Re: Re: Re: Re: Re: Re: Re: Re: Output impedance of Hafler DH220
Cool down, Glen. This all started with me just providing the community with the output impedance characteristics of the Hafler, which I said earlier that I would do. I did not even mention your name in that post, and I was not trying to rub your nose in anything. You're the one who jumped on it and decided to be paranoid about it.
If David Hafler had decided to design, say, a 50-watt lateral MOSFET amplifier with only a single pair of output transistors, I am quite sure he would have biased each transistor near 200 mA, so as to maintain his distortion at a reasonable value. With probably 40V rails for a 50 watt amp, each channel at idle would still dissipate only about 16 watts, still not a lot. As a result, he would have still ended up with about a 0.25 uH effective output inductance.
If, instead, he had followed your scenario, he would have doubled his output distortion and his effective output inductance would have risen to a mere 0.5 uH, still not a lot. That higher distortion would probably be a much bigger issue than an increase of 0.25 uH in the effective output impedance before the use of the coil.
I have not looked at it or thought about it a lot, but it is also not clear that the effective output inductance will rise that much when a MOSFET output stage is driven directly by the VAS (all else remaining equal, including total output bias current and amount of NFB at 20 kHz). The reason for this uncertainty on my part is that the output impedance of a Miller-compensated VAS will be decreasing as frequency increases, so the MOSFET source followers may see a source impedance that is sufficiently low enough not to matter a lot. Why don't you try SPICing such an arrangement and let us know if your speculation that driving directly from the VAS really does increase effective output impedance.
Perhaps you can explain to us your assertion that MOSFET power amplifiers in which the output MOSFETs are driven directly by the VAS will exhibit higher effective output inductance.
BTW, the Hafler has no where near 40 dB of NFB at 20 kHz. Where did you get that idea?
You are right, there are plenty of well-designed MOSFET amplifiers out there with different architectures and tradeoffs. Maybe you should go measure some of them for effective output inductance.
Cheers,
Bob
G.Kleinschmidt said:
Excuse me? All I did was to highlight a rather specific set of design constraints which can result in an amplifier having a relatively high effective output inductance - namely relating to designs which drive the MOSFET's from a high impedance source (eg directly from the VAS) and/or use limited negative feedback.
What you provided is the measured results of an amplifier which doesn't qualify, followed by an accusation of “back-peddling” on my part. And now you characterise my specific statements pertaining to output impedance as “overly-broad generalisation about MOSFETs”. Give me a break.
There are many perfectly decent and competently designed MOSFET amplifiers in existence that drive the MOSFET’s directly from the VAS and/or use limited global negative feedback (especially not as high as 40dB at 20kHz). In fact, I’d say that the majority of MOSFET amplifiers designed and built by the DIY community would qualify as such, so I hardly see what is wrong with my attempt to raise the issues of these designs with respect to output impedance here at DIYaudio.
Such designs are the sole domain of fools according to you. Well fine, I’m sure others can decide on the merits of that declaration with out my need to comment.
Also, following my rebuttal, I notice that you’d rather not clarify any of your comments with regards to the biasing current/method used in the DH-220, or the alleged misunderstanding of how source followers behave on my part.
Thanks,
Glen
Cool down, Glen. This all started with me just providing the community with the output impedance characteristics of the Hafler, which I said earlier that I would do. I did not even mention your name in that post, and I was not trying to rub your nose in anything. You're the one who jumped on it and decided to be paranoid about it.
If David Hafler had decided to design, say, a 50-watt lateral MOSFET amplifier with only a single pair of output transistors, I am quite sure he would have biased each transistor near 200 mA, so as to maintain his distortion at a reasonable value. With probably 40V rails for a 50 watt amp, each channel at idle would still dissipate only about 16 watts, still not a lot. As a result, he would have still ended up with about a 0.25 uH effective output inductance.
If, instead, he had followed your scenario, he would have doubled his output distortion and his effective output inductance would have risen to a mere 0.5 uH, still not a lot. That higher distortion would probably be a much bigger issue than an increase of 0.25 uH in the effective output impedance before the use of the coil.
