some hyperbole in claiming the OPS THD is inconsequential in my examples. Mea maxima culpa 🙂
But the evil practice of using perfect OPS to test earlier blocks must stop. At least use a resistive load which approximates what the OPS + 8R load would present to the VAS.
As you say, the interaction is a fact of life. I'll go further and say the interaction is the most important 'feature'.
There are many ways to look at the problem. Some more useful than others.
______________
In the discrete-opamp-open-design.html thread, the two top performers are Guru Wurcer's SWOPA and my FET990.
Scott uses a 2 block approach An Operational Amplifier Architecture with a Single Gain Stage and Distortion Cancellation as in his AD797. One block is IPS/VAS and the other OPS.
I also use a 2 block approach but one is the IPS and the other VAS/OPS. They have very similar THD figures on similar loads.
But because I deal with the VAS/OPS interaction explicitly, FET990 has far fewer and lower level high order harmonics than Scott's approach .. ie much less xover distortion. And more important for me, a much simpler topology for da supa dupa performance. 🙂
Scott's complex supa dupa diamond OPS still shows xover distortion. So in this case, I'll claim I'm dealing with the more important 'interaction' by considering VAS/OPS as one 'block'.
________________________
Trying to organise Linuxguru to get some samples to Mr. Marsh to confirm dis SPICE world stuff. 😉
Last edited:
Richard, I would not call MC slugging the TIS/VAS. This is straight feedback, all be it frequency dependent due to Xc.
Even if you close the front end with TPC or MIC, the effect is still to lower the output impedance of the front end - which is exactly what we need in order to deal with the non linear input impedance of the OPS.
The other option is to raise the OPS IP impedance, for example by going to an EF3.
My approach is to make each stage as linear as possible, and then close the overall NF loop with due regard to the stability requirements. See Bob's book where he shows the evolution of a low distortion design (chapter 4 IIRC).
I seem to recall Self saying that MC was remarkably effective given its simplicity. Now, this is not my endorsement of MC as the b all and end all, but it does a good job and we know in engineering terms why.
Even if you close the front end with TPC or MIC, the effect is still to lower the output impedance of the front end - which is exactly what we need in order to deal with the non linear input impedance of the OPS.
The other option is to raise the OPS IP impedance, for example by going to an EF3.
My approach is to make each stage as linear as possible, and then close the overall NF loop with due regard to the stability requirements. See Bob's book where he shows the evolution of a low distortion design (chapter 4 IIRC).
I seem to recall Self saying that MC was remarkably effective given its simplicity. Now, this is not my endorsement of MC as the b all and end all, but it does a good job and we know in engineering terms why.
Last edited:
Not exactly on topic but if you look at the thd of the driver (or any other stage) to the output stage you will measure the amount the ips/vas have to "distort" in in order to correct for the ops. So I think I am understanding Richard to be saying that these nodes must have the bandwidth to do so and linearity is secondary.
Easier (for me) to start with an real output stage and determine the amount of gain/bandwidth (and slew rate if dominated by the cross-over) needed from and ideal driver (with the ideal driver having adjustable gain and bw and other parameters to achieve some specific thd) and then work my way to realizing those parameters with real drivers.
Bonsai, what do you mean by entering class b on the input stage (for a cfa)?
For example, I would for practical simplified purposes, use when the "offs" transistor current has dropped to 1/4 its nominal value or are you only worried about stored charge effects?
Thanks
-Antonio
Easier (for me) to start with an real output stage and determine the amount of gain/bandwidth (and slew rate if dominated by the cross-over) needed from and ideal driver (with the ideal driver having adjustable gain and bw and other parameters to achieve some specific thd) and then work my way to realizing those parameters with real drivers.
Bonsai, what do you mean by entering class b on the input stage (for a cfa)?
For example, I would for practical simplified purposes, use when the "offs" transistor current has dropped to 1/4 its nominal value or are you only worried about stored charge effects?
Thanks
-Antonio
By taking the compensation from the VAS collector, you immediately limit the HF response AND slew of the amp up to that point.
I want the HF response at that point to be as fast as possible so the amp gets across the yucky xover region as quickly as possible.
The result is much less higher harmonics and also the amp becomes MUCH less critical as to Iq. The residual THD goes from spiky to small bumps. I show this in my sims of MikeK's cr*p TPC/TMC suggestions.
