G.Kleinschmidt said:
What happen to measured differences on low output power, if they were measured on high power? How capacitive character of load change them?
Why BJT, MOSFET, tube outputs with the same THD level on relatively high power sound so different on low power?
The question was, if such non-significant according to measurements differences are audible....
G.Kleinschmidt said:
Hello all readers and music lovers!
I have read at several different sources that
0.1% THD is at the limit to human hearing.
This means both those amplifiers are perfect
by THD criteria only
at a specific level, at a specific test frequency and with a specific test load
and no one will truly be able to hear anything different under those specific circumstances
When we talk levels of 0.50% THD or more, in relation to one other 0.01% THD amplifier
we can most probably say one is more HiFi, high fidelity,
( in true meaning of word, which is like: more accurate in signal reproduction )
than the other
--------------
Note!High Fidelity does not necessary mean sounding better, a more pleasing sound.
It means you get output in accordance what you input.
For better or for worse.
--------------
Say an artist produce a recorded song, which is very dissonantly mixed and produced
and both lyrics and musical sound is intended to express a worrying feeling - discomfort.
Now some NOT HiFi amplifier may add some even harmonies,
2nd 4th harmonics into this song.
This would destroy the intended artwork in its original form.
An HiFi, High Fidelity, amplifier is fidel, =true, to the original at such level
it is impossible to tell difference from original recording by ears.
-------------
A very good sounding amplifier may well be Not HiFi - high fidelity,
but InFi - infidelity, which also is called LowFi - low fidelity.
But this infidelity is such, that in most listened kinds of music, it is more pleasing.
Easier to your ears and mind.
-------------
Finally, we can not judge either HiFi or LowFi amplifiers as being all bad.
As long as anyone likes one or the other
and wants to use them, it is good to those who enjoys them.
Taste is difference, thinking is different and isn't this very good and interesting thing!
Indifference would be extremely boring
One might just as well go put an end to such an excuse for a life.
thinks lineup
you may think totally differentG.Kleinschmidt said:
I’d like to elaborate a bit further in the levels of power dissipation under quiescent conditions.
That 1kW into 4 ohm hypothetical design with a 89V peak voltage swing may have +/-110V rail voltages. With 10 paralleled MOSFET pairs biased to 200mA each we have a total of 440W of power dissipation when the amplifier is just idling – that’s a hell of a lot of quiescent power dissipation for an amplifier only rated at 1000W RMS into the load.
Using BJT devices of similar current and voltage ratings instead, that bias current could easily be cut down to one-fifth this level without sacrificing crossover distortion performance, giving an idle dissipation of 88W. That is a dramatic increase in efficiency.
Another thing that should be said, is that just because an amplifier may be rated to deliver 1/8 of maximum rated power on a continuous basis (endless hours) without over heating, it does not mean that such an amplifier does not have reserves to operate at much higher ‘continuous’ levels for briefer periods of time – say 10 or thirty minutes.
Now the hypothetical 1kW amplifier, with a BJT output stage and 88W of quiescent power dissipation would have much greater reserves to deliver high average power levels to the load before over heating than the equivalent MOSFET design with 440W of quiescent power dissipation - assuming both amplifiers have the same degree of heatsinking.
The MOSFET design will idle much hotter and will therefore thermally limit much sooner when called upon to deliver high output power levels. The recovery time of the MOSFET amplifier after thermally limiting will also be much greater than the BJT design, as at 5 times the quiescent power dissipation, it will take much longer to cool down.
The only way to alleviate this disadvantage of the MOSFET design is to heat sink it much more heavily than the equivalent BJT design – and that is a significant economical disadvantage.
john curl said:
I think we are in pretty good agreement here, John. I have liked 20 kHz THD in the past because it is a tougher test, and it correlates well with HF non linearity that will also result in in-band IM. But you are absolutely right about the difficulty of seeing the higher-order harmonics of THD-20 with ordinarily-available spectral analysis tools. You are also right in suggesting that not all THD-20 is the same; the same number composed of benign 2nd and 3rd is much less objectionable than the same number containing equal amounts of harmonics out to the 7th or 10th.
For this reason, these days I tend to focus more on CCIF 19 kHz + 20 kHz with full spectral analysis, since it still has reasonably good slew rate and HF stress, and readily shows the high-order IM products.
Bob
lineup said:
Well, I guess neither Bob Cordell or John Curl, two very well known audio designers,
can answer my question.
Or they just wont bother. Not even a comment ....
Maybe because my, lineup status, in the inner circle of audio society
is close to ZERO = 0.000000001.
Humbly while under a deep bow to these cruel facts
And with 1000 Regards
lineup
G.Kleinschmidt said:
Glenn,
You've made a HUGE mistake here. Your assumption that 10 pairs of paralleled MOSFETs would have to have a bias of 200 mA each for reasonable crossover distortion into 4 ohms is completely wrong.