I have not looked at it or thought about it a lot, but it is also not clear that the effective output inductance will rise that much when a MOSFET output stage is driven directly by the VAS (all else remaining equal, including total output bias current and amount of NFB at 20 kHz). The reason for this uncertainty on my part is that the output impedance of a Miller-compensated VAS will be decreasing as frequency increases, so the MOSFET source followers may see a source impedance that is sufficiently low enough not to matter a lot. Why don't you try SPICing such an arrangement and let us know if your speculation that driving directly from the VAS really does increase effective output impedance.
Perhaps you can explain to us your assertion that MOSFET power amplifiers in which the output MOSFETs are driven directly by the VAS will exhibit higher effective output inductance.
BTW, the Hafler has no where near 40 dB of NFB at 20 kHz. Where did you get that idea?
You are right, there are plenty of well-designed MOSFET amplifiers out there with different architectures and tradeoffs. Maybe you should go measure some of them for effective output inductance.
Cheers,
Bob
PMA said:
You are exactly right. Indeed, in the case of the lateral MOSFETs, if you bias them above the 100 mA number, as Hafler does in the smaller amplifier, you begin to move ever so slightly into the negative TC region of the MOSFETs, which will, if anything, improve bias stability.
Cheers,
Bob
G.Kleinschmidt said:
Glen, there is no mandate that people have to bias lateral MOSFETs right at the zero TC point. The TC changes fairly slowly, and one can depart quite a bit before having to apply a lot of bias temperature compensation.
Of course, with bipolars, we never have that option in the first place, do we?
Bob
Bob Cordell said:
OK, so now I am paranoid as well. Another compliment. You also have a very short memory. The first line of your post was:
"In an earlier post, Glen had expressed concern about the HF output impedance.........."
I'm not sure what you were trying to achieve by accusing me of back-pedaling after I simply interjected to point out that the DH-220 was not such a design of which I am “concerned” about.
He did design such an amp. It is called the DH120. The output devices are biased at approximately 100mA.
The published THD specifications for the DH-120 do not indicate a doubling of distortion over the DH-220 and the 100mA per device pair biasing scheme is not my invention. I am not promoting it or otherwise. It is something that was commonly done with the devices used in the Haffler amps and in the Hafler amps.
I am not saying the 0.5uH is a lot. Suppose that the DH-120 had 6dB less NFB, the value might be up to about 1uH. THD would be worse, but I'm not arguing otherwise and I wouldn't design an amplifier in such a way. I also don't think that 1uH is a lot either. I am simply pointing out the fact that such values are achievable.
This topic of effective output inductance stemmed from a discussion on output coil audibility. I suggest that anyone concerned with the audibility of a 0.5-2uH output inductor should also be concerned about the effective output inductace of their amplifiers. Especially those who prote the use of low negative feedback.
Is that OK with you?
Here is another design of a "fool", described in a paper discussing the implications of possible higher (with respect to BJT's) output impedance of MOSFET output stages:
http://users.ece.gatech.edu/~mleach/papers/Feedforward.pdf
It has a considerably higher effective output inductance than the "pedestrian" DH-120 and DH-220 combined.
I think that you need to think about it a little harder. I'm through with wasting my time.
I did not say that it did. Pay closer attention to what I wrote. The 150mA for the DH-120 is the total current consumption for the entire channel. The actual output devices see considerably less.
Well of course there isn't and I never said that there was such a mandate. Did you even read this post of mine: ?
http://www.diyaudio.com/forums/showthread.php?postid=1229173#post1229173
Funnily enough, I don't recall saying otherwise.
G.Kleinschmidt said:
Hi Glen,
My bad, I DID mention your name. Us old geezers do have memory problems at times
. Anyway, I really was not trying to make you look bad, just presenting some data that I thought was interesting. I'm sorry if it came across that way.With regard to my question to you about direct driving of the VAS causing increased effective output inductance, I seriously want to know why you think that is. Please don't duck the question.
