Rikard
if you look into this topic
you will find one error correction output circuit
http://www.diyaudio.com/forums/showthread.php?threadid=69782
Schematic:
http://www.diyaudio.com/forums/attachment.php?s=&postid=790790&stamp=1134638934
And one paper refered to is my PDF download link
Current dumping audio amplifier by P. J. Walker 1975
Link at lineupaudio (my PDF collection)
http://lineupaudio.freehostia.com/pdf/contents.php
(password= diyaudio)
if you look into this topic
you will find one error correction output circuit
http://www.diyaudio.com/forums/showthread.php?threadid=69782
Schematic:
http://www.diyaudio.com/forums/attachment.php?s=&postid=790790&stamp=1134638934
And one paper refered to is my PDF download link
Current dumping audio amplifier by P. J. Walker 1975
Link at lineupaudio (my PDF collection)
http://lineupaudio.freehostia.com/pdf/contents.php
(password= diyaudio)
Contrary to popular belief, nature is not symmetrical. 
Hi Folks, I just saw this thread. Lots of good conversation here, although many of the topics have been covered in the EC thread. Here are a few comments.
The Toshiba MOSFETs are definitely vertical MOSFETs, in the same vein as IRF HEXFETs, but with a lower Vgs.
The Toshiba parts are nice parts, but they appear to be quite expensive and less widely available than other vertical power MOSFETs. Looks to me like the Toshibas are in the $8-$12 each range. Please correct me if I am wrong.
The lower threshold voltage of the Toshiba parts is nice, as it helps preserve power supply headroom. However, my measurements indicate that in regard to the transconductance figure of merit, they are no better than, for example, IRFP240/9240. This FOM is simply the amount of transconductance you get for a given idle bias current, that current often being on the order of 150 mA per pair.
In regard to error correction, my circuit, which largely comes from Hawksford, is compatible with the Toshiba parts as long as some adjustments are made to give the EC circuit more headroom than it ordinarily would have with devices having lower Vgs. In many cases, just the addition of an additional tier of emitter followers ahead of the output devices will do. Other approaches using diodes and Vbe multipliers can also be used. In no case need one use a voltage divider that effectively reduces the transconductance of the output devices. The paper describing my EC circuit can be found on my web site at www.cordellaudio.com.
The discussion about whether Hawksford Error Correction (HEC) is just another form of negative feedback, or a legitimate form of error correction, or a form of variable negative feedback, has gone round and round on the EC thread. How it is viewed really doesn't matter. Some views provide different insights than others. The negative feedback view is a handy reminder of the need for attention to stability, as was mentioned in my original paper. If one wants to view it as a version of negative feedback wherein the forward gain is obtained by use of positive feedback, that is perfectly fine. The Devil is in the details and the proof is in the pudding.
As shown in my EC paper, the HEC circuit reduces output stage distortion by a factor of about 30 (30 dB) BEFORE the application of global negative feedback. In my EC amplifier, I employed a 2 MHz gain crossover frequency with a straight 6-dB/octave rolloff, achieving 40 dB of global negative feedback at 20 kHz. If one wants to argue that the output stage then has effectively 70 dB of NFB around it that may be OK. The important point is that it is stable and produces very high performance.
Some people don't like my version of the HEC circuit because it eats up VAS headroom, using a 22 volt fixed bias spreader to power the HEC circuits. At least that was the approach used in my original design. This inevitably leads to the need for boosted VAS power supplies, which adds cost and complexity. That is a fair criticism.
However, high-performance MOSFET power amplifiers often want to have a boosted supply anyway in order to make the most efficient use of the main power supply rails due to the inevitable need for significant forward drive voltage by the MOSFETs, especially when they are delivering high currents. This is slightly less the case for the Toshiba parts whose Vgs is about 2V less. So, the way I look at it, if you are really going for a high-performance MOSFET amplifier using verticals, you'll probably be considering the use of boosted rails for the VAS anyway. Needing to make those boosted rails higher by a few volts to accomodate HEC as opposed to no HEC seems not to be much of a big deal, but to each his own. Once you decide to pay the price for boosted rails, you will also reap other rewards because of the higher quality rails you can provide with better isolation from the dirty main rails.
