Re: What could be done better...
Why you haven't used the (so much applauded by you) LT1166? This chip seems to solve all the bias problems at once.
Perhaps you can also try the FQA19N20/FQA12P20 pair from Fairchild. I've ordered them from Digi-Key and the Vgs mismatch varied from 10 to 30mV. As you see, not that much.
That's the price if you don't use your simulator exhaustively. 😀 I know it's time consuming, but at least you don't get dirty hands.
I'm afraid there might be another design error: the gate protection Zeners. Normally they are directly connected to the gates (and you'll need 6 of them). Why? Because a very fast transition of the drain might trigger a fault condition via Cgd. This happens when Vd/dt > Vgsmax / ( Rg * Cgd). Probably this will never happen in your application, but I would prefer to stay on the safe side.
syn08 said:
Why you haven't used the (so much applauded by you) LT1166? This chip seems to solve all the bias problems at once.
Perhaps you can also try the FQA19N20/FQA12P20 pair from Fairchild. I've ordered them from Digi-Key and the Vgs mismatch varied from 10 to 30mV. As you see, not that much.
That's the price if you don't use your simulator exhaustively. 😀 I know it's time consuming, but at least you don't get dirty hands.
I'm afraid there might be another design error: the gate protection Zeners. Normally they are directly connected to the gates (and you'll need 6 of them). Why? Because a very fast transition of the drain might trigger a fault condition via Cgd. This happens when Vd/dt > Vgsmax / ( Rg * Cgd). Probably this will never happen in your application, but I would prefer to stay on the safe side.
diamond driver
Now we are on cross purposes (my fault). An essential word was missing in my post. I mean: "a fair comparison of the two topologies is not possible when using at the same time two totally different output devices. IOW, using laterals in the one topology and verticals in the other topology gives no clue about the merits of the topology itself.
No, I 'quickly' replaced the wrong resistors values (supplied by you) by the correct ones. 😀 Now the circuit performs better.
Without C34 the turn-on drive is determined (and limited!) by the current sources, about 10mA. With C34 in place the drive current is boosted by Q30 or Q35 (only at fast transient, of course). A simulation would be very revealing.
That's right. Happily, I'm not indoctrinated by traditional designs. 🙂 But in case of the diamond driver, I'm still not convinced of the advantage over a traditional driver stage. Admittedly, it works well, but it adds to complexity.
syn08 said:
Now we are on cross purposes (my fault). An essential word was missing in my post. I mean: "a fair comparison of the two topologies is not possible when using at the same time two totally different output devices. IOW, using laterals in the one topology and verticals in the other topology gives no clue about the merits of the topology itself.
No, I 'quickly' replaced the wrong resistors values (supplied by you) by the correct ones. 😀 Now the circuit performs better.
Without C34 the turn-on drive is determined (and limited!) by the current sources, about 10mA. With C34 in place the drive current is boosted by Q30 or Q35 (only at fast transient, of course). A simulation would be very revealing.
That's right. Happily, I'm not indoctrinated by traditional designs. 🙂 But in case of the diamond driver, I'm still not convinced of the advantage over a traditional driver stage. Admittedly, it works well, but it adds to complexity.
Bob Cordell said:
Hi Bob,
I have no trouble if you are highlighting the other ways of looking at HEC, as you know better. However, I have trouble with people who do ignore or dismiss the NFB view, as they will never understand why frequency compensation is necessary.
I'm not happy with the 'free' bias generator.
As for higher frequencies (say 2 to 20kHz) I fully agree with that.
Indeed, that's a nice property. Also notice my remarks on BW in one my previous post, please: http://www.diyaudio.com/forums/showthread.php?postid=1603443#post1603443
True. I've seen that too, even the compensation!
BTW, someone who visited the Halcro factory once, told me that they don't have an AP or equivalent piece of equipment to check the distortion of every amp before it leaves the factory. What about that in connection with their sub ppm claims?
Cheers,
Edmond.
Re: Re: What could be done better...