Finally, yes, my bad again; when I quickly looked at the manual and saw the 150 mA number, I did not look to see how it was being measured. Apparently the input stage, VAS and pre-drivers together consume 50 mA.
Cheers,
Bob
Bob Cordell said:
Hi Glen,
My bad, I DID mention your name. Us old geezers do have memory problems at times
With regard to my question to you about direct driving of the VAS causing increased effective output inductance, I seriously want to know why you think that is. Please don't duck the question.
Finally, yes, my bad again; when I quickly looked at the manual and saw the 150 mA number, I did not look to see how it was being measured. Apparently the input stage, VAS and pre-drivers together consume 50 mA.
Cheers,
Bob
If the VAS is not buffered from the input capacitance of the MOSFETs the high frequency gain of the VAS is seriously compromised, as is the bandwidth of the output stage. This constrains the open loop gain, dictates a lower dominant pole frequency for stability; thus resulting in less global negative feedback, particularly out to high frequencies.
Since the effective output inductance is directly proportional to the degree of negative feedback applied, it increases proportionately.
G.Kleinschmidt said:
As long as you don't try to go ultra-low on the VAS Miller cap (as in the Leach design), you might find the output impedance of the VAS at a frequency of, say, 1 MHz to be surprisingly low. So the situation vis-a-vis stability may not be as severe as it appears at first glance.
Re: Re: Re: Re: Re: Re: Re: Re: Re: Output impedance of Hafler DH220
Hi Glen,
I'm sorry if I have missed your point. I suppose it was about buffering the VAS and its influence on the effective output inductance, right?
Although I utilized buffers, I don't think it's that important with respect to the output impedance, as I regard them as a simple gain stage, just like any other gain stage. Removing them and inserting a separate VAS, would probably yields the same results, because the available NFB loop gain counts in the first place and this has the greatest impact on the output inductance.
I know your objections on MOSFET's. So, my primary reason was to show you what you can do with MOSFET's, even in a simple amplifier comprising only four gain stages.
As for the effective output inductance, my simulation (neglecting all stray inductances etc.) reveals a mere 5nH, opposing to 50nH of a typical 'blameless' amp. (indeed, suspect low values, but I can't get them higher).
Cheers, Edmond.
G.Kleinschmidt said:
Hi Glen,
I'm sorry if I have missed your point. I suppose it was about buffering the VAS and its influence on the effective output inductance, right?
Although I utilized buffers, I don't think it's that important with respect to the output impedance, as I regard them as a simple gain stage, just like any other gain stage. Removing them and inserting a separate VAS, would probably yields the same results, because the available NFB loop gain counts in the first place and this has the greatest impact on the output inductance.
I know your objections on MOSFET's. So, my primary reason was to show you what you can do with MOSFET's, even in a simple amplifier comprising only four gain stages.
As for the effective output inductance, my simulation (neglecting all stray inductances etc.) reveals a mere 5nH, opposing to 50nH of a typical 'blameless' amp. (indeed, suspect low values, but I can't get them higher).
Cheers, Edmond.
G.Kleinschmidt said:
Thanks, Glen.
I see your reasoning. However, I thought that you were asserting that directly driving with the VAS would somehow increase output inductance, even when the amount of NFB at 20kHz was held the same. That was how I posed my question, when I said "all else remaining equal". I guess we were just talking about two different things.
If designed right, there is nothing wrong with an amplifier where the VAS drives the output MOSFETs directly, but it is easy to do it wrong. If the load capacitance presented by the MOSFETs is allowed to significantly control the HF gain, then distortion will result because that capacitance is quite non-linear.
On the other hand, if one uses a VAS that is well-biased and maintains a low impedance at its output at high frequencies, as with Miller compensation, then such an amplifier can be fine.
Bob
Re: ....ignore the problem...
Upupo aopa & Bobby
a welding machine amp .. Super!