At the same time, I will happily admit that I have designed vertical MOSFET power amplifiers that do not use boosted rails and do not use HEC. It really just depends on the immediate objectives.
I should also add that others on this board, like Edmond, have in recent years demonstrated very high performance through the use of what might be considered more conventional, yet local, negative feedback around the output stage. Indeed, what Edmond has called Transitional Miller Compensation (TMC) has a great deal of bang for the buck when it comes to distortion reduction versus added number of parts needed. If I am not mistaken, Baxandall first described that technique.
Cheers,
Bob
The Toshiba MOSFETs are definitely vertical MOSFETs, in the same vein as IRF HEXFETs, but with a lower Vgs.
The Toshiba parts are nice parts, but they appear to be quite expensive and less widely available than other vertical power MOSFETs. Looks to me like the Toshibas are in the $8-$12 each range. Please correct me if I am wrong.
The lower threshold voltage of the Toshiba parts is nice, as it helps preserve power supply headroom. However, my measurements indicate that in regard to the transconductance figure of merit, they are no better than, for example, IRFP240/9240. This FOM is simply the amount of transconductance you get for a given idle bias current, that current often being on the order of 150 mA per pair.
In regard to error correction, my circuit, which largely comes from Hawksford, is compatible with the Toshiba parts as long as some adjustments are made to give the EC circuit more headroom than it ordinarily would have with devices having lower Vgs. In many cases, just the addition of an additional tier of emitter followers ahead of the output devices will do. Other approaches using diodes and Vbe multipliers can also be used. In no case need one use a voltage divider that effectively reduces the transconductance of the output devices. The paper describing my EC circuit can be found on my web site at www.cordellaudio.com.
The discussion about whether Hawksford Error Correction (HEC) is just another form of negative feedback, or a legitimate form of error correction, or a form of variable negative feedback, has gone round and round on the EC thread. How it is viewed really doesn't matter. Some views provide different insights than others. The negative feedback view is a handy reminder of the need for attention to stability, as was mentioned in my original paper. If one wants to view it as a version of negative feedback wherein the forward gain is obtained by use of positive feedback, that is perfectly fine. The Devil is in the details and the proof is in the pudding.
As shown in my EC paper, the HEC circuit reduces output stage distortion by a factor of about 30 (30 dB) BEFORE the application of global negative feedback. In my EC amplifier, I employed a 2 MHz gain crossover frequency with a straight 6-dB/octave rolloff, achieving 40 dB of global negative feedback at 20 kHz. If one wants to argue that the output stage then has effectively 70 dB of NFB around it that may be OK. The important point is that it is stable and produces very high performance.
Some people don't like my version of the HEC circuit because it eats up VAS headroom, using a 22 volt fixed bias spreader to power the HEC circuits. At least that was the approach used in my original design. This inevitably leads to the need for boosted VAS power supplies, which adds cost and complexity. That is a fair criticism.
However, high-performance MOSFET power amplifiers often want to have a boosted supply anyway in order to make the most efficient use of the main power supply rails due to the inevitable need for significant forward drive voltage by the MOSFETs, especially when they are delivering high currents. This is slightly less the case for the Toshiba parts whose Vgs is about 2V less. So, the way I look at it, if you are really going for a high-performance MOSFET amplifier using verticals, you'll probably be considering the use of boosted rails for the VAS anyway. Needing to make those boosted rails higher by a few volts to accomodate HEC as opposed to no HEC seems not to be much of a big deal, but to each his own. Once you decide to pay the price for boosted rails, you will also reap other rewards because of the higher quality rails you can provide with better isolation from the dirty main rails.