1. You asked before - I answered before:
http://www.diyaudio.com/forums/showthread.php?postid=1478092#post1478092
LT1166 is not for the ultimate performance. Remember that I am using it in my 7x (actually 8x) home theater system, that has other performance criteria than very low THD20.
2. I did, both in PGP and in YAP and I don't like them at all, as much as I also dislike IRF240/IRF9240. The problem is the high level of crossover distortion, in particular at HF. I'm suspecting the core reason for this is the large difference in transconductance between the N and the P types, they are built for switching and optimized for matching Ciss. I wish I had the time to see how using these devices could be optimized for linear audio. Until somebody comes up with a proven solution, I don't see them fit for anything but switching and maybe Class A.
Don't fool yourself about the Vgs matching - at 5S transconductance, 10mV is a lot and in fact it's more than +/-20% in terms of Id at bias.
3. Well, admittendly I did not simulate the power dissipation under transient conditions when the protection is on. And I need no simulation to see that resistor burning up when the output is at zero - if I would look for such. But the MP930 Caddock power resistor will solve the problem nicely, without a PCB revision.
4. True, I have looked into the other option and I have decided is not worth the trouble to add 12 zener diodes to the PCB (that is, another 1.5 sq. in. to the PCB area)
Edmond Stuart said:
1. You asked before - I answered before:
http://www.diyaudio.com/forums/showthread.php?postid=1478092#post1478092
LT1166 is not for the ultimate performance. Remember that I am using it in my 7x (actually 8x) home theater system, that has other performance criteria than very low THD20.
2. I did, both in PGP and in YAP and I don't like them at all, as much as I also dislike IRF240/IRF9240. The problem is the high level of crossover distortion, in particular at HF. I'm suspecting the core reason for this is the large difference in transconductance between the N and the P types, they are built for switching and optimized for matching Ciss. I wish I had the time to see how using these devices could be optimized for linear audio. Until somebody comes up with a proven solution, I don't see them fit for anything but switching and maybe Class A.
Don't fool yourself about the Vgs matching - at 5S transconductance, 10mV is a lot and in fact it's more than +/-20% in terms of Id at bias.
3. Well, admittendly I did not simulate the power dissipation under transient conditions when the protection is on. And I need no simulation to see that resistor burning up when the output is at zero - if I would look for such. But the MP930 Caddock power resistor will solve the problem nicely, without a PCB revision.
4. True, I have looked into the other option and I have decided is not worth the trouble to add 12 zener diodes to the PCB (that is, another 1.5 sq. in. to the PCB area)
Re: diamond driver
I think I have mentioned this before: a simple driver with whatever C34 didn't do it. The amount of crossover conduction (I guess, the effect was increasing THD20 with the bias) was to high. Add to this some stability issues as described here:
http://www.diyaudio.com/forums/showthread.php?postid=1478092#post1478092
and you'll get the whole picture. But then I'm pretty sure if you are targeting 2.5MHz ULG, then a simple driver will do. It's simply that YAP relies on high(er) ULG rather than on linearization tricks (CMCL), high gain and steep rolloff as in your designs (as far as I can grasp so far). Both approaches are adding to complexity and which one is more effective is debatable. Some prefer to count the trannies in the signal path (which I think is stupid) others prefer to look at the costs (mostly in PCB area) and price/performance ratio.
Edmond Stuart said:
I think I have mentioned this before: a simple driver with whatever C34 didn't do it. The amount of crossover conduction (I guess, the effect was increasing THD20 with the bias) was to high. Add to this some stability issues as described here:
http://www.diyaudio.com/forums/showthread.php?postid=1478092#post1478092
and you'll get the whole picture. But then I'm pretty sure if you are targeting 2.5MHz ULG, then a simple driver will do. It's simply that YAP relies on high(er) ULG rather than on linearization tricks (CMCL), high gain and steep rolloff as in your designs (as far as I can grasp so far). Both approaches are adding to complexity and which one is more effective is debatable. Some prefer to count the trannies in the signal path (which I think is stupid) others prefer to look at the costs (mostly in PCB area) and price/performance ratio.