At the same Iq, high order harmonics are ALWAYS less with 'pure Cherry'.
The best alternative to 'pure Cherry' was Dadod's Little Gem TMC but that was still 6dB worse in THD.
With 'pure Cherry', only Cob of the main VAS device & strays limit the HF response at that point ... so I think 'slugged' is appropriate. 🙂
[/QUOTE]Even if you close the front end with TPC or MIC, the effect is still to lower the output impedance of the front end - which is exactly what we need in order to deal with the non linear input impedance of the OPS.
The other option is to raise the OPS IP impedance, for example by going to an EF3.
My approach is to make each stage as linear as possible, and then close the overall NF loop with due regard to the stability requirements. See Bob's book where he shows the evolution of a low distortion design (chapter 4 IIRC).
I seem to recall Self saying that MC was remarkably effective given its simplicity. Now, this is not my endorsement of MC as the b all and end all, but it does a good job and we know in engineering terms why.[/QUOTE]I know and have studied, simmed, tried in 'real life', bla bla all that. Self does an excellent job showing Miller's good points ... but also shows the clear limits to better performance with Miller.
I'm taking a completely opposite tack but sorta in line with Prof. Cherry to go beyond these limits.
At the end of the day, its what can be achieved with each approach which is important.
I've challenged yus to do better in my tpc-vs-tmc-vs-pure-cherry thread. I'll add da MIC gurus to da TMC/TPC challengers. Shouldn't be any threat as MIC is an 'evil LoZ drive to OPS' technique and has no chance of beating Holy HiZ 'pure Cherry' 😀
The Heretical HiZ champion to beat is in #499 of this thread with performance in #823
If you can equal or beat this performance with equal or less complexity I will grovel at your feet. 😱
I only wish I knew how to do this with CFAs 😡
____________________
BTW, what's wrong with a CFA IPS entering Class B? What evils does this bring?
Self, on MC, p64 Fifth edition
"It is often said that the use of a high VAS collector impedance provides a current drive to the output stage, often with the implication that this somehow allows the stage to skip quickly and lightly over the dreaded crossover region. This is a misconception -" etc.
Hm.
I have the excellent Cherry article that JCX referenced in post 1449 and have mulled it over but no conclusion yet. Anyone else worked it out?
Best wishes
David
Last edited:
Well, I think we are mixing up output stage drive, local feedback loops and global negative feedback.
The bandwidth of the front end has to be managed in a GNFB amp, because you have to deal with the OPS phase shift - whether thats through PC ('pure cherry'), MC, TMC, TPC or MIC - take your pick.
Not one of these comp methods can operate without considering the OPS phase shift - they all find a way to deal with it.
I think MC is wasteful, because it places a 20 dB/decade pole at LF, whereas TPC for example allows higher feedback at HF because of the dual slope roll off after the pole. So, if you call MC 'slugging it' in that context then I would agree. But, this is a different discussion when considering the TIS output impedance and we should not confurse the two IMV.
😎
The bandwidth of the front end has to be managed in a GNFB amp, because you have to deal with the OPS phase shift - whether thats through PC ('pure cherry'), MC, TMC, TPC or MIC - take your pick.
Not one of these comp methods can operate without considering the OPS phase shift - they all find a way to deal with it.
I think MC is wasteful, because it places a 20 dB/decade pole at LF, whereas TPC for example allows higher feedback at HF because of the dual slope roll off after the pole. So, if you call MC 'slugging it' in that context then I would agree. But, this is a different discussion when considering the TIS output impedance and we should not confurse the two IMV.
😎
Last edited:
CFA class B operation.
some people have commented that if the small signal stages transition into class B operation this is probably not a good thing because it will affect the sonics.
I have investigated this, and its very easy to ensure the front end stays in class A.
For an IC designer, where power consumption is an important specification, I can see that they would be ok with class B operation as you can run the front end really lean and still get very high slew rates. This of course is counter intuitive for for VFA where the expectation (esp. with MC designs) is that lower tail currents mean lower slew rates.
some people have commented that if the small signal stages transition into class B operation this is probably not a good thing because it will affect the sonics.
I have investigated this, and its very easy to ensure the front end stays in class A.
For an IC designer, where power consumption is an important specification, I can see that they would be ok with class B operation as you can run the front end really lean and still get very high slew rates. This of course is counter intuitive for for VFA where the expectation (esp. with MC designs) is that lower tail currents mean lower slew rates.