At the same time, I will happily admit that I have designed vertical MOSFET power amplifiers that do not use boosted rails and do not use HEC. It really just depends on the immediate objectives.
I should also add that others on this board, like Edmond, have in recent years demonstrated very high performance through the use of what might be considered more conventional, yet local, negative feedback around the output stage. Indeed, what Edmond has called Transitional Miller Compensation (TMC) has a great deal of bang for the buck when it comes to distortion reduction versus added number of parts needed. If I am not mistaken, Baxandall first described that technique.
Cheers,
Bob
Hi Bob! Nice of you to pop in on! Really appreciate your comments, good to hear you think the Toshibas are ok. About your original EC circuit, I recently found it on your site, and started thinking about my "old" amp design that has been running my speakers for the last 8 or 10 years. I measured the distorsion and was quite pleased, but at the same time noticed that crossover distorsion was very idle dependant. Increasing the idle to the point where second and third harmonic became lower than the noise level was simply not an option with the heat sinks used.
Besides, a more "elegant" solution would be nice, so your EC seemed to (still seems) be the thing! I did some simulations, but as I don´t have the models for the Toshiba FET´s I didn´t get that close to reality. Today I had just enough time to asseble a bird nest which is mounted above the orignal amps board, and after a quick check it seems to be stable, the idle was easily adjusted and very solid. Tomorrow I hope to be able to test it further, and also determine the cause of the slight DC offset I noticed today, som 300-400 mV at the output. I could have messed something up though.
If you think adjusting the voltage be means of diodes would do the trick for the EC I would be happy to go through with it.
Don´t know if you have seen the zipped screenshot I uploaded earlier in this thread, but the input and predrivers are way simpler arranged than your original amp, but still seems to perform ok. Would be very nice to hear your commets about it. I have seen a few versions of this circuitry around, so nothing special really.
Also, what would be your opinion about biasing the FETs in a way that they never shut completely? In my world, this would be another great way of reducing crossover distorsion. Comparing this with your EC raises yet another question, the dampening factor (don´t know the correct english term). The EC improves this quite a lot as I understand it. "Dynamic biasing" would not, so it´s all up to the global NFB and supply to cope with it.
Once again, thank´s Bob for your comments!
Besides, a more "elegant" solution would be nice, so your EC seemed to (still seems) be the thing! I did some simulations, but as I don´t have the models for the Toshiba FET´s I didn´t get that close to reality. Today I had just enough time to asseble a bird nest which is mounted above the orignal amps board, and after a quick check it seems to be stable, the idle was easily adjusted and very solid. Tomorrow I hope to be able to test it further, and also determine the cause of the slight DC offset I noticed today, som 300-400 mV at the output. I could have messed something up though.
If you think adjusting the voltage be means of diodes would do the trick for the EC I would be happy to go through with it.
Don´t know if you have seen the zipped screenshot I uploaded earlier in this thread, but the input and predrivers are way simpler arranged than your original amp, but still seems to perform ok. Would be very nice to hear your commets about it. I have seen a few versions of this circuitry around, so nothing special really.
Also, what would be your opinion about biasing the FETs in a way that they never shut completely? In my world, this would be another great way of reducing crossover distorsion. Comparing this with your EC raises yet another question, the dampening factor (don´t know the correct english term). The EC improves this quite a lot as I understand it. "Dynamic biasing" would not, so it´s all up to the global NFB and supply to cope with it.
Once again, thank´s Bob for your comments!
Bob, a question about your EC: How about reducing the 22 volts to say 16 volts as the Toshiba FETs use less gate voltage? It would reduce heat dissipation from the transistors, and save some headroom.
Also, I would like to reduce the current through all of the EC transistors in order to use ordinary ones like 2N5401/5551 except for the last driver pair that might be a bit tougher. I know this will slow down the circuit, but as I understand it the Toshiba FETs need about 17 mA to reach 100 V/us.