... and true, who has got several thousands bucks
in this forum.
let's build amplifiers ( www.firstwatt.com )
that 99% can build, and still have food money left
to buy something to eat
the rich one's have so many choices, anyway,
without www.diyaudio.com give them service
thinks
lineup - who is extremely rich, by some standard
.... regarded poor by the other standard
Upupa Epops said:What about design an amp like welding machine, John ?
By nominal output power 3 W
it will be not necessary price several thousands bucks...
.
Upupo aopa & Bobby
a welding machine amp .. Super!
... and true, who has got several thousands bucks
in this forum.
let's build amplifiers ( www.firstwatt.com )
that 99% can build, and still have food money left
to buy something to eatthe rich one's have so many choices, anyway,
without www.diyaudio.com give them service
thinks
lineup - who is extremely rich, by some standard
.... regarded poor by the other standard
Bob Cordell said:
To address both yours and Andy’s point with regards to VAS impedance; All else remaining equal, a VAS Miller-compensated so as to have a low output impedance at high frequencies to negate the effects of MOSFET input capacitance will generally need to have it’s frequency response rolled off much earlier than would be necessary if the VAS was buffered from the MOSFET output devices with an emitter follower.
One can always run the VAS with a higher standing current to compensate, but then one inevitably runs into the practical problem of providing (with acceptable linearity) the necessary drive current for the VAS miller capacitance from the input stage.
Cheers,
Glen
G.Kleinschmidt said:
I think we are all agreeing that a VAS that is directly driving the MOSFETs wants to run at higher standing currents for a multiple of good reasons. However, that, in itself in practice is not that difficult to do, especially if the VAS is a Darlinton. I would also argue that it does not necessarily have to have its response rolled off early. Running the VAS hot is not something I like to do, but many others do a fine job with it. It does mean that you have to deal with its power dissipation, and it may limit slightly your selection of transistors for use in the VAS, but this should not be a problem.
However, even if you don't run the VAS hot, you might be surprized how low its output impednace can get at high frequencies due to the shunt feedback effect of the Miller compensation capacitor. This is especially the case in a VAS that uses a Darlinton and which is fed by a current-mirror-loaded input stage.
Let's say you run the VAS at 10 mA and degenerate it by 10:1. It will then have an emitter resistor of about 25 ohms. At frequencies where the Miller capacitor is effectively a short, essentially all of the VAS output voltage appears back at the base, and the resulting 100% shunt feedback results in a VAS output impednace of a mere 25 ohms. Since the input impednace of the Darlinton is high due to beta-squared, and the output impedance of the driving current mirror is high, one has to go to a faily low frequency to get to the point where the Miller capacitor feedback voltage is attenuated very much against the input impedance of the Darlington VAS.
With betas on the order of 50, the impedance at the input node of the VAS will probably be on the order of 50K. Suppose the Miller capacitor is 50 pF. The 50 pF corners with the 50K at about 60 kHz; this is where the VAS output voltage fed back to the VAS input is down 3 dB if I have done my math right. The VAS output impedance will rise at 6 dB per octave from this point as we go down in frequency. So at 20 kHz, the VAS output impedance may be up to 100 ohms or so (this is even lower than I thought, hope I did the math right).
The last thing that must be considered is that the Miller capacitor will form a capacitance voltage divider with any shunt input capacitance at the input node of the Darlinton VAS. If there is 20 pF there, for example, there will be some fractional loss against the 50 pF Miller capacitor which will increase impedances a bit, at least at the high frequencies. Of course, this is a bunch of hand-waiving here, and this sort of thing is what SPICE simulations are really good for confirmation of.
Cheers,
Bob
Bob Cordell said:
Yes, I agree, but this is a rather ideal scenario and not all MOSFET amplifiers have been built this way. Suppose your input stage is a symmetrical differential pair driving a push-pull VAS. Such a design may have a collector load resistor for each differential pair in the vicinity of 1k. 50pF corners with 1k at approximately 3MHz.
Still, a miller capacitance of not much greater that 50pF may only be required due to the much lower gain of the input stage.