What matters in the creation of static crossover distortion is the variation in total transconductance of the output stage in relation to the impedance of the load. In very rough terms, the re if a bipolar transistor is on the order of 26 ohms/Ic in mA, and that of a MOSFET is on the order of 250 ohms/Id in mA. That is why we normally say that a MOSFET needs about ten times the bias current, all else remaining equal.
Transconductance of both types of devices is roughtly proportional to current in this current range. A single MOSFET biased at 200 mA will have rs on the order of 1.2 ohm, the complementary pair on the order of 0.6 ohm. Let's say this ratio of 0.6 ohm against 4 ohms was OK for a 50W amplifier using a single pair of MOSFETs. The interesting thing is that this TOTAL amount of idle bias is STILL just as OK for the 1000 W amplifier using 10 pairs of MOSFETs paralleled each idling at 20 mA. Your logic in assuming that each of the ten pairs of MOSFETs would have to be biased at 200 mA was totally wrong.
I'm not saying I'd bias them that low in practice in a 1000W amplifier, but I hope you get the point. I would probably bias them at some level such that the idle bias dissipation was still below the 1/8 power dissipation. That is why I asked you about whether your amplifier was rated to work continuously at 1/8 power. If it has to be adequately heat-sinked to withstand 1/8 power continuously, then it can ceratinly withstand a few hundred mA of idle bias.
Cheers,
Bob
Bob Cordell said:
Sorry, I still disagree! I've worked on commercial high power MOSFET amps and have several articles of published designs which all run the individual bias currents of multiple parallel connected MOSFET pairs at anywhere from 100-200mA each. Perhaps my experience is limited, but I've never seen a high power MOSFET amp biased anywhere near as low as what you are suggesting.
I've always seen the need for 100-200mA of bias current for each audio MOSFET not strictly for the reduction of x-over distortion but to bias each device nearest possible to the zero temperature coefficent point of vgs Vs Id so as to keep the bias current stable with the wild temperature variations that can occur in an abused high power amplifier.
I tend to agree Glen. It would be almost impossible for me to insert Mosfets where I now have power transistors and get the same power in a practical way.
Also, I would think that I would want to bias the power Mosfets at 80ma ea or more, just to keep them from increasing the higher order distortion component due to the rate of change of the Gm at very low currents. What is the advantage of power mosfets in high powered amps anyway?
Lineup, I don't know exactly why you should get very high 9 harmonic distortion in a class A design. It would seem to be impossible. Maybe it is a computer program glitch.
Also, I would think that I would want to bias the power Mosfets at 80ma ea or more, just to keep them from increasing the higher order distortion component due to the rate of change of the Gm at very low currents. What is the advantage of power mosfets in high powered amps anyway?
Lineup, I don't know exactly why you should get very high 9 harmonic distortion in a class A design. It would seem to be impossible. Maybe it is a computer program glitch.
lineup said:
Well, I guess neither Bob Cordell or John Curl, two very well known audio designers,
can answer my question.
Or they just wont bother. Not even a comment ....
Maybe because my, lineup status, in the inner circle of audio society
is close to ZERO = 0.000000001.
Humbly while under a deep bow to these cruel facts
And with 1000 Regards
lineup
lineup,
I'm sorry I did not get back to you sooner. These last four days have been very busy with family and around the house, and its been difficult to keep up with the threads. Yesterday my wife had me spending the whole day putting up Christmas lights.
What you describe seems pretty unusual. Even with potentially sharp-edged distortions like crossover distortion, a strong high-order harmonic in my experience is almost inevitably accompanied by second and/or third harmonics of similar level (or higher). I may be off on the theory a bit here, but I think the worst case for harmonic spectral distribution is an impulse, in which all of the harmonics are about the same level. For a square wave, of course, we have all odd harmonics and they decrease in amplitude with increasing order.
Bob
G.Kleinschmidt said:
Glenn, OK, one thing at a time. You have now brought up a different issue, i.e., thermal bias stability.
Let me first re-iterate that there is absolutely no way one has to operate 10 pairs of MOSFETs at a total bias current of 2A to avoid crossiver distortion, as you had assumed. If you operate the total group at a few hundred mA, that will be adequate, since at that current level the transconductance of the group as a whole is adequate, and that is what counts for crossover distortion, whether the amplifier is a 50W or 1KW unit.
Now let's talk about the new issue you have brought up as a reason why you think that each of the ten MOSFET pairs needs to be biased at 200 mA, for a total idle of 2A. This is the issue of thermal bias stability in MOSFETs. I have covered this in detail in my MOSFET amplifier paper on my web site at www.cordellaudio.com. You cited instances where you claim others have paralleled multiple pairs of MOSFETs and run each pair at 100-200 mA. I can only think that you may have been referring to a design that used lateral MOSFETs.