Re: Re: Re: Re: Re: HEC vs NFB
You appear to have a short memory. You may recall that I simulated your 1984 OS before you did. Have you simulated your own circuit yet? I also explained to you what FB loop gain you had (because you didn't know) and I verified it against your measured distortion reduction results. I provided you with mathematical models and simulations.
You have a lot vested in this old HEC circuit of yours. Maybe it is time to move on...come up with something new. 30dB at 40kHz isn't very special. And the manual trimming thing is a bit of a nonsense.
I think you are confusing diversity, which is respect for different facts, with support for different specious hypotheses.
Bob Cordell said:
You appear to have a short memory. You may recall that I simulated your 1984 OS before you did. Have you simulated your own circuit yet? I also explained to you what FB loop gain you had (because you didn't know) and I verified it against your measured distortion reduction results. I provided you with mathematical models and simulations.
You have a lot vested in this old HEC circuit of yours. Maybe it is time to move on...come up with something new. 30dB at 40kHz isn't very special. And the manual trimming thing is a bit of a nonsense.
I think you are confusing diversity, which is respect for different facts, with support for different specious hypotheses.
Re: Re: Re: What could be done better...
I'm sorry to say so, but your remarks are not 100% consistent: If you bypass the LT1166 with a large capacitor, which you did, see: http://www.diyaudio.com/forums/showthread.php?postid=1600377#post1600377
then the HF performance would be the same as in the case with a Vbe multiplier. So, what's the problem according to you? As a matter of fact, there is another problem, but you pretended to not understand that.
>LT1166 is not for the ultimate performance.
True, but delivers YAP the ultimate performance?
So I should be warned when using FQAxxx. Thanks for the hint.
I don't think so. The Toshiba's do 5S at Id=5A (the FQAxxx slightly more). 10mV * 5S = 50mA or 1% WRT 5A. Something wrong with my math?
Unfortunately 150ohm is not in stock, as you told us.
Moreover, if you had realized this before building the amp, then I don't exclude that you would have opted for a protection circuit that did not raise the dissipation in R52 to such a high level (which means a different PCB layout)
Indeed, even 12 zeners, not only 6 as I said. Under these circumstances I understand your trade-off arguments.
syn08 said:
I'm sorry to say so, but your remarks are not 100% consistent: If you bypass the LT1166 with a large capacitor, which you did, see: http://www.diyaudio.com/forums/showthread.php?postid=1600377#post1600377
then the HF performance would be the same as in the case with a Vbe multiplier. So, what's the problem according to you? As a matter of fact, there is another problem, but you pretended to not understand that.
>LT1166 is not for the ultimate performance.
True, but delivers YAP the ultimate performance?
So I should be warned when using FQAxxx. Thanks for the hint.
I don't think so. The Toshiba's do 5S at Id=5A (the FQAxxx slightly more). 10mV * 5S = 50mA or 1% WRT 5A. Something wrong with my math?
Unfortunately 150ohm is not in stock, as you told us.
Moreover, if you had realized this before building the amp, then I don't exclude that you would have opted for a protection circuit that did not raise the dissipation in R52 to such a high level (which means a different PCB layout)
Indeed, even 12 zeners, not only 6 as I said. Under these circumstances I understand your trade-off arguments.
Re: Re: Re: Re: What could be done better...
1. It depends on your definition of "ultimate performance". So far I've measured, YAP is very close close to the PGP amp, although PGP seems to be a better tradeoff. One area where PGP beats YAP hands down is the S/N ratio. YAP is not better than 110dB, while PGP was very close to 120dB. But YAP beats down PGP in stability: YAP is rock solid, while PGP is not stable without Zoebel and output coil. Etc...
2. That is correct at 5A, I was thinking about low current. 50mA at 5A makes theoretically about 10mA at 150mA (practically the dispersion is about double because of the subthreshold current dispersion contribution and a few other things), that is say 14% and this is assuming the errors are cumulating on the same side (is it +/-5mV or +/-10mV?). It may or may not be critical, but I certainly don't like it. Isn't it the few first watts that define the amp quality? When the drain current is 5A and the SPL out of my speakers reaches 110dB, I think I can forget about any 50mA mismatches anyway.
