I was reading an article yesterday by lynn olsen (spelling??) about why tube amplifiers are often preferred (by listeners) to solid state amplifiers even though their specs aren't as good. Lynn hypothesised that this difference which measurements such as THD (etc. etc.) could not pick up was in fact due to miller capacitance of solid state devices. He said something along the lines of transisters being very poor capacitors (worse than electrolytics) and when the distortion of the capacitance within the device is amplified by the gain of the circuit... well you can see where he was going with this.
So does anyone have any thoughts on this.? Possibly circuit suggestions which could combat this miller capacitance. I was thinking that some devices have lower capacitance than others, are these the ones we should be using?
Dan
P.S. At present I don't have an opinion on this it's just what I read and I would like to hear what others have to say.
So does anyone have any thoughts on this.? Possibly circuit suggestions which could combat this miller capacitance. I was thinking that some devices have lower capacitance than others, are these the ones we should be using?
Dan
P.S. At present I don't have an opinion on this it's just what I read and I would like to hear what others have to say.
Sorry,
The last thing I want to do is misrepresent Lynn. Grey, he did mention this, in fact I think he said it very similarly to what you just did. He basically said that the dielectric of the transistor was very poor in comparison to tubes. I will post the link to the article tonight when I get home so I don't cause anyone any confusion.
Dan
The last thing I want to do is misrepresent Lynn. Grey, he did mention this, in fact I think he said it very similarly to what you just did. He basically said that the dielectric of the transistor was very poor in comparison to tubes. I will post the link to the article tonight when I get home so I don't cause anyone any confusion.
Dan
Hi Ding.
A circuit wich can, if not circumvent the Miller-capacitance, then at least minimize this, is tha cascode design, of wich mr. Pass has an article:
http://www.passlabs.com/articles/cascode.htm
It'll not only increase bandwidth, the slewrate will benifit from it, as well as the risk of self-oscilation will be less.
A circuit wich can, if not circumvent the Miller-capacitance, then at least minimize this, is tha cascode design, of wich mr. Pass has an article:
http://www.passlabs.com/articles/cascode.htm
It'll not only increase bandwidth, the slewrate will benifit from it, as well as the risk of self-oscilation will be less.
I really value and respect Lynn's opinions (and his designs). He's authored some fine articles, and I highly recommend anything you come across with his name on it.
That said, there are of course numerous precautions which can be taken to mitigate the aforementioned effects... including cascode design. I wouldn't worry too much about the miller capacitance though. In the primary gain stage of most solid state amplifiers, you'll have a compensation capacitor (usually of high quality) whose value should swamp the miller capacitance of the gain transistor, possibly in combination with a cascode. On top of that, you've usually got plenty of feedback to help linearize the system as a whole.
There are several more subtle effects I've come across which i think deserve somewhat more attention. One of these is transient thermal distortion. There's an interesting site on that here: http://peufeu.free.fr/audio/
If you look at many of the Pass designs from the perspective of thermal memory distortion, you'll see that they're probably rather immune to these effects... high-biased class-A MOSFETs aren't going to feel much parametric change from very small instantaneous power dissipation changes. A close look also reveals that MOSFETs have much higher gate capacitance (miller capacitance) than BJTs. You may draw your own conclusions from this...
That said, there are of course numerous precautions which can be taken to mitigate the aforementioned effects... including cascode design. I wouldn't worry too much about the miller capacitance though. In the primary gain stage of most solid state amplifiers, you'll have a compensation capacitor (usually of high quality) whose value should swamp the miller capacitance of the gain transistor, possibly in combination with a cascode. On top of that, you've usually got plenty of feedback to help linearize the system as a whole.
There are several more subtle effects I've come across which i think deserve somewhat more attention. One of these is transient thermal distortion. There's an interesting site on that here: http://peufeu.free.fr/audio/
If you look at many of the Pass designs from the perspective of thermal memory distortion, you'll see that they're probably rather immune to these effects... high-biased class-A MOSFETs aren't going to feel much parametric change from very small instantaneous power dissipation changes. A close look also reveals that MOSFETs have much higher gate capacitance (miller capacitance) than BJTs. You may draw your own conclusions from this...
Here is the URL for the article
http://whoville.ucsd.edu/~stark/audio/PF1.html
You'll have to scroll down quite a way to find the Q&A on tubes Vs. SS.
Thanks for the links Hoffmeyer and HifiZEN I will have a look at these sites.
Dan
http://whoville.ucsd.edu/~stark/audio/PF1.html
You'll have to scroll down quite a way to find the Q&A on tubes Vs. SS.