All power MOSFETs start out with a positive TC of Ids at a given gate voltage. As current increases, the TC decreases, goes through zero TC, then goes negative. The current at which the TC crossover occurs depends on the type of MOSFET. With laterals, this TC crossover occurs around 100-200 mA, and provides for a nice characteristic that provides good thermal bias stability virtually for free (no heat sink temperature tracking devices needed, for example). So what you saw may have been a lateral design where they biased each pair at 100-200 mA to stay at this zero TC point. In that limited context, you are right: if one just kept the TOTAL idle current at, say, 200 mA, and ran half the current through each pair, one would give up at least part of the zero TC behavior, since each device would then be operating at a current below the TC crossover point.
As explained in my 20+ year old paper, this TC crossover occurs in the low ampere region for HEXFETs like the IRFP-240, meaning that in usual designs we are still stuck with a positive TC of Ids with these devices. This is no different than with bipolars, where the effect is actually far worse (see my paper). Therefore, there is no need or advantage in keeping each of the HEXFEt pairs operating in the 100-200 mA range for thermal stability purposes. Anything you say about thermal stability and concern about program-related temperature swings affecting operating point, it is usually going to be worse with bipolars than with HEXFETs.
Bottom line, there is no reason to operate ten pairs of HEXFETs in a 1000W amplifier at anything remotely approaching 2A idle.
Having said all of that, it is certainly true that the TOTAL amount of idle bias in a small amplifier using MOSFETs without EC almost always wants to be more than one with bipolars. As the amplifier gets bigger, the relatively fixed amount of needed MOSFET idle bias gets smaller, dissipation-wise, in comparison with the dissipation of the amplifier at 1/8 power or 1/3 power, so it is actually less of an issue in higher-power amplifiers. In the example you cited, even if we bias the output stage at a healthy 400 mA (40 mA per MOSFET pair), with 100V rails we are only dissipating 80 watts.
Let's look at the other extreme, however, in fairness. Suppose you are a manufacturer who wants to build a 5 X 75 wpc home theater amplifier with minimal heat sinking to meet a price point. There you WILL get burned, so to speak, if you choose MOSFETs over bipolars, because those MOSFETs WILL need substantially more idle bias to achieve low THD. Of course, in that crappy design, you probably don't have a design that can do 1/8 power with all five channels driven (but you might be able to do so with only two channels driven). Anyway, that I think is a good case where you would definitely suffer from the MOSFEts' need for greater idle bias to achieve a given transconductance.
Bob
john curl said:
Hi John,
Could you elaborate on why it would be almost impossible for you to insert MOSFETs where you now have bipolars and get the same power in a practical way?
I assume you would make the usual necessary circuit changes and accomodations.
Are you saying that because of the amount of idle bias you would have to use? Five paralleled pairs in a 250W channel, each pair biased at 80 mA, for a total of 400 mA would seem quite reasonable for achieving very good quality without resort to EC. This would result in 60 watts idle dissipation if you had 75 V rails. Is this more than the amount of idle dissipation you currently use in your better bipolar amps? How much do you idle your 250 watt amps at, per channel? I thought you were pretty decently into Class AAB. Maybe not as much as I thought?
If you are referring to a 5-channel amplifier, I do understand. I'm thinking in terms of one of your better stereo amplifiers.
I also do understand if your concern is the added cost of providing boosted supplies for the input and VAS stages. That I agree on for sure. Without boosted front-end supplies (by at least, say, 10V), the HEXFETs give up too much efficiency in use of the main power supplies because of their gate drive voltage needs. Boosted supplies don't cost a lot, but in a consumer product where every dollar counts, I agree that their cost impact cannot be ignored. An extra $10 in cost for the boosted supplies would boost the retail price by probably $50.
Or is there some other reason than idle bias that you believe it would be impossible for you to use MOSFETs?
Thanks,
Bob
lineup:
> 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 s imulations.
> Question now.
> What can cause this 9th harmonics distortion?
> Power supply, some type of feedback or some special topology?
I don't take Spice FFTs at -120 dB too serious. Remember, during transient analysis, the real result is approximated by a piecewise linear function by numeric integration with embedded linearization around the instantaneous operating point. The quality of this approximation depends depends on some variables you can set, but also on the ratio of fastest/slowest time constants, lucky pivoting of the admittance matrix, sharp nonlinearities etc. Sometimes, it doesn't even converge at all.
lineup:
> ...can answer my question.
> Or they just wont bother. Not even a comment ....
