Nico Ras said:
This was in the context of running multiple parallel complementary pairs, perhaps 10 pairs each with 0.22 ohm RE. So the current flow is not all concentrated in one area of a small piece of track. Of course, one does have to be aware of track resistances and track widths in any of these arrangements, since total currents in the amperes or tens of amperes may flow under some signal or fault conditions.
Cheers,
Bob
Nico, output FET's do NOT follow the same guideline as bipolar transistors.
Also, changing the quiescent current WILL change the order of the harmonic series, but it takes special measurement to easily see it. I have a low bias-high bias switch on my Parasound JC-1, and I can measure more 5th harmonic at a few watts out compared to high bias.
Also, changing the quiescent current WILL change the order of the harmonic series, but it takes special measurement to easily see it. I have a low bias-high bias switch on my Parasound JC-1, and I can measure more 5th harmonic at a few watts out compared to high bias.
Mr Curl,
the Re's of the JC-1 look like 0.15 ohm/5% types.
88mA per device in the low bias setting corresponds to a voltage drop across Re of a little over 13mV.
The Re value seems less critical with 9 output devices in parallel, is there a trade-off ?
Your last post doesn't seem to justify a low bias switch.
(ps: i take it that the opa134 is an upgrade compared to the original AD7eleven)
the Re's of the JC-1 look like 0.15 ohm/5% types.
88mA per device in the low bias setting corresponds to a voltage drop across Re of a little over 13mV.
The Re value seems less critical with 9 output devices in parallel, is there a trade-off ?
Your last post doesn't seem to justify a low bias switch.
(ps: i take it that the opa134 is an upgrade compared to the original AD7eleven)
lumanauw said:
I'm not sure if there is a magic number for MOSFETs. First of all, many MOSFET designs do not use an RE resistor at all, even when devices are placed in parallel. Secondly, as has been discussed earlier, MOSFETs want to run at a higher bias current so as to get the gm up high enough to reduce crossover distortion. Remember, static crossover distortion is essentially the result of the output impedance of an emitter follower or a source follower changing as a function of operating current into the load (this, against the load impedance, changes the incremental gain as a function of signal current level, and this is what causes distortion). The smaller this varying impedance is in the first place, and the smaller the percentage change, the less distortion.
The output impedance of each half of the MOSFET stage is about equal to 1/gm of the device. GM of a MOSFET is on the order of 1/10 that of a bipolar at a given current, so rs = 1/gm is on the order (very roughly) of 250/Id in mA. What this means is that it is unlikely that the bias current in the MOSFET will be set high enough to get into gm doubling, given that you still don't want to use too high a value of RE. If, for example, you used an RE of 0.33 ohm for a single MOSFET, and you ran the MOSFET at 200 mA, you would see about 66 mV across RE, but you would still be well under the point where so-called gm doubling might be coming in.
I guess what I'm not doing a very good job of saying here is that for a MOSFET, in very rough terms, the more idle current the better. If an "optimum" exists, it is probably at a value of current that is higher than you would want to use.
There is, I believe, a mitigating factor in the static crossover distortion situation with MOSFETs. Because the transconductance is lower for a given current, the change in transconductance is a bit less sudden than with a bipolar, so that the region of overlap in handoff from one sex device to the other is a little less abrupt. This can lead to the order of the harmonics being a bit lower.
Cheers,
Bob
Yes, there is NO magic number for fets. (That should confuse things even more. 
)
It is important to note that the optimum number for complementary bipolar transistors is only a guideline. If you are off by a little, it may not make that much difference. It is, in fact, impossible to have an optimum resistance with both a high and a low current setting, such as the JC-1 power amp, but the measured difference is rather small, because we are still close, in either setting.
Also, the OPA134 appears to be an excellent servo IC, and we can buy them today. 711's became difficult to buy, for some reason.
)It is important to note that the optimum number for complementary bipolar transistors is only a guideline. If you are off by a little, it may not make that much difference. It is, in fact, impossible to have an optimum resistance with both a high and a low current setting, such as the JC-1 power amp, but the measured difference is rather small, because we are still close, in either setting.
Also, the OPA134 appears to be an excellent servo IC, and we can buy them today. 711's became difficult to buy, for some reason.
Hi, Mr. Cordell, Mr. Curl,
Thanks for the explenation
I'm interested in this :
If it is hard to get "gm doubling", is it means that mosfet in not so biased mode, is actually in "gm lacking" around the area of crossover from N to P / P to N?