3. Don't wory, I got the right value on order 😀
4. That one I don't know. Wait until you'll build your amp then tell us what you would do different!
Edmond Stuart said:
1. It depends on your definition of "ultimate performance". So far I've measured, YAP is very close close to the PGP amp, although PGP seems to be a better tradeoff. One area where PGP beats YAP hands down is the S/N ratio. YAP is not better than 110dB, while PGP was very close to 120dB. But YAP beats down PGP in stability: YAP is rock solid, while PGP is not stable without Zoebel and output coil. Etc...
2. That is correct at 5A, I was thinking about low current. 50mA at 5A makes theoretically about 10mA at 150mA (practically the dispersion is about double because of the subthreshold current dispersion contribution and a few other things), that is say 14% and this is assuming the errors are cumulating on the same side (is it +/-5mV or +/-10mV?). It may or may not be critical, but I certainly don't like it. Isn't it the few first watts that define the amp quality? When the drain current is 5A and the SPL out of my speakers reaches 110dB, I think I can forget about any 50mA mismatches anyway.
3. Don't wory, I got the right value on order 😀
4. That one I don't know. Wait until you'll build your amp then tell us what you would do different!
traderbam said:
traderbam, does no matter much, in 100 years from now,
even less in the year of 3008, 1000 years from now.
I have mudded a lot of waters & minds, in my life, too
god forgive some of my words
because i was not my self ... my real good human self
Keep on posting your knowledge, traderbam
you have much to contrubute, as well as have syn08, Edmond and Bob
In this topic I am more one observer and one interested reader. Because I have not too much to contribute when comes to in depth discussion of correction algoritms.
But sure, I have other things I am better at .. no doubt 🙂
.. not necessarily related to audio in specific.
But maybe more into understanding social interaction between human individuals.
A few times, however, I can not understand some people & their actions, at all 😀
Through YOUR Self you can Know other Selfs.
And that we are so alike in many ways.
He who is without fault, may throw First Stone .. upon me, Lineup.
Lineup
Re: Re: Re: Re: Re: What could be done better...
1. All kinds of distortions, noise as well as THD and IMD. I thought that THD and IMD of PGP was also better than YAP.
As for stability, you tested YAP with 1uF, before the Zoebel network I assume. BUT .. what's the ESR etc. of that cap? Quite often an amp starts oscillating when loading with an ultra low ESR cap (multi layer ceramics) of say 47 nF. Did you tried that too (with short leads of course)?
3. You're a lucky boy. 🙂
4. I'm afraid I also have to tell things like that. Not only because I've a few things overlooked (despite all my simulations), but also my insights may have been changed in the meantime, based on your experiences, for example.
syn08 said:
1. All kinds of distortions, noise as well as THD and IMD. I thought that THD and IMD of PGP was also better than YAP.
As for stability, you tested YAP with 1uF, before the Zoebel network I assume. BUT .. what's the ESR etc. of that cap? Quite often an amp starts oscillating when loading with an ultra low ESR cap (multi layer ceramics) of say 47 nF. Did you tried that too (with short leads of course)?
3. You're a lucky boy. 🙂
4. I'm afraid I also have to tell things like that. Not only because I've a few things overlooked (despite all my simulations), but also my insights may have been changed in the meantime, based on your experiences, for example.
Re: Re: Re: Re: Re: Re: What could be done better...
At the same 150mA bias, PGP has THD20 1ppm, YAP has 2ppm. However, pushing the bias at 250mA brings YAP at 1ppm as well. IMD19+20 follows.
It's stable in 100nF metal stacked @200KHz and 10Vpp before coil and without Zoebel.
Edmond Stuart said:
At the same 150mA bias, PGP has THD20 1ppm, YAP has 2ppm. However, pushing the bias at 250mA brings YAP at 1ppm as well. IMD19+20 follows.