The output impedance of the VAS is now significantly higher. At the gain cross over frequency it may be in the vicinity of 100 ohms or so. 100 ohms may not seem like a lot, but it is much higher than what the output stage driving impedance would be it the VAS was buffered with an emitter follower. MOSFET input capacitance is large, and you do not need to drive it with an overly large impedance to incur a significant low-pass frequency pole.
This frequency pole will inevitably need to be combated with a dominant pole roll-off frequency significantly less than would be the case if the VAS was buffered.
Cheers,
Glen
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Output impedance of Hafler DH220
I did not say the an amplifier with low output impedance cannot be made if MOSFETs are used in the output stage.
I said that low power amplifiers using simple, common topologies with MOSFET output stages can have effective output inductances in the range of concern of those worried about output coil/inductance audibility.
Since you agree that NFB plays the greatest part in determining output impedance, I don’t see how you can generally dismiss the utility of VAS buffering in this regard, which has the benefit of extending the high frequency response of the output stage by presenting it with a very low drive impedance.
Cheers,
Glen
estuart said:
I did not say the an amplifier with low output impedance cannot be made if MOSFETs are used in the output stage.
I said that low power amplifiers using simple, common topologies with MOSFET output stages can have effective output inductances in the range of concern of those worried about output coil/inductance audibility.
Since you agree that NFB plays the greatest part in determining output impedance, I don’t see how you can generally dismiss the utility of VAS buffering in this regard, which has the benefit of extending the high frequency response of the output stage by presenting it with a very low drive impedance.
Cheers,
Glen
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Output impedance of Hafler DH220
Hi Glen,
Okay, you didn't say that, but still you don't like MOSFET's very much.
Happily, you say "can have". So, we agree that not all simple topologies suffers from a too high output output impedance.
No, I didn't say it in that way. But first, look at my schematic again. Then you'll see that the VAS is part of the current mirror, i.e. the VAS has a current gain of only 1x. So, what I'm trying to explain is this: if one dismiss the VAS buffering and replace the VAS with something that do have gain (a traditional VAS), then the performance will be roughly the same, because the NFB loop gain hasn't changed much.
I verified this by simulation and got about 10nH output inductance (compared to 5nH). In other words, the output inductance does not depends on VAS buffering per se, it's the loop gain that counts, as we both agree on that. Theoretically it doesn't matter where you put (most of) the gain, in the input stage, before the VAS or after the VAS (as a buffer), however, I agree with you that the most appropriate place is between the VAS and O/P stage. Besides, it has other advantages too, like more drive current to the gates.
Cheers, Edmond.
G.Kleinschmidt said:
Hi Glen,
Okay, you didn't say that, but still you don't like MOSFET's very much.
G.Kleinschmidt said:
Happily, you say "can have". So, we agree that not all simple topologies suffers from a too high output output impedance.
G.Kleinschmidt said:
No, I didn't say it in that way. But first, look at my schematic again. Then you'll see that the VAS is part of the current mirror, i.e. the VAS has a current gain of only 1x. So, what I'm trying to explain is this: if one dismiss the VAS buffering and replace the VAS with something that do have gain (a traditional VAS), then the performance will be roughly the same, because the NFB loop gain hasn't changed much.
I verified this by simulation and got about 10nH output inductance (compared to 5nH). In other words, the output inductance does not depends on VAS buffering per se, it's the loop gain that counts, as we both agree on that. Theoretically it doesn't matter where you put (most of) the gain, in the input stage, before the VAS or after the VAS (as a buffer), however, I agree with you that the most appropriate place is between the VAS and O/P stage. Besides, it has other advantages too, like more drive current to the gates.
Cheers, Edmond.
Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Re: Output impedance of Hafler DH220
Hi Gen,
Just to be clear, I don't think many of us are saying that it is not a good idea to buffer the VAS before the output stage of a MOSFET amplifier. I think we only got onto the topic of the un-buffered VAS architecture when you brought it up as a legitimate case of a MOSFET amplifier that might incur an elevated effective output inductance.