Thanks for the links Hoffmeyer and HifiZEN I will have a look at these sites.
Dan
more audio quality theories...
oh yes, something else which may be working against solid-state designs (mainly those with class-B outputs) is non-monotonic distortion. Class-A designs should be relatively immune to this problem, so long as the driver stages don't contain anything too funky...
I have another theory which I've been investigating... focussing on open loop output impedence. This is very different from damping factor. If the open loop output impedence is non-zero, the load impedence affects the open-loop transfer function of the amplifier, prior to the application of feedback. You can see how open-loop output impedence could play an important role in how an amplifier sounds, and why OL output impedence also plays a critical role in amplifier stability (especially in high-feedback designs.. eg solid-state). With low feedback or OL designs, this problem is much less of a factor, perhaps contributing to better sound. It's interesting to note the parallel between the subjective sound of higher-feedback tube designs and SS amps.
If we don't want to reduce the feedback, the only option is to reduce the open-loop output impedence as much as possible, something which seems to have escaped the attention of many SS amp designers. Notably, the many paralleled output devices used in Pass amplifiers have the effect of reducing open-loop output impedence, in addition to providing more output capacity...
My most recent test designs all feature very low open-loop output impedence, and i'm very pleased with the results so far... In addition, some of the circuit's i've tried happen to minimize memory distortion too, and my limited experimentation with this concept seems to indicate some subjective improvement. I'm eager to try more experiments with various high-power circuit configurations to help reduce open-loop output impedence further, with the goal of eliminating the output inductor ordinarily necessary for power amp stability.
oh yes, something else which may be working against solid-state designs (mainly those with class-B outputs) is non-monotonic distortion. Class-A designs should be relatively immune to this problem, so long as the driver stages don't contain anything too funky...
I have another theory which I've been investigating... focussing on open loop output impedence. This is very different from damping factor. If the open loop output impedence is non-zero, the load impedence affects the open-loop transfer function of the amplifier, prior to the application of feedback. You can see how open-loop output impedence could play an important role in how an amplifier sounds, and why OL output impedence also plays a critical role in amplifier stability (especially in high-feedback designs.. eg solid-state). With low feedback or OL designs, this problem is much less of a factor, perhaps contributing to better sound. It's interesting to note the parallel between the subjective sound of higher-feedback tube designs and SS amps.
If we don't want to reduce the feedback, the only option is to reduce the open-loop output impedence as much as possible, something which seems to have escaped the attention of many SS amp designers. Notably, the many paralleled output devices used in Pass amplifiers have the effect of reducing open-loop output impedence, in addition to providing more output capacity...
My most recent test designs all feature very low open-loop output impedence, and i'm very pleased with the results so far... In addition, some of the circuit's i've tried happen to minimize memory distortion too, and my limited experimentation with this concept seems to indicate some subjective improvement. I'm eager to try more experiments with various high-power circuit configurations to help reduce open-loop output impedence further, with the goal of eliminating the output inductor ordinarily necessary for power amp stability.
"If we don't want to reduce the feedback, the only option is
to reduce the open-loop output impedence as much
as possible, something which seems to have escaped the
attention of many SS amp designers. Notably, the
many paralleled output devices used in Pass amplifiers have
the effect of reducing open-loop output
impedence, in addition to providing more output capacity..."
Some designers avoid multiple output devices, as they feel
these compromise sound quality. For instance, Naim have
always used single output transistors (hugely over-specified
in their new NAP500!). This may be one reason why many
prefer the sound of relatively low-powered amplifiers.
If this were true, it would seem to counter the output
impedance argument (although I can certainly see sense
in the latter).
Any comments?
Alex
to reduce the open-loop output impedence as much
as possible, something which seems to have escaped the
attention of many SS amp designers. Notably, the
many paralleled output devices used in Pass amplifiers have
the effect of reducing open-loop output
impedence, in addition to providing more output capacity..."
Some designers avoid multiple output devices, as they feel
these compromise sound quality. For instance, Naim have
always used single output transistors (hugely over-specified
in their new NAP500!). This may be one reason why many
prefer the sound of relatively low-powered amplifiers.
If this were true, it would seem to counter the output
impedance argument (although I can certainly see sense
in the latter).
Any comments?
Alex
Actually, i think it works in favour of the output impedence argument... In general, multiple output device stages require emitter (or source) resistors to keep quiescent currents balanced between transistor pairs. So, if one were to use single, very large output devices, it is entirely possible to reduce the open-loop output impedence by eliminating the need for these current-sharing resistors...
