Bob Cordell Interview: BJT vs. MOSFET

[poynton]

Originally posted by Upupa Epops
Best of the both worlds give cascode connection of Mosfet + BJT, as was shown in TAA by authors Pass/ Hagard....

Got a copy for posting ??

Andy

[Upupa Epops]

I'm sorry Andy, I don't have it, but some guy will have...

[GK]

Originally posted by john curl
Glen, the MJL21193 series devices are interesting for certain applications. For example, if I wanted to make a really powerful woofer amplifier for a PA or even a high end audio system. Motor drives would be ideal for this device.
It isn't 'perfect' for high frequency reproduction, but I am pretty sure from past experience that I could make an acceptable power amp with a slew rate of 50-100V/us with these devices. The Levinson ML-2 power amp (JC-3) used similar devices, e.g. the 2N5684 and 2N5884 series.

G’day John.

Just out of curiosity, as I don’t design for or have any professional connection with the mass produced or audiophile markets, what kind of THD figures would you expect of a “high-end” amplifier? For instance that Douglas Self design I provided a link too in a previous post - THD typically .001% to 10kHz, less than 0.02% well beyond 20kHz.
These are figures indicative of performance I’d personally consider up there in the “high-end” class. In other words, they are THD figures well below those produced by the majority of “HIFI” amplifiers on the market.
I’ve achieved very similar results (in fact slightly better) in a bridged (I actually designed a differential op-amp topology with a common mode voltage servo) 600W design using the MJL21193/MJL1194 devices. It would be fair to say that with a better pair of transistors the results I achieved could have been even better, but I’d still say what I got was “high-end” performance.

Cheers,
Glen

[Bob Cordell]

Originally posted by G.Kleinschmidt


I’ve designed and built various bipolar A, AB designs from 500-5000W with THD levels well below perceptible limits. These are not mass produced units, but mostly one-off designs either for my own use or for specific applications.[/B]

Hi Glenn,

First of all, I hope everyone had a very enjoyable Thanksgiving break. I had lots of company at my home, so I was not able to participate much on the board for awhile.

Glenn, If you've designed amplifiers up to 5000W in Class AB, rather than H or G, it seems inconsistent that you would be worrying that much about idle bias power consumption. Unless your amplifiers are not rated for continuous operation at 1/3 power or even 1/8 power, the dissipation of the amplifier under those conditions is far more than that of idle bias.

You don't tell us your THD number, but assert it is "well below perceptible limits". Maybe you assert that 0.1% of THD-20 is below perceptible limits. How should I know? There are those who do. While 0.1% THD-20 may be OK for a tube amp, it is atrocious for a solid state high end amp.

I’m sorry, but I don’t agree with your assertion that a BJT amplifier biased to 20-50mA per device will only deliver “fair sonics”. That’s just plain absurd. The THD figure you quote isn’t particularly difficult to achieve in non-EC amplifiers with bipolar output stages and ultra low THD figures at 20kHz are superfluous anyway. If you can hear the second or third harmonic of 20kHz then you’re doing better than me.[/B]

First of all, one can make a very fine amplifier out of BJTs, and there are plenty of great ones out there. John Curl and I may disagree on a few things, but he makes a very fine BJT amplifier. He has his own good reasons for choosing that technology and knows how to tame its individual challenges. But you can bet he does not achieve his good sonics by biasing his output stage at only 20 mA.

I merely quoted my MOSFET non-EC 0.02% THD-20 to demonstrate that at only 150 mA, MOSFETs can do quite well, even without EC.

By making the point that one cannot hear the harmonics of 20 kHz, and that 20 kHz THD does not matter so much, you are showing that you do not understand that THD-20 is a symptom of underlying HF nonlinearity, not what one hears as the resulting sonic degradation. This is a very important distinction that you seem not to grasp. THD-20 is just a very convenient and well-known measurement technique, but what you actually are hearing when the cymbal spits rather than shimmers is intermodulation products, but the tendency to producing those IM products is quite well correlated to THD-20.

OK, but I don’t design for the esoteric audiophile market. I’m interested in producing efficient cost effective designs. Heat dissipation must be kept to a minimum and in that regard BJT’s with their much lower bias current requirements and ability to run at high temperatures without throttling back rule supreme. [/B]

This is baloney in super high power amplifiers if you are still using Class-AB rather than Class H or Class G. Do the power dissipation math at 1/3 power or 1/8 power. You never did tell us if your designs are able to operate continuously at 1/3 power or 1/8 power without overheating.