Rgs Rikard
Also, I would like to reduce the current through all of the EC transistors in order to use ordinary ones like 2N5401/5551 except for the last driver pair that might be a bit tougher. I know this will slow down the circuit, but as I understand it the Toshiba FETs need about 17 mA to reach 100 V/us.
Rgs Rikard
Bob Cordell said:
As an engineer, I say that how a circuit is viewed matters very much indeed. Legitimizing conjecture that is unproven, or disproven, leads to bad design decisions and misinformation.
The audio world is rife with tantalizing, bad conjectures, and we are all vulnerable to them, and one value I stand for in the context of diyAudio is to help members to distinguish that which stands up to scrutiny from that which is bogus. For this is what leads to good design and good sound.
For this reason I am compelled to repeat that there is no engineering proof that Bob's circuit incorporates anything other than feedback to reduce output errors. The conjectures that "feedback on demand" and "error cancellation" are involved, have been disproved by mathematical analysis and by simulation, and this is all published in the EC thread and elsewhere. Nor does Hawksford make such claims.
I recognize the conflict of interest that Bob faces, but I will continue to urge him to be conscientious in his writings to avoid legitimizing mystique.
This is nothing we normal diyaudio guys
need to get too much involved in.
Let's build good amplifiers (sometimes with help of highly educated men )
and let us leave the academic discussions to the ones interested in splittin' hairs.
This topic is about best use of 2SK1530 & 2SJ201
no matter if they are vertical, lateral, horisontal or diagonal MOSFETs
We will see what way Rikard Nilsson will arrange his global feedback.
If he wants to use global FB .....
He is experienced enough to decide this for him self,
if I am not mistaken .. which can happen some time
Lineup
need to get too much involved in.
Let's build good amplifiers (sometimes with help of highly educated men
)and let us leave the academic discussions to the ones interested in splittin' hairs.
This topic is about best use of 2SK1530 & 2SJ201
no matter if they are vertical, lateral, horisontal or diagonal MOSFETs
We will see what way Rikard Nilsson will arrange his global feedback.
If he wants to use global FB .....
He is experienced enough to decide this for him self,
if I am not mistaken .. which can happen some time
Lineup
traderbam said:
As an engineer, I say that how a circuit is viewed matters very much indeed. Legitimizing conjecture that is unproven, or disproven, leads to bad design decisions and misinformation.
The audio world is rife with tantalizing, bad conjectures, and we are all vulnerable to them, and one value I stand for in the context of diyAudio is to help members to distinguish that which stands up to scrutiny from that which is bogus. For this is what leads to good design and good sound.
For this reason I am compelled to repeat that there is no engineering proof that Bob's circuit incorporates anything other than feedback to reduce output errors. The conjectures that "feedback on demand" and "error cancellation" are involved, have been disproved by mathematical analysis and by simulation, and this is all published in the EC thread and elsewhere. Nor does Hawksford make such claims.
I recognize the conflict of interest that Bob faces, but I will continue to urge him to be conscientious in his writings to avoid legitimizing mystique.
It is unfortunate that you cannot distinguish between "bogus", "mystique", and legitimate differences in viewing an engineering problem. You are certainly right that there is a lot of unfortunate misleading and downright fraudulent stuff running around in audio, but to lump the various ways of looking at HEC in with that is completely wrong. I think you just have a very bad case of NIH. Remember, it it not what we call it, but rather the end result that matters. In this case the end result was rather outstanding 25 years ago, even by today's standards. It is sad that as an engineer you cannot bring yourself to give credit where credit is due. For you to insinuate that I am being driven by some kind of conflict of interest and not being conscientious in my patient explanations is truly absurd and insulting.