One salient aspect of low-power amps is that the magnitude of the most difficult problems is drastically reduced (usually by the square of output amplitude or more). So, I think it is a much easier proposition to design a good low-power amp than a high-powered amp. This, perhaps even more than output impedence, works very much in favour of the low-powered amp. To list a few of the advantages:
- lower output currents (for greater linearity)
- lower rail voltages (more robust outputs)
- lower slew rates
- more feedback available (similar open-loop gain)
- less thermal stress
- power supplies can be proportionally bigger
etc. these are just a few that come to mind, but all are potentially very important to sound quality.
Anyway, my theory basically lies on the premise that it seems desireable to make the operation of the amplifier as independent as possible from the load, and this includes not only the measured output of the amp, but also what's happening inside the amp circuit... eg, what kind of error signals are floating around in there, and how is the load affecting the currents, voltages and thermal stresses being seen by components inside the amp circuit?
One salient aspect of low-power amps is that the magnitude of the most difficult problems is drastically reduced (usually by the square of output amplitude or more). So, I think it is a much easier proposition to design a good low-power amp than a high-powered amp. This, perhaps even more than output impedence, works very much in favour of the low-powered amp. To list a few of the advantages:
- lower output currents (for greater linearity)
- lower rail voltages (more robust outputs)
- lower slew rates
- more feedback available (similar open-loop gain)
- less thermal stress
- power supplies can be proportionally bigger
etc. these are just a few that come to mind, but all are potentially very important to sound quality.
Anyway, my theory basically lies on the premise that it seems desireable to make the operation of the amplifier as independent as possible from the load, and this includes not only the measured output of the amp, but also what's happening inside the amp circuit... eg, what kind of error signals are floating around in there, and how is the load affecting the currents, voltages and thermal stresses being seen by components inside the amp circuit?
I think many of these discoveries (that single transistor / many parallel transistor / high bias current / single-ended -output stages sound better) may simply be serendipitous, because no-one seems to have come out and pointed the finger squarely at open-loop output impedence. Yet, in all of these cases, it can be seen that investigation will likely reveal that there is a lower open-loop output impedence than the alternative. I find it very appealing that all of these cases can be explained with one easy-to-understand theory.
Dan,
I finally stole enough time to read the article you mentioned. Unlike many on this site, I've long been a fan of tubes as they do things that solid state simply can't--for whatever reason. I've heard a lot of reasons postulated for the differences between tubes and solid state over the years, but this is the first time I'd heard anyone bring up the difference in quality of the Miller capacitance. However, as soon as you mentioned it, I could see the direction Olsen was going and it made perfect sense. Now, whether it actually is an audible effect, I don't know, but it certainly is an attractive concept.
C Simpson's points I find unconvincing:
--Compensation caps. One, tube circuits frequently don't have them, so strike one up for simplicity. Two, it's not a question of the quality of the compensation cap, it's a question of the quality (or lack thereof) of the capacitance in the gain device; bypassing an electrolytic with a film cap still isn't as good as a bank of pure film capacitance.
--Feedback as a cure. The use of feedback is rather a Faustian bargain. Sure you get some good things up front, but you pay in the end. The more feedback you use, the more dry and lifeless the music becomes. This has been known for years in contexts ranging from subjective assessment of the sound quality to the TIM/slew rate thing.
--MOSFET capacitance. Sure MOSFETs have capacitance. So do bipolars and tubes. Again, the question Olsen raises isn't the quantity--it's the quality.
Hoffmeyer's point about cascodes may or may not be applicable, as it's based on where the signal enters the gain device, i.e. base/gate/grid, or emmiter/source/cathode and how the circuit's used. It's not that the capacitance goes away, it's that you're trying to render it irrelevant. I'll have to think about this after I've gotten more sleep. In any event, not all circuits use cascodes or are able to. I don't know that I could say that I've ever noticed a 'cascode sound' that was identifiable as consistently better. For all that the topology has theoretical benefits, it doesn't seem to sound any better in practice. For what it's worth, the differential circuit also accomplishes the same end, but only if you take the output from the second (uninverted) side.
In summary, I think Olsen's point has possible merit. I'll roll it around in the back of my mind over the next few days.
Grey
I finally stole enough time to read the article you mentioned. Unlike many on this site, I've long been a fan of tubes as they do things that solid state simply can't--for whatever reason. I've heard a lot of reasons postulated for the differences between tubes and solid state over the years, but this is the first time I'd heard anyone bring up the difference in quality of the Miller capacitance. However, as soon as you mentioned it, I could see the direction Olsen was going and it made perfect sense. Now, whether it actually is an audible effect, I don't know, but it certainly is an attractive concept.