I didn’t say that the gm of Mosfets wasn’t plenty for linear amplification, I was making a comparison to BJT’s which have a much higher transconductance and therefore suffer from a particular distortion mechanism to a lesser degree than Mosfets devices. In source follower configurations the instantaneous output voltage is equal to the voltage at the gate minus the gate to source voltage, which for some devices can become significantly non linear at higher currents.[/B]

It sounds to me like you've never bothered to build a MOSFET amplifier. The relationship of Vgs to Id at higher currents in MOSFETs is not a problem at all.

I didn’t cite the MJL21193 as the best audio transistor out there, but as a unit with comparable robustness, as that is what you had asked for.

Things are obviously better where you are. I was comparing it to the TO3 IRF240, which from an Australian supplier costs approximately 50% more than both the MJ21193 and MJL21193. I didn’t get a price on the IRFP240.


The data sheets only provide limited information. That does not mean the MJL21193 will not withstand much high dissipation levels for brief 10mS bursts. I haven’t subjected one to destructive testing, but going by it’s robust 1 second rating I doubt that it would be much less able, if at all, than the IRF240 device.[/B]

I'm just saying you need to compare apples with apples. While it is true that 10 ms SOA will tend to be higher than 1 second SOA, if they don't spec it, you are taking a bit of a risk in assuming it.

Cheers.
Bob

[Bob Cordell]

Originally posted by G.Kleinschmidt
At 150 mA bias, my MOSFET amplifier did 0.02% THD-20 into 8 ohms without EC.




Bob, Just to further elaborate on the results that can be achieved with bipolar output stages, I provide a link to a rather basic, albeit refined Douglas Self design that does not use error correction and is of comparable power specification to your MOSFET design:


http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm


It produces less than 0.02% THD at 40kHz delivering 40W into 8 ohms.
I have never published my designs, but I have achieved similar results using paralleled MJL21193/MJL21194 complementary pairs at 15 times this power figure - devices which you assert are not suitable for high-end audio.

Cheers,
Glen

Hi Glenn,

I'm very familiar with virtually all of Doug Self's work. Overall he has done a fine job and contributed greatly to the understanding of good power amplifier design practices in our community. His designs tend to measure quite well on the bench, also.

However, one area in which I tend to disagree with Doug has to do with idle bias in Class AB emitter follower output stages. He has been very thorough in plotting net GM vs net output current, but has reached a conclusion that I do not see producing the best sonics. He ends up concluding that a quite low standing current is the optimum. He is correct in showing that distortion is high under under-biased conditions, then goes through a minimum, then increases as bias is increased further. He prefers that we bias at the optimum, balacing the distortion produced by under-biased conditions against the distortion produced by over-biased conditions (what he calls gm doubling distortion).

It is my view that in the real world he has made not the best tradeoff, ending up with too little idle bias. The distortion due to gm doubling is far more tame distortion and easier to deal with than the horrid distortion created due to under-bias. Given the vagaries of bipolar bias-setting and the fact that it is so sensitive to junction temperature variations due to program material, I believe that he foolishly operates too close to the edge on the low side of idle bias. That is my view on why so many high-end audiophile BJT amplifiers deliberately err on the high side when it comes to idle bias. I'm sure John would have a comment here.

Indeed, some practitioners like to operate Class-A up to a couple of watts if they can get away with it dissipation-wise. Sure, they will get some so-called GM doubling when they exit Class-A into Class-AB, but they deal with it, and it is the lesser Devil. Also, if you have numerous BJTs in paralllel, with maybe 0.27 ohm emitter ballast resitor on each, and each biased a bit toasty, the GM doubling effect is really quite small.

Cheers,
Bob

[Bob Cordell]

Originally posted by lumanauw
Hi, Bob Cordell,

What is your opinion about quasi complementary output stage (that uses all NPN devices)?

It can be seen that the positive part is common collector, while the negative part is common emitor usage mode of output transistors, so it is not balanced between positive and negative half.

What is the merit and drawback of this quasi complementary output stage?

The last time I had anything to do with a quasi-complementary output stage was in 1971. They are inferior on almost every level.