Last edited:
You can test this easily.
Sim a simple Blameless with Miller like fig 6.16 of Self 4 ed. This is what my tpc vs tmc vs pure Cherry efforts are based on. Drop Iq to 20mA or less to increase the dreaded crossover region.
Feed a 20kHz high level signal and look at the distortion residual. Look also at the voltage at the VAS output. Try different voltage levels.
Do the same with my #502 circuit with 'pure Cherry'.
Vary the compensation caps on these circuits to see what effect these have.
Better still, do this in 'real life'. You can look at the VAS output without a THD analyser. Just need to reduce Iq more to see stuff on a scope. Try a fast square wave instead.
You'll see the Self Miller circuit has serious slew limitations across the dreaded xover region while 'pure Cherry' will 'skip quickly & lightly'.
I've done all this in 'real life'.
Du.uuh!
Stirring Jurassic memories brings back a dim recollection that this is easier to see in MOSFET OPS cos the 'dreaded xover region is larger'.
(Probably why Self hates MOSFET OPS 🙂)
This example shows what you get if Iq is very small. If you are versed in the art, you can extrapolate what is happening from the THD residuals at more sensible Iq & THD levels but I'm just going to show what you might expect from a scope on the VAS output.
The MOSFET example is fairy tale cos I don't have any EKV models (essential if you are looking at THD)
Cherry.gif is the circuit used. To sim the Miller version, C3 to the output end of R8 is transferred to Q4 collector.
The Red curve in MillerWave.gif is Q4 collector while the Blue curve is the junction of the 2 MOSFET sources. There is clear slew limiting at the xover point at Q4; bad enough to cause a bump in the output.
MillerXover.gif shows this in detail.
CherryXover.gif shows the transition is faster & sharper; 'skipping quickly & lightly over the dreaded xover region' so the final output is less wonky.
My apologies for these crude examples. I no longer have my Jurassic notes on the 'real life' experiments.
But as I've said, the real proof of the pudding is the final result ... for which I refer you to #499 with performance in #823 and invite you to do better using your own prejudices & pet theories. 🙂
If you can do this simply with a CFA, I will grovel even lower 😱
Attachments
Last edited:
Richard, I would not look at this from a speed POV or describe it as 'skipping' over the cross over region more rapidly.
By using a more sophisticated comp design, you have simply been able to apply more feedback around the loop at HF. And, one of the requirements to be able to do this is to have a wider loop bandwidth - which is exactly what TMC, TPC et allow you to do.
Another way to look at this is to think about MC desings that have very high slew rates (easy to make a VFA with 150 V/us). You do not necsessarily get lower distortion than a design with a lower slew rate. All you have to do to show this is change the tail current and close the loop at some fixed frequncy - say 1 MHz.
Of course, in testing this, you have to make sure the amplifier does not slew.
By using a more sophisticated comp design, you have simply been able to apply more feedback around the loop at HF. And, one of the requirements to be able to do this is to have a wider loop bandwidth - which is exactly what TMC, TPC et allow you to do.
Another way to look at this is to think about MC desings that have very high slew rates (easy to make a VFA with 150 V/us). You do not necsessarily get lower distortion than a design with a lower slew rate. All you have to do to show this is change the tail current and close the loop at some fixed frequncy - say 1 MHz.
Of course, in testing this, you have to make sure the amplifier does not slew.
I was too lazy to explain... Let's look at the upper side.
If there is not enough current in the (traditional) symmetrical input stage, reproducing a sinus, the voltage can reach the point where it is lower than the BE diode of the VAS.
No more conduction: class B. My question was: What the problem if at this time, the VAS was yet in class B ?
In some sims, it seems the HD can be reduced that way.
Last edited:
In a traditional symmetrical VAS you are unlikely to get one half turning off completely. The biger issue there is that one half of the LTP switches off - which is whay happens if your amp slews. But to fix it is very easy through the correct compensation.
For CFA, its also easy to fix.
I thing for high quality audio, you always want the front end to be operating in class A. Recovery from class B is never going to be without some artifacts getting into the signal - but thats just my personal view - I have not checked this in a sim.
For CFA, its also easy to fix.
I thing for high quality audio, you always want the front end to be operating in class A. Recovery from class B is never going to be without some artifacts getting into the signal - but thats just my personal view - I have not checked this in a sim.