C Simpson's points I find unconvincing:
--Compensation caps. One, tube circuits frequently don't have them, so strike one up for simplicity. Two, it's not a question of the quality of the compensation cap, it's a question of the quality (or lack thereof) of the capacitance in the gain device; bypassing an electrolytic with a film cap still isn't as good as a bank of pure film capacitance.
--Feedback as a cure. The use of feedback is rather a Faustian bargain. Sure you get some good things up front, but you pay in the end. The more feedback you use, the more dry and lifeless the music becomes. This has been known for years in contexts ranging from subjective assessment of the sound quality to the TIM/slew rate thing.
--MOSFET capacitance. Sure MOSFETs have capacitance. So do bipolars and tubes. Again, the question Olsen raises isn't the quantity--it's the quality.
Hoffmeyer's point about cascodes may or may not be applicable, as it's based on where the signal enters the gain device, i.e. base/gate/grid, or emmiter/source/cathode and how the circuit's used. It's not that the capacitance goes away, it's that you're trying to render it irrelevant. I'll have to think about this after I've gotten more sleep. In any event, not all circuits use cascodes or are able to. I don't know that I could say that I've ever noticed a 'cascode sound' that was identifiable as consistently better. For all that the topology has theoretical benefits, it doesn't seem to sound any better in practice. For what it's worth, the differential circuit also accomplishes the same end, but only if you take the output from the second (uninverted) side.
In summary, I think Olsen's point has possible merit. I'll roll it around in the back of my mind over the next few days.
Grey
I guess my point was that with careful design, the miller capacitance effect can at least be reduced or circumvented in some cases, and that IMHO there are greater gremlins to be dealt with in the solid-state world. I am certainly not contesting the obviously superior characteristics of tubes in this regard, nor that there may be an audible effect. Indeed Lynn's hypothesis brings the quality of capacitance to the fore, and both MOSFETs and BJTs have horrid capacitive characteristics.
As far as feedback goes, I used to believe that high feedback = bad sound too. But, my latest experiments just havn't supported this belief, and i've been forced to re-evaluate my opinions. I think it's all too easy to assign bad sound to the presence of feedback without taking the time to wrestle with the mechanisms behind it. So far, I havn't heard many (any?) arguments WHY feedback can lead to poor sound... it's always stated as self-evident, then dismissed as some mysterious effect we can't explain, and given no further consideration...
I postulate that there is some specific mechanism(s) behind this effect, and that this mechanism interacts with or manifests itself through feedback, leading to a subjective decrease in sound quality. I further hypothesize that if we can identify and eliminate this mechanism, then feedback will no longer be the sonic culprit people make it out to be...
The error correction concept seems perfectly valid to me, so we need to investigate why it doesn't always work the way we want it to and attack the problems if we're ever to come to some understanding and advance the art and science of solid-state amplification. So far, the open-loop output impedence explaination seems the most plausible to me, because it introduces the potential for error signals which are not consistent across frequency, amplitude and time, very probably leading to a constantly changing distortion spectra, which is then percieved by the human ear as artificially reproduced sound. Just my theory... criticisms and further analysis welcome...
As far as feedback goes, I used to believe that high feedback = bad sound too. But, my latest experiments just havn't supported this belief, and i've been forced to re-evaluate my opinions. I think it's all too easy to assign bad sound to the presence of feedback without taking the time to wrestle with the mechanisms behind it. So far, I havn't heard many (any?) arguments WHY feedback can lead to poor sound... it's always stated as self-evident, then dismissed as some mysterious effect we can't explain, and given no further consideration...
I postulate that there is some specific mechanism(s) behind this effect, and that this mechanism interacts with or manifests itself through feedback, leading to a subjective decrease in sound quality. I further hypothesize that if we can identify and eliminate this mechanism, then feedback will no longer be the sonic culprit people make it out to be...
The error correction concept seems perfectly valid to me, so we need to investigate why it doesn't always work the way we want it to and attack the problems if we're ever to come to some understanding and advance the art and science of solid-state amplification. So far, the open-loop output impedence explaination seems the most plausible to me, because it introduces the potential for error signals which are not consistent across frequency, amplitude and time, very probably leading to a constantly changing distortion spectra, which is then percieved by the human ear as artificially reproduced sound. Just my theory... criticisms and further analysis welcome...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.