I'm not an expert about this, but the best thing would be a continious same gm, not "doubling" and not "lacking", wouldn't it?
Mr. Cordell,
Back again to Cordell's EC. Is that your EC mosfet making relatively smooth gm of output stage for audio usage (eliminating gm doubling or gm lacking without the need of classA mode)?
Thanks for the explenation
I'm interested in this :
If it is hard to get "gm doubling", is it means that mosfet in not so biased mode, is actually in "gm lacking" around the area of crossover from N to P / P to N?
I'm not an expert about this, but the best thing would be a continious same gm, not "doubling" and not "lacking", wouldn't it?
Mr. Cordell,
Back again to Cordell's EC. Is that your EC mosfet making relatively smooth gm of output stage for audio usage (eliminating gm doubling or gm lacking without the need of classA mode)?
lumanauw said:Hi, Mr. Cordell, Mr. Curl,
Thanks for the explenation
I'm interested in this :
If it is hard to get "gm doubling", is it means that mosfet in not so biased mode, is actually in "gm lacking" around the area of crossover from N to P / P to N?
I'm not an expert about this, but the best thing would be a continious same gm, not "doubling" and not "lacking", wouldn't it?
Mr. Cordell,
Back again to Cordell's EC. Is that your EC mosfet making relatively smooth gm of output stage for audio usage (eliminating gm doubling or gm lacking without the need of classA mode)?
Yes, you are exactly right. In most cases, the MOSFET can be said to be operating in the "gm-lacking" range in the crossover region. This is what I referred to as transconductance droop in my original MOSFET power amp with EC paper (available on my web site at www.cordellaudio.com).
This transconductance droop is really what Self is referring to when he states that MOSFETs are more nonlinear than BJTs in an output stage. They just have a lot less transconductance than BJTs at a given operating current.
Indeed, the single biggest reason that I applied the error correction circuit to the MOSFET power amp was to greatly reduce the crossover distortion due to transconductance droop while being able to bias the output stage at a reasonable current level.
The cost of the error correction circuit is very small, perhaps an additional 4 or 6 small-signal transistors IF you already are using boosted supplies for the driver circuits.
This error correction technique can also be applied to BJT output stages to virtually eliminate the effects of gm doubling. That would allow you to operate more freely in the region of gm doubling, allowing you to operate in Class AAB with higher values of RE for better thermal stability if you want.
Cheers,
Bob
Error Correction on BJT
Hello BoB
Would error work on a BJT ouput stage. I would of thought that if you applied it like you did with Mosfet stage you would have stability problems . Can you reduce crossover distortion with this error corrction method on bipolars.
My impression is that lowest distortion amp is built using bipolars all the way up to the output where you would use Mosfets with error correction, assuming you wanted a practical amp that did not need class A bias.
Regards
Arthur
Hello BoB
Would error work on a BJT ouput stage. I would of thought that if you applied it like you did with Mosfet stage you would have stability problems . Can you reduce crossover distortion with this error corrction method on bipolars.
My impression is that lowest distortion amp is built using bipolars all the way up to the output where you would use Mosfets with error correction, assuming you wanted a practical amp that did not need class A bias.
Regards
Arthur
Re: Error Correction on BJT
Yes, error correction will work with a bipolar output stage, although I have not personally tried it with BJTs. Actually, Hawksford's original description of the technique assumed bipolar output transistors. Yes, one would want to use BJT's with good ft, and the stability compensation would have to be appropriate for the BJTs used, but it should not have a stability problem if it is done right. Also, as a minor circuit detail, some voltage drop needs to be placed in the drive chain in front of the BJTs, since their lower turn-on voltage than the MOSFETs would otherwise not provide enough working voltage headroom for the error correction circuits. We're only talking about adding a drop of maybe 2-3V, however.
You're impression of how to build the lowest distortion amplifier is largely correct as far as if it was me doing it, except that I would still tend to use JFETs in the input differential pair. However, that would just be my personal preference. But that is certainly not the only way to do it, as has been pointed out quite well on this thread.