Cheers,
Bob
it seems the best method of operating MOSFETs (or bipolars) in the most linear part of their curve is to parallel them and bias them for somewhere between 100-200mA per device. this also tends to mask the normal device-to-device variations in characteristics.
there were amps made back in the '70s and '80s that used a higher rail voltage for the drivers than was used for the outputs, as well as a "triple darlington" output stage. these amps were nothing to sneeze at, as their distortion at full rated power was around 0.02% or better. there's nothing wrong with having a higher rail voltage for the driver stage (except of course you need extra windings on the power transformer and another low current bridge rectifier) to maximize the use of MOSFET outputs. if one decides to use a "triple darlington" output stage, be aware that the miller capacitance is multiplied by the total beta of the darlingtons, and compensation is going to really have to slow down your input/VAS stages to stabilize it, even with MOSFET outputs.
there were amps made back in the '70s and '80s that used a higher rail voltage for the drivers than was used for the outputs, as well as a "triple darlington" output stage. these amps were nothing to sneeze at, as their distortion at full rated power was around 0.02% or better. there's nothing wrong with having a higher rail voltage for the driver stage (except of course you need extra windings on the power transformer and another low current bridge rectifier) to maximize the use of MOSFET outputs. if one decides to use a "triple darlington" output stage, be aware that the miller capacitance is multiplied by the total beta of the darlingtons, and compensation is going to really have to slow down your input/VAS stages to stabilize it, even with MOSFET outputs.
unclejed613 said:
Hi unclejed613,
You are right about using paralleled devices. With BJTs, it especially helps reduce the effects of beta droop and ft droop at high currents. With MOSFETs, 150 mA per device is a very good region to operate in, and paralleling them increases total transconductance of the output stage, mitigating the effects of transconductance droop in the crossover region. Of course, paralleing multiple MOSFET pairs each operating at 150 mA increases idle bias current and dissipation, so there is a price to be paid. Overall, however, one receives the benefit of a larger class A operating region.
With regard to the use of output Triples like the Locanthi T circuit, it is not quite right to suggest that their use requires a change in the value of the Miller compensation capacitance or that the input stage must be slowed down. It is true, however, that the use of the Triple lightens the load on the VAS and increases VAS voltage gain at low frequencies. This does result in a lower open-loop bandwidth, but does not change the amount of open-loop gain at high frequencies. In any case, the value of the Miller compensation capacitor needed to obtain a given gain crossover frequency remains essentially the same when going from an out put Double to and output Triple. Similarly, the introduction of a Triple does not reduce slew rate, all else remaining the same.
Cheers,
Bob
I tried measuring the prototype today, and ended up with slightly worse distorsion figures than before I implemented the EC network. Second harmonic was at -77 dB at 10 watts into 4 ohms resistive load and 150 mA idle.
Increasing the idle radically reduced the distorsion, I had no problems reaching -85 dB at about 300 mA idle. But the thing is, the amp was slightly better before! So I assume the EC network may still be "operating" outside its good range.
I measured 3,3 volts between the gates, and the zener at the input is now 16 volts instead of 22 volts.
I put 500 ohm trimmers in place of R34 and R35 in my schematic to be able to adjust the feedback. Turning these trimmers didn´t affect the distorsion at all, but the idle slightly. So something is still wrong, but I can´t figure out what until I have some time to simulate the thing again.
/Rikard
Increasing the idle radically reduced the distorsion, I had no problems reaching -85 dB at about 300 mA idle. But the thing is, the amp was slightly better before! So I assume the EC network may still be "operating" outside its good range.
I measured 3,3 volts between the gates, and the zener at the input is now 16 volts instead of 22 volts.
I put 500 ohm trimmers in place of R34 and R35 in my schematic to be able to adjust the feedback. Turning these trimmers didn´t affect the distorsion at all, but the idle slightly. So something is still wrong, but I can´t figure out what until I have some time to simulate the thing again.