Bob

[john curl]

Bob, when it comes to distortion, itself: It is almost impossible to set a single number of what 20KHz thd should be with any amp. at full output. Actually, who cares? Who listens to 20KHz sine waves (or their equivalent) at full power at some very low distortion, just to have the amp clip with a 1dB increase in the output?
Personally, I like to concentrate on high frequency distortion at relatively high listening levels which are 1-25W for example. I usually use 5KHz, because my oscillator is happy at that frequency, it is in a more realistic high frequency range, and it is easy for my FFT analyser that follows my THD analyser to see all of the harmonics up to the 20th. Of course, I usually can do well enough with just 10 harmonics, but sometimes higher is useful.
It is in my GREATEST interest to note any 7th or 9th harmonic distortion, especially at listening levels. This is what I think separates the solid state sound from the tube sound, all else being equal. Your opinion may differ.
I must make good measurements at 20KHz, because Tom Holman insisted on putting that into his THX specs, and I make THX rated amps. I think that it is a poor choice of a standard of quality, but I commend you for doing so well with it.
I do not criticise Charles Hansen for not caring about it, because he is after a more 'authentic' sound than you or I are trying for in power amp design. Does he go too far, perhaps? Probably, but what the heck. "You pay your money and you take your choice.";-)

[lineup]

Originally posted by john curl

Of course, I usually can do well enough with just 10 harmonics, but sometimes higher is useful.
It is in my GREATEST interest to note any 7th or 9th harmonic distortion, especially at listening levels.
This is what I think separates the solid state sound from the tube sound, all else being equal.
Your opinion may differ.

A question.
I have run some discrete amplifier simulation tests.

And a Fourier analys at 1 kHz, can show 2nd, 3rd and 4th harmonic in falling steps nicely
and 5th, 6th can be very low.

But the 9th suddenly can be comparatively high, almost like 4th.
Also sometimes 7th sticks up a bit higher than 5th-6th.

But one preamplifier I simulated had very low of almost every harmonic 3-10, except for this 9th.
Didn't matter what levels, loads I used for running analysis.
And there was no way I could tweak away that 9th peak.

Some other circuits does not have this 9th problem, at least in simulations.


Question now.
What can cause this 9th harmonics distortion?
Power supply, some type of feedback or some special topology?

As I recall it, it was a no global feedback that had most of this constant high 9th.
But, then again, most circuits I put together are no feedback and all stages in Class A.


lineup

[Wavebourn]

Originally posted by G.Kleinschmidt
Wavebourne:
Even more gifted than the rest of people who don't care of spectral analyzing and other measurements, they just say: "I don't know why, and don't want to know why, but this sounds more natural than that".

I’m not particularly musically talented. When it comes to amplifier evaluation, specifications and measurements mean more to me than personal subjective evaluations.

Cheers,
Glen

We don't need to be musically talented to understand that length of piano's legs matter less than other parameters, right? And suppose, the whole industry is working hard to create perfect legs for piano because people think that length of piano legs impact significantly on sound quality because it have to stay strictly horizontally...

...and if ignorant in perfect piano design people subjective evaluate and say that piano with bad legs sounds better, shall we show them measurements of legs, or try to figure out what parameters are more significant to measure?

[GK]

Bob:
Glenn, If you've designed amplifiers up to 5000W in Class AB, rather than H or G, it seems inconsistent that you would be worrying that much about idle bias power consumption.


Absolutely not! Had I designed the same amplifiers with MOSFET’s in the output stages the dissipation would have been a great deal higher. It’s my very choice of class AB instead of the more efficient G or H configurations that has dictated my need to use BJT’s - the more efficient output devices.


Bob:
Unless your amplifiers are not rated for continuous operation at 1/3 power or even 1/8 power, the dissipation of the amplifier under those conditions is far more than that of idle bias.


I’m sorry, but that’s just plain wrong. A quick example with some basic calculations to put things into perspective - a 1kW RMS in 4 ohms AB amplifier would be required to deliver a peak voltage of SQRT(1000*4*2) = 89V with a peak load current of 89/4 = 22A. The average haft sinusoid current for one half of the output stage working at full output power is equal to Ipeak/pi = 22A/3.14 = 7A.
Lets assume this amplifier was made with a MOSFET output stage with 10 paralleled output pairs biased at 200mA each. That would give an idle current of 10*0.2A = 2A.