Cheers,
Bob
PHEONIX said:
Yes, error correction will work with a bipolar output stage, although I have not personally tried it with BJTs. Actually, Hawksford's original description of the technique assumed bipolar output transistors. Yes, one would want to use BJT's with good ft, and the stability compensation would have to be appropriate for the BJTs used, but it should not have a stability problem if it is done right. Also, as a minor circuit detail, some voltage drop needs to be placed in the drive chain in front of the BJTs, since their lower turn-on voltage than the MOSFETs would otherwise not provide enough working voltage headroom for the error correction circuits. We're only talking about adding a drop of maybe 2-3V, however.
You're impression of how to build the lowest distortion amplifier is largely correct as far as if it was me doing it, except that I would still tend to use JFETs in the input differential pair. However, that would just be my personal preference. But that is certainly not the only way to do it, as has been pointed out quite well on this thread.
Cheers,
Bob
Re: Re: Error Correction on BJT
Hello Bob
I have never seen a super low distortion amp using error correction on the output with bipolar output stage. I know of the Halcro this uses fets but not an amp with bipolars.
Do you think you could go as low distortion as the Fet amp 1ppm THD at 20Khz. How low is the lowest THD you a measured it must be close to this.
Regards
Arthur
Bob Cordell said:
Hello Bob
I have never seen a super low distortion amp using error correction on the output with bipolar output stage. I know of the Halcro this uses fets but not an amp with bipolars.
Do you think you could go as low distortion as the Fet amp 1ppm THD at 20Khz. How low is the lowest THD you a measured it must be close to this.
Regards
Arthur
suzyj said:Just read on the Cordell Audio site about the RMAF level matched valve/transistor comparison.
I like! That's very cool - keep slaughtering those sacred cows
Cheers,
Suzy
Thanks, Suzy. There is an interesting history to this. I didn't start out trying to prove anything or slaughter any sacred cows, but rather just out of curiosity (although my curiosity often gets me in trouble that way). Back in high school in the Sixties I had built numerous tube amps, and one of them survived to this day due to my unfortunate reluctance to throw anything out. Anyway, about this time last year I wanted to do a decent one-on-one comparison of a tube amp with a solid state amp so I could honestly hear for myself the dramatic difference I "knew" existed from all that I have read. So I resurrected that old tube amp. It was a 35 wpc design based on 7591A output tubes.
The amp wasn't anything special at the time, since when I designed it I was just a high school kid. It was your typical push-pull Class AB "classic" design. All the electrolytics were shot, of course, and many of the components were not of really great quality. The power transformer had actually been liberated from an old black-and-white TV (but was still good!).
So I pretty much gutted it and started all over again. I decided to obtain new output tubes, and settled on my old friend the KT-88. This tube, of course, is just loafing at 35 watts. I also decided to use a more modern technology in some areas. For example, I used MOSFET voltage regulators for the screen supplies (it was a pentode design, not an Ultra-linear design), and used transistor current sources for the vacuum tube differential pair tails. No solid state stuff in the signal path, although I suppose some purists would still argue that I "lost" the tube sound by putting anything solid state inside the chassis.
Anyway, I finished building it and measured it, and it measured quite decently for a tube design, with "Classic" characteristics. About 20 dB of feedback and a damping factor of about 20. I hooked it up and it sounded pretty good. I then AB'd it against a Hafler P230 MOSFET amp, and was shocked that the audible differences were quite subtle when I level-matched them properly (within 0.1 dB at 1 kHz as read at the speaker terminals as loaded by the speaker being used for audition). I then listened to it against a Crest bipolar 200 wpc pro amplifier. Again, the differences were remarkably subtle. In both cases I had really been expecting to hear night and day differences. Maybe I just have tin ears? Maybe I had inadvertantly given the tube amp a solid state sound? Go figure. Differences could be heard, but they were subtle and it was not at all clear which sounded "better".
I then took the amplifier to a "bottlehead" meet. I wanted to hear it against other tube amps, and also hear what the tube guys though of its sound. I said nothing of my own listening comparisons with solid state designs. The guys there immediately loved the look and glow of those big-@ss KT88's, before they even listened to it. Bottom line, they all liked the sound of it, and though it was one of the best sounding tube amps there. So I guess it was a legitimate tube-sound amp. Certified by bottlehead tube gurus. It kicked the cr@p out of a 300B SET, but that's another story.
Anyway, that's why we decided to make it one of our RMAF workshops. We not only wanted to share our experience with others but, perhaps more importantly, have even more high-end afficiandos listen to the comparison and see how much difference they could hear. To this day, I remain really, really surprized.