/Rikard
I ran this in the sim, it works perfectly! Note, this is of course only the output. I can adjust the values of R34 and R35 for minimum distorsion, and the values are quite critical. 20 ohms in either direction immediately shows up as distorsion. I managed to get a simulated distorsion as low as 0.003% !So the EC network should be in the correct operating area with these values. Only thing is I don´t know how close the FETs are to the 2SJ201/2SK1530 ones.I´ll tweak the values in my prototype and try again. I do have another one so I could build the EC once again if I did mess something up now.
Attachments
And here is the complete amp with VAS. I use boosted rails in the prototype, filtered through "gyrators". Works ok. In the sim I added 2 volts of 50 Hz AC at the final rails to simulate ripple. Seems to be unaffected by this amount. Had to zip the picture since it is too wide to upload.
I tried cascoding the predrivers Q1 and Q2, but could not get rid of oscillating.
Lets here what you think guys!
I tried cascoding the predrivers Q1 and Q2, but could not get rid of oscillating.
Lets here what you think guys!
Complete amplifier with VAS
... except for the input section .. it is complete
Made my head go crazy for a while
http://www.diyaudio.com/forums/attachment.php?s=&postid=1584621&stamp=1218649902
I am sure you have input section in some ZIP ....
Rikard,
Will be a great amplifier, no doubt .. as you know what you are doing.
Very well you know.
Thanks for attaching your circuits & ZIP files
I have downloaded them all!
Will have closer look, when have time
Olympic Days, you know
Lineup - got 2 olympic silver medals, by now
... except for the input section .. it is complete
Made my head go crazy for a while
http://www.diyaudio.com/forums/attachment.php?s=&postid=1584621&stamp=1218649902
I am sure you have input section in some ZIP ....
Rikard,
Will be a great amplifier, no doubt .. as you know what you are doing.
Very well you know.
Thanks for attaching your circuits & ZIP files
I have downloaded them all!
Will have closer look, when have time
Olympic Days, you know
Lineup - got 2 olympic silver medals, by now
Thanks, maybe someone can solve the problem with cascoding Q1 and Q2. I tried adding a transistor to each of them connecting the base to the base of Q11 and Q12 resp since these nodes have a constant voltage in respect to the rails. The bandwidth increases a lot, but after a few cycles it gets unstable. Didn´t help adding freq compensation at the FET gates. Don´t think that is where is oscillates. Tried a small feedback cap from the cascode output back to the inverting input of the diff amp as well, but still oscillates. Removing the extra transistors makes everything stable, and as it seems to perform well maybe I will just leave it this way.
If somebody have good EWB models for the 2SJ201/2SC1530 I would be thrilled!
Just have to get the thing working in practice too.....
If somebody have good EWB models for the 2SJ201/2SC1530 I would be thrilled!
Just have to get the thing working in practice too.....
Rikard Nilsson said:
If somebody have good EWB models for the 2SJ201/2SC1530 I would be thrilled!
Just have to get the thing working in practice too.....
see this post
and also the post BEFORE & AFTER in the Spice Simulation thread:
http://www.diyaudio.com/forums/showthread.php?postid=1255538#post1255538
Bob Cordell was also annoyed, after contacting Hitachi to get those models
Rikard,
frankly, I am having considerable difficulties finding anything I like about this design. The correction occurs over many stages and unlinear diodes strongly reducing its accuracy, on the whole probably doing more damage than good. Numerous other topological annoyances and massive negative feedback effects will keep the subjective qualities at a less than mediocre level.
frankly, I am having considerable difficulties finding anything I like about this design. The correction occurs over many stages and unlinear diodes strongly reducing its accuracy, on the whole probably doing more damage than good. Numerous other topological annoyances and massive negative feedback effects will keep the subjective qualities at a less than mediocre level.
i kind of ran out of time this morning as i had to leave for work, but i was going to mention that with triples, EF is the type that has high Miller capacitance, but CFP puts the Miller (of the output devices) capacitance in series with the load, rather than in parallel as the EF does. with the output device capacitance in series with the load, rather than in parallel, the amp will be easier to stabilize with capacitive loads.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.