I certainly have always advocated a buffer between the VAS and the output MOSFETs, but have also allowed that with proper design one can make a good MOSFET amplifier without such buffering.
Cheers,
Bob
G.Kleinschmidt said:
Hi Gen,
Just to be clear, I don't think many of us are saying that it is not a good idea to buffer the VAS before the output stage of a MOSFET amplifier. I think we only got onto the topic of the un-buffered VAS architecture when you brought it up as a legitimate case of a MOSFET amplifier that might incur an elevated effective output inductance.
I certainly have always advocated a buffer between the VAS and the output MOSFETs, but have also allowed that with proper design one can make a good MOSFET amplifier without such buffering.
Cheers,
Bob
G.Kleinschmidt said:
Hi Glen,
The architecture I cited as an example is certainly not ideal or unusual, and is very similar in many ways to the blameless architecture. The use of a current-mirror load and a Darlinton VAS is pretty common.
Nevertheless, you make a very legitimate point that not all MOSFET amplifiers use that architecture. Indeed, your example of the full-complementary dual differential pair driving the push-pull VAS is probably just as common. And yes, in that case there is a very finite load resistance at the input of the VAS.
Indeed, consider a simple amplifier with a single differential pair where there is such a load at the input to the VAS. Let the input pair be biased at 1 mA per transistor and let each be degenerated by 250 ohms. Then assume a 680 ohm load on the base of a non-darlington VAS. Let the VAS be operated at 10 mA and be degenerated 10:1 by a 25 ohm emitter resistor. Assume an amplifier closed loop gain of 26 dB and a closed loop bandwidth of 400 kHz and 26 dB of NFB at 20 khz. The Miller capacitor will need to be about 36 pF.
The 36 pF Miller capacitor will corner with the 680 ohm VAS base resistor at about 6.5 MHz, meaning that the output impedance will rise from 25 ohms at 6 dB/octave as we go down in frequency starting at 6.5 MHz. Zout of the VAS will be up to about 8 k at 20 kHz, corresponding to an effective output impedance of about 1000 pF.
Even given the bootstrapping of the MOSFET Cgs, the capacitive load of the MOSFET output stage may be comparable to the 1000 pF of effective VAS output impedance, so there will be some vulnerability to the nonlinearity of the MOSFET capacitive load. It is also true that, if nothing else were changed, the bandwidth of the amplifier would naturally tend to come in somewhat, and there might be some excess phase due to the fact that the capacitive load of the output stage is decreasing the amount of pole-splitting we are used to getting from the Miller compensation. I think this is your point. I can't disagree.
But I would say that the distortion performance of this example amplifier has been compromised. I would not consider such an amplifier to be a very good design. The rise in effective output inductance to closer to a microhenry would be the least of its problems.
Your point actually highlights one of the reasons why I do not use the full complementary dual differential pair architecture. As has been covered before, most of the time this architecture requires the use of the VAS input load resistor, and does not normally permit the use of the current mirror load if VAS standing current is to be reasonably defined. This means that we are giving up the benefit of the VAS operating with the higher tight-loop gain of the Miller compensation, resulting in a high VAS effective output impedance and more susceptibility to load capacitance.
This can be avoided to some extent by using the form of compensation that I used in my amplifier instead of Miller compensation, where what was the Miller capacitor is now fed back to the input stage input. This tends to get back the feedback we had around the VAS at high frequencies, and in so doing again lowers the output impedance of the VAS so that it is more suitable for driving a MOSFET output stage directly. But this is no longer a particularly ordinary amplifier. BTW, it looks like you have chosen to use this form of compensation in your amplifier. If so, good choice.
Cheers,
Bob
Nelson Pass said:
Perhaps I'm missing something, but it seems to me that this
has been around a long time. I was using it in '73, and I doubt
very much that I invented it. Is there something else that makes
it out of the ordinary?
Hi Nelson, no, I don't think you're missing anything. I don't know who invented this form of compensation. I am a little surprised that I have not seen it used more often, however.
Bob