Bob
The problem with these sorts of comparisons is that they
are not usually optimized to throw audible differences into
sharp relief. It is important to have lived with a reference
system for a period of time and then toss in a variable.
I have found that, as with wine, the differences are most
apparent when you step down in quality from a familiar
reference.
are not usually optimized to throw audible differences into
sharp relief. It is important to have lived with a reference
system for a period of time and then toss in a variable.
I have found that, as with wine, the differences are most
apparent when you step down in quality from a familiar
reference.
Nelson Pass said:The problem with these sorts of comparisons is that they
are not usually optimized to throw audible differences into
sharp relief. It is important to have lived with a reference
system for a period of time and then toss in a variable.
I have found that, as with wine, the differences are most
apparent when you step down in quality from a familiar
reference.
I agree, Nelson. And I certainly agree if one is comparing nominally similar components. But I was set up to expect night-day differences between a 35 wpc tube amp and a 200 wpc solid state amp.
Bob
Re: Re: Re: Error Correction on BJT
There are not many EC amplifiers out there that I know of, anyway, but I am not specifically aware of a commercial EC amp with bipolar outputs. Yes, the Halcro uses EC with HEXFETs, as far as I know. If you look at Candy's patent, it looks a lot like the output stage I designed about 10+ years prior. Maybe he was unaware of the JAES publications on EC by Hawksford and myself, because he cited neither as prior art in that patent. Curiously, in a later patent of his where he was promoting a different kind of error reduction scheme (gobs of high-order NFB loops), he did cite the prior art of Hawksford and me. This later patent gives rise to the question: Is Halcro still using the original Hawksford-type error correction scheme in their amplifiers, or are they instead using the scheme referenced in that later patent (gobs of NFB).
1 ppm THD20 is 0.0001%, or -120 dB. I got down to 0.0006% in my original design. I've never gotten better than that, but I can't say as I've tried hard. I'm looking at some circuits now that will probably do better. It is notable that Halcro claims 0.00005% out to 20 kHz, if I've got the numbers right. That's 500 parts per Billion. So, if you believe them, 1 ppm can be done. The only problem is, I've never seen a review where the Halcro came even close to their advertized spec, but that always raises the question about the ability of the instrumentation. The latest DM38 Follow-Up test in Stereophile is pretty thorough and still appears to show credible evidence of Halcro distortion well above their spec, even at 1 kHz.
Bob
PHEONIX said:
There are not many EC amplifiers out there that I know of, anyway, but I am not specifically aware of a commercial EC amp with bipolar outputs. Yes, the Halcro uses EC with HEXFETs, as far as I know. If you look at Candy's patent, it looks a lot like the output stage I designed about 10+ years prior. Maybe he was unaware of the JAES publications on EC by Hawksford and myself, because he cited neither as prior art in that patent. Curiously, in a later patent of his where he was promoting a different kind of error reduction scheme (gobs of high-order NFB loops), he did cite the prior art of Hawksford and me. This later patent gives rise to the question: Is Halcro still using the original Hawksford-type error correction scheme in their amplifiers, or are they instead using the scheme referenced in that later patent (gobs of NFB).
1 ppm THD20 is 0.0001%, or -120 dB. I got down to 0.0006% in my original design. I've never gotten better than that, but I can't say as I've tried hard. I'm looking at some circuits now that will probably do better. It is notable that Halcro claims 0.00005% out to 20 kHz, if I've got the numbers right. That's 500 parts per Billion. So, if you believe them, 1 ppm can be done. The only problem is, I've never seen a review where the Halcro came even close to their advertized spec, but that always raises the question about the ability of the instrumentation. The latest DM38 Follow-Up test in Stereophile is pretty thorough and still appears to show credible evidence of Halcro distortion well above their spec, even at 1 kHz.
Bob
Hey Nelson, Bob thinks that LOW distortion is what is all important. We know better, don't we?
Of course, one should match their levels. In 1969, I did a comparison between my reference solid state complementary diff power amp and a noted triode tube design on a single K-horn. We used a differential subtraction method to match the outputs. Guess what? The triode won! Then, I measured the differences between my design and the triode:
At one Watt, virtually unmeasureable IM distortion on both, smoothly rising to .1% at 10W. For a K-horn, this is enough. Damping factor was about 30 in both units. Order of distortion was primarily 3'rd in both units. Frequency response was at least 100KHz in both units. Why then was there an audible difference to experienced ears?