It's stable in 100nF metal stacked @200KHz and 10Vpp before coil and without Zoebel.
Bob Cordell said:
Not the same result - there are too many other variables involved.
If you start with (just for an example at one extreme) a class A RET output stage with 0.06% open loop THD the EC circuit required to get the THD down to .002% will have to be significantly more sophisticated (with less non linearity of its own) than that needed to attain 0.02% THD from a OPS with 0.6% open loop THD.
Cheers,
Glen
G.Kleinschmidt said:
Hi Glen,
I understand; you raise an interesting point. Putting HEC on a RET output stage that already does 0.06% open loop might be pushing diminishing returns. I have never built a BJT HEC amplifier and I'm not even sure I've simulated a BJT HEC amplifier. I've usually thought of HEC as primarily a blessing for MOSFETs, and thats where my simulation and implementation efforts have been.
However, permit me to speculate and think out loud here.
I think that the HEC might still do reasonably well in something not too far from its current form in the BJT situation you cite. While it is true that the linearity of the HEC differencing circuits themselves is a legitimate concern, I think it is also the case the the amplitudes they have to deal with (and consequent distortion they create) will be quite a bit lower in a case like the one you cite. I say this for two reasons. First, the larger amount of gate drive voltage required by MOSFETs as compared to BJTs is seen as an error that must be corrected, and this is a significant factor in the amplitudes that the differencing circuits must handle. Secondly, the smaller distortion of the BJT RETs to begin with would seem to further reduce the amplitudes that the differencing circuits would have to handle.
I think there is also another interesting question or opportunity in connection with BJT class A amplifiers. I could be wrong, but it seems to me that most class A push-pull amplifiers will go into class AB when they must drive a load to a level that demands more than twice their idle current. When this transition from class A to class AB occurs, I believe that they will suffer from gm-doubling distortion. The net gm of the output stage will be cut in half when the "other" side goes into cut-off. If this is the case, then HEC in such an output stage might significantly mitigate the gm-doubling distortion. What do you think?
I think it is also true that any speculation about how big a benefit HEC is for BJT applications needs to be tempered by the observation that the equivalent ft of the vertical MOSFETs may often be considerably larger than that for even RET BJT power devices, especially at high currents.
Cheers,
Bob
Bob Cordell said:
Hi Bob
I can agree with this thinking. I was playing with a version of HEC driving hexfets and I took notice of how much higher output in frequency vs magnitude the error amplifier has to produce with regard to crossover and cut-off. ( here) BJT's may have a more linear transconductance in that boundary around cut-off than the hexfets but there will still be lots of high order garbage introduced by any transistor related to cut-off, mosfet or BJT. IF the bandwidth of the HEC is large enough, I would think it would provide significant distortion reduction to the higher order frequency components generated by cut-off. If the BJT amp is always in class A, I doubt it would really be worth the effort, other than maybe giving a boost to the damping factor while not asking the VAS to work any harder. But with BJT's in class A, is that really asking that much of the VAS?
Ok guys, may I jump into this hot debate and post something about YAP?
I'm busy finishing the YAP measurements and here's the distortion final results.
1. 20KHz spectrum (full power 200W into 4ohm). THD20 is somewhere between 1 and 2ppm.
2. 10KHz spectrum (full power 200W into 4ohm). THD10 is again somewhere between 1 and 2ppm, slightly lower than at 20KHz.
3. IMD CCIF 19+20KHz. Again, it's between 1 and 2ppm. Fundamentals are attenuated by the Amber instrument (I got lately the IMD extension).
4. IMD SMPTE 60Hz+7KHz
5. The pulse response at 100KHz. As mentined before, the rise and fall times are defined exclusively by the input filter.
I'll be back with some details... Meantime, comments are welcomed!
I'm busy finishing the YAP measurements and here's the distortion final results.
1. 20KHz spectrum (full power 200W into 4ohm). THD20 is somewhere between 1 and 2ppm.
2. 10KHz spectrum (full power 200W into 4ohm). THD10 is again somewhere between 1 and 2ppm, slightly lower than at 20KHz.