I'm not an Audio Deity and to my excuse I can say that I spent this weekend in Dresden, where I enjoyed a nice classical concerto in the reconstructed Frauenkirche.
That combination of circle and ZERO reminds me at a word by Dirac (the guy with this pulse named after him):
"Some people have a range of sight of radius zero. They call that their point of view."
regards, Gerhard
> 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 s imulations.
> Question now.
> What can cause this 9th harmonics distortion?
> Power supply, some type of feedback or some special topology?
I don't take Spice FFTs at -120 dB too serious. Remember, during transient analysis, the real result is approximated by a piecewise linear function by numeric integration with embedded linearization around the instantaneous operating point. The quality of this approximation depends depends on some variables you can set, but also on the ratio of fastest/slowest time constants, lucky pivoting of the admittance matrix, sharp nonlinearities etc. Sometimes, it doesn't even converge at all.
lineup:
> ...can answer my question.
> Or they just wont bother. Not even a comment ....
I'm not an Audio Deity and to my excuse I can say that I spent this weekend in Dresden, where I enjoyed a nice classical concerto in the reconstructed Frauenkirche.
That combination of circle and ZERO reminds me at a word by Dirac (the guy with this pulse named after him):
"Some people have a range of sight of radius zero. They call that their point of view."
regards, Gerhard
Bob Cordell said:
Thanks Bob.
Yes, most simulations where I got the specific 9th problem
also had 2nd, and 3rd that were even a level higher than 9th.
But it is strange, I think, to find 9th ( or 7th) that is higher
than 4-5-6-7-8.
It is not higher than 2nd /3rd , but there is something like
a peak of 10 dB compared to to the previous and surrounding ones.
And I could simply not trim or tweak down this 9th whatever I tried.
I am not very interested if this 7th or 9th is good or not.
I want to know what can be done to eliminate.
To me, lineup, all distortion is bad,
and no number harmonics dist is better than the other.
They all mean that the output signal is not in line with THE INPUT signal
And my question, again was:
What can cause this 9th (and sometimes 7th) peak in harmonics?
somebody may have some clue to cause, reason we get this unexpected test data
somebody may have done trial and error investigations
to see what is the mechanism behind these higher harmonics peaks
Regards
lineup
Lineup,
it is quite common to get a lot of high order peaks in the spectrum due to simulation/FFT artefacts. Maybe that is what you are seeing? There are a few things you can check:
1) Do an FFT not only of the signal of interest, but also of the source signal. This shows if you choosen parameters for simulation and FFT are sufficient to get a good result. If your distorsion appear also in the source, then it is a simulation/FFT problem.
2) Try simulating with shorter step size and/or simulating more cycles. That will take more time but also give a better result. This will show if your high-order distorsion tend to dissapear (in which case it is probably a simulation artefact) or stay (in which case it could be a real result). I usually run at least 11 cycles and don't start collecting data until after the first cycle.
3) Also try different parameters for FFT.
it is quite common to get a lot of high order peaks in the spectrum due to simulation/FFT artefacts. Maybe that is what you are seeing? There are a few things you can check:
1) Do an FFT not only of the signal of interest, but also of the source signal. This shows if you choosen parameters for simulation and FFT are sufficient to get a good result. If your distorsion appear also in the source, then it is a simulation/FFT problem.
2) Try simulating with shorter step size and/or simulating more cycles. That will take more time but also give a better result. This will show if your high-order distorsion tend to dissapear (in which case it is probably a simulation artefact) or stay (in which case it could be a real result). I usually run at least 11 cycles and don't start collecting data until after the first cycle.
3) Also try different parameters for FFT.
Bob, please don't second guess me. I use more than 100ma/device in my JC-1 amp, BUT I operate at +/ 90V with a +/- 100V driver supply. You run out of practical voltage headroom with your caps, predrivers, etc, above this. Why do you insist that you can do something better than what we already do?
Christer said:
Thanks Christer.
Yes.
As almost nothing I changed in the circuit would effect the 9th,
I had a little though it may have to do with the analysis, simulation itself.
I will try your good advice #1), to test the INPUT signal, the source.
In real life, using a multimeter or an oscilloscope or whatever,
with an error / distortion LARGER than circuit / device under test,
is not good.
When changes in your circuit is at the test gear limit or below
you can't verify any effects. Not with accuracy, anyway.
lineup
john curl said:
Thanks for comment, John Curl
Maybe I haven't read the post where this was told
because I can't remember seeing that information
Please, Can you refer to post #number or link to it, Curl.
I promise i will read it with interest and hopefully learn about this.
When you say 'staircase', in FFT analysis,
reading from left to right, and harmonics are 1-2-3-4-5-6-7-8-9-10
Do you mean with this,
.... a staircase that goes down in steps first and on the other side goes up a bit?
Regards
lineup