I came to the conclusion that I had to use 40dB of negative feedback and the triode used 20dB. Thus, I hypothesized that negative feedback might have something to do with it. I still think this is true today.
We know better, don't we? Of course, one should match their levels. In 1969, I did a comparison between my reference solid state complementary diff power amp and a noted triode tube design on a single K-horn. We used a differential subtraction method to match the outputs. Guess what? The triode won! Then, I measured the differences between my design and the triode:
At one Watt, virtually unmeasureable IM distortion on both, smoothly rising to .1% at 10W. For a K-horn, this is enough. Damping factor was about 30 in both units. Order of distortion was primarily 3'rd in both units. Frequency response was at least 100KHz in both units. Why then was there an audible difference to experienced ears?
I came to the conclusion that I had to use 40dB of negative feedback and the triode used 20dB. Thus, I hypothesized that negative feedback might have something to do with it. I still think this is true today.
John, don't be so presumptuous in characterizing what others believe. If some body asks me about distortion, I'll answer their question. If somebody expresses interest in how low we can go, I'll offer my opinion. But I don't believe that is all there is to it. Why else would I be advocating the use of soft clipping, for example, which INTRODUCES distortion in a controlled way.
Cheers,
Bob
Cheers,
Bob
Re: Re: Re: Re: Error Correction on BJT
Hello Bob
Interesting point you make about his distortion specs , I once quizzed him in a forum about how he measured better than 1ppm at 20Khz, he did not want to talk about it, and it was implied(not by Bruce) they had a special setup, of which there is not a single reference anywhere on their website.
Regards
Arthur
Bob Cordell said:
Hello Bob
Interesting point you make about his distortion specs , I once quizzed him in a forum about how he measured better than 1ppm at 20Khz, he did not want to talk about it, and it was implied(not by Bruce) they had a special setup, of which there is not a single reference anywhere on their website.
Regards
Arthur
john curl said:Hey Nelson, Bob thinks that LOW distortion is what is all important.
Of course, one should match their levels. In 1969, I did a comparison between my reference solid state complementary diff power amp and a noted triode tube design on a single K-horn. We used a differential subtraction method to match the outputs. Guess what? The triode won! Then, I measured the differences between my design and the triode:
At one Watt, virtually unmeasureable IM distortion on both, smoothly rising to .1% at 10W. For a K-horn, this is enough. Damping factor was about 30 in both units. Order of distortion was primarily 3'rd in both units. Frequency response was at least 100KHz in both units. Why then was there an audible difference to experienced ears?
I came to the conclusion that I had to use 40dB of negative feedback and the triode used 20dB. Thus, I hypothesized that negative feedback might have something to do with it. I still think this is true today.
Dear John, Bob and Nelson,
This thread has intrigued me so much if you recall I presented an A/B test comparison between two amplifiers both utilising identical complimentary symmetry except one amplifier used 2SK1058/2SJ162 mosfets and the other MJ15003/MJ15004 BJTs.
As mentioned before, both amps had low THD, bias currents were set to 90 mA and used RE=0.22 Ohm, both used identical amount of NFB, all transistors were matched to single digit variances, all resistors were 1% metal film and there were no capacitors in the signal path.
During the initial trials, four musicians found subtle differences variations in character between percussion and strings but no real preference between amps. This result was not good enough for me.
During the holidays, we had 83 visitors at some time or other passing through and I made a point that everyone listened and gave a verdict of which sounded "better".
The ages ranged between 18 and 76 years, none even remotely connected to the audio industry, just normal people. Some did not even like listening to music at all.
I chose a good quality recording that I thought no-one would be familiar with and played 60 seconds of the same track one each amp.
The verdict was that 77% could not tell any difference at all. 8% thought that they preferred the BJT while the remainder thought they preferred the MOSFET.
I have also exhausted my own ears listening to the two amps and the more I listen to them the more they sound the same and even the initial differences in character of instruments detected previously became an uncertainty.
Prior to this experience, I was a staunch follower of the BJT regime and regarded MOSFET amplifiers as inferior. I have never thought that I could be so wrong.
I think different sounding amplifiers are different simply because of the different design approaches.
The fact is, an amplifier sounds like its designer.
Kind regards
Nico