3. IMD CCIF 19+20KHz. Again, it's between 1 and 2ppm. Fundamentals are attenuated by the Amber instrument (I got lately the IMD extension).
4. IMD SMPTE 60Hz+7KHz
5. The pulse response at 100KHz. As mentined before, the rise and fall times are defined exclusively by the input filter.
I'll be back with some details... Meantime, comments are welcomed!
Originally posted by lineup
[snip]
He who is without fault, may throw First Stone .. upon me, Lineup.
Originally posted by Edmond Stuart
Perhaps JC?
-------------------------------------------------------------------
My Comment:
I am not too afraid there are very many stones coming.
If they are rightenous stones ot not, does not matter much.
To me. One stone is always a stone.
I am not afriad, because there will be some real heavy stones thrown back at each agressor trying something with me.
***
There is one thing practiced at every court room, where mankind is trying to do justice between people.
That is, circumstance
The rule is:
nobody have the right to and should use MORE FORCE / VIOLENCE than was appropriate in the situation at hand.
And the second rule here is:
The right for any body or any country to Act In Self Defence
I can and have the right to defend my self & my actions.
This is called Self Defence.
( in some cases i would be a fool to defend my stupidness ... better to admit and regret .... 😉 )
I practice the noble kind of self defence, where I use words.
Without trying to humiliate my opponent.
Not too much.
Not more than this person DESERVES, in my opinion.
Let me stop here by quoting one olden goldie Jimmy Cliff Reggae song:
The Harder They Come ... THE HARDER THEY FALL .. one and all
syn08
Good 20 kHz squarewaves 😎
Do not think my amplifiers can beat those nice tests ...
PS.
You ought to do something about those BIG BIG images.
Make them fit 1024x768 screens! (max width ~ 900 pixels)
Thank you 🙂
Lineup
Considerably More WILD Forum .. run by lineup
[snip]
He who is without fault, may throw First Stone .. upon me, Lineup.
Originally posted by Edmond Stuart
Perhaps JC?
-------------------------------------------------------------------
My Comment:
I am not too afraid there are very many stones coming.
If they are rightenous stones ot not, does not matter much.
To me. One stone is always a stone.
I am not afriad, because there will be some real heavy stones thrown back at each agressor trying something with me.
***
There is one thing practiced at every court room, where mankind is trying to do justice between people.
That is, circumstance
The rule is:
nobody have the right to and should use MORE FORCE / VIOLENCE than was appropriate in the situation at hand.
And the second rule here is:
The right for any body or any country to Act In Self Defence
I can and have the right to defend my self & my actions.
This is called Self Defence.
( in some cases i would be a fool to defend my stupidness ... better to admit and regret .... 😉 )
I practice the noble kind of self defence, where I use words.
Without trying to humiliate my opponent.
Not too much.
Not more than this person DESERVES, in my opinion.
Let me stop here by quoting one olden goldie Jimmy Cliff Reggae song:
The Harder They Come ... THE HARDER THEY FALL .. one and all
syn08
Good 20 kHz squarewaves 😎
Do not think my amplifiers can beat those nice tests ...
PS.
You ought to do something about those BIG BIG images.
Make them fit 1024x768 screens! (max width ~ 900 pixels)
Thank you 🙂
Lineup
Considerably More WILD Forum .. run by lineup
Stupid?
It depends on someones motive. If he thinks that the 'sonic degradation' is proportional to the number of trannies or something of similar import, then that's nonsense of course. However, there exist other motives. If one can avoid an extra gain stage by optimizing the gain of other stages, for example, then the overall phase shift will probably be less. That means more headroom for NFB etc. etc. In this case not that stupid.
syn08 said:
It depends on someones motive. If he thinks that the 'sonic degradation' is proportional to the number of trannies or something of similar import, then that's nonsense of course. However, there exist other motives. If one can avoid an extra gain stage by optimizing the gain of other stages, for example, then the overall phase shift will probably be less. That means more headroom for NFB etc. etc. In this case not that stupid.
- Status
- Not open for further replies.