That’s an idle current of 2A Vs 7A average current when delivering full output power.

At 1/3 power output, the peak output voltage is equal to SQRT((1000/3)*2*4) = 51.6V, giving an average half sinusoid current for one half of the output stage of (51.6/4)/pi = 4.1A

Now we’ve got an idle current of 2A Vs 4.1A average current when delivering 1/3 of full output power.

At 1/8 power output, the peak output voltage is equal to SQRT((1000/8)*2*4) = 32V, giving an average half sinusoid current for one half of the output stage of (32/4)/pi = 2.5A.

Now we’ve got an idle current of 2A Vs 2.5A average current when delivering 1/8 full output power.

As you can deduce from these figures, the 1/8 power dissipation is by no means far less than the idle dissipation by any stretch of the imagination. Incidentally, my really high power designs were designed for continuous operation at 1/8 output power, as that was adequate for their intended application. I feel entirely justified in my choice of bipolar transistors in order to keep dissipation levels at a minimum. Even at 1/3 continuous output power, I believe I have shown that dissipation at idle conditions can be rather significant.


Bob:
You don't tell us your THD number, but assert it is "well below perceptible limits". Maybe you assert that 0.1% of THD-20 is below perceptible limits. How should I know? There are those who do. While 0.1% THD-20 may be OK for a tube amp, it is atrocious for a solid state high end amp.


OK, I wouldn’t be particularly enamoured with 0.1% either.


Bob:
First of all, one can make a very fine amplifier out of BJTs, and there are plenty of great ones out there. John Curl and I may disagree on a few things, but he makes a very fine BJT amplifier. He has his own good reasons for choosing that technology and knows how to tame its individual challenges. But you can bet he does not achieve his good sonics by biasing his output stage at only 20 mA.


Good. I don’t bias my output stages as low as 20mA either.


Bob:
I merely quoted my MOSFET non-EC 0.02% THD-20 to demonstrate that at only 150 mA, MOSFETs can do quite well, even without EC. By making the point that one cannot hear the harmonics of 20 kHz, and that 20 kHz THD does not matter so much, you are showing that you do not understand that THD-20 is a symptom of underlying HF nonlinearity, not what one hears as the resulting sonic degradation. This is a very important distinction that you seem not to grasp. THD-20 is just a very convenient and well-known measurement technique, but what you actually are hearing when the cymbal spits rather than shimmers is intermodulation products, but the tendency to producing those IM products is quite well correlated to THD-20.


You sure do infer a lot. I didn’t say that THD at 20kHz isn’t important, or to have a significant relation to an amplifiers IMD performance. I questioned the validity of attempting to achieve ultra low THD figures at 20kHz. In my opinion low THD is of lesser importance at 20kHz than at 10kHz or 1kHz. Many audio designers agree, many don’t. That’s what it’s like in the audio engineering domain.


Me:
OK, but I don’t design for the esoteric audiophile market. I’m interested in producing efficient cost effective designs. Heat dissipation must be kept to a minimum and in that regard BJT’s with their much lower bias current requirements and ability to run at high temperatures without throttling back rule supreme.

Bob:
This is baloney in super high power amplifiers if you are still using Class-AB rather than Class H or Class G. Do the power dissipation math at 1/3 power or 1/8 power. You never did tell us if your designs are able to operate continuously at 1/3 power or 1/8 power without overheating.


No it isn’t baloney. Did the math. Addressed this contention above. My super high power design had to fulfil a requirement of being powered by a remote 12Vlead acid battery bank with a commercially available SMPS at the battery bank end delivering the rail voltages. It wasn’t an economical proposition to provide for the multiple rail voltages required for either class G or H at the time, so I was stuck with class AB. Using BJT’s, I designed my amp as economically as I could.


Bob:
It sounds to me like you've never bothered to build a MOSFET amplifier. The relationship of Vgs to Id at higher currents in MOSFETs is not a problem at all.


I’ve used Exicon, Hitachi and Tosbiba MOSFETs all with a great deal of success in low (<100W designs). I did not say that the relationship of Vgs to Id was a very significant problem or insurmountable problem - I was just pointing out the fact that many MOSFETs have a lower intrinsic linearity in source follower applications than do many high power BJT’s.


Cheers,
Glen