Gents, maybe you'll permit me to take a sightly different approach.

It seems to me that the last question posed by Anaxilus is really not a proper question. It's kind of like asking what happens when an irresistible force pushes on an immovable object. The question is framed such that you really can't answer it. But, I think it's possible to answer sub-questions that might illuminate something.

I think the first thing is that the active devices don't stand alone in a circuit. It's not just that there are other passive components around them, it's also that they are in a particular circuit topology. This circuit topology has as much to do with the sound as do the devices themselves.

As someone pointed out BJTs are current controlled devices, FETs are voltage controlled, tubes are voltage controlled. In fact, tubes are the hollow state analog of depletion fets. In an enhancement FET the gate must be positive (for an N FET) with respect to the source. In a depletion FET the gate must (in general) be negative. In a tube the grid must be negative (for normal operation) with respect to the cathode.

Each device type, to some degree, dictates parts of the circuit design.

For the moment, let's assume that you're building an amp entirely using one type of device, BJT or FET or Triode. Greatly simplified thoughts.

For BJTs some elements of the circuit have to push current in and out of the BE junctions. In addition, if you want the devices to be turned on at the quiescent op point, then you must have some BE idle current. This dictates certain conditions and driving circuitry. For example, with very high power output BJTs a lot of base current is required. If you're pushing peak 10A and the hfe is only 100 you need peak 100mA into the bases. The drivers for these transistors have to be selected and configured for this.

Since the BE junction is a forward biased diode it typically doesn't exhibit a lot of capacitance. Thus, you don't necessarily have to account for significantly increased loading as the frequency goes up.

BJTs exhibit thermal runaway. They also operate happily at fairly low voltages.

The characteristics of the devices, to some degree, already set some conditions on the design.

Mosfets don't require gate current to bias to an operating point. But, they can have large gate/source and gate/drain capacitances. In high current or high voltage mosfets these capacitances can be fairly large, large enough to be relevant in the audio band. What this means is that whatever is driving the gate may not have to push much charge at low frequencies, but quite a lot at high freqs. The driver stage has to be designed to handle this capacitative load. And if this is a totally FET design, then there are capacitors everywhere.

If using enhancement FETs (by far the most common today) then there has to be a positive (negative for P) gate/source voltage to establish an idle current.

These characteristics dictate, to some degree, the circuit design.

Tubes are high voltage, low current devices. The grids are negative with respect to cathode. The are no P channel tubes because holes don't exist in a vacuum. We can't build complementary circuits. In general tubes exhibit fairly low capacitances because their elements are far from each other and there is no dielectric between them.

These device characteristics dictate, to some degree, the circuit design (e.g., the need for output transformers to achieve high output currents).

It is rare, in my experience, that you can take an amp circuit designed solely for one of these types of active device and simply wholesale replace all the active devices with another type. The only situation that I can think of is where you could put depletion fets in for tubes assuming they can stand the voltages and would achieve similar quiesc ent currents. Maybe someone knows of another situation.

So, why all this? Well, if we were comparing three amps, each made entirely with one type of active device, we are comparing not just the behavior of the devices but the specific circuit topologies needed to make them work as a practical amplifier. MHO is that these circuit topologies are equally as relevant to the comparison. And because there are multiple, effective topologies for each type of device, the comparisons form a set of combinations which are hard to get a handle on.

OK, so what to do? We can take the purely external, black box approach and set some conditions on performance.

Any of us could design a 10W amplifier with any of these devices with 20-30% THD. If three of us did this and we did a blind test we might actually think that one sounded better than the other two. But, all three would still be junk.

Part of this question has to do with what we might call a "well designed" amplifier. There are decades of conversation, theory, and practice on this topic, although maybe not using this exact language. 

We can make an arbitrary definition for discussion. Let's say we want an amp which can deliver 10WRMS into 4R (resistive) with .1% THD at 1kHz and -1db from 20Hz to 100kHz and a gain of 25. This is not a difficult target, but let's call this a "well designed" amp. These are entirely externally measurable characteristics and don't tell us anything about what's inside the amp.

Our first problem is that all three of these active device types are non-linear except at very small excursions around operating points. So, our job is take these non-linearities and linearize them to achieve the "well designed" result. And, being whiz designers we do this and build our amps. Then we compare them.

Regardless of the result, a dominant part of the questions as to why one "sounds" better than the other two (if there is one) is not which particular type device is used, but rather what techniques were used in the design to linearize the gain and to achieve the bandwidth?

The techniques for doing this have been around for decades in one form or another. They include local NFB, global NFB, using CCSs for high dynamic resistances, reducing interstage loading, using diff amps for common mode rejection, etc. Most of the time these techniques are used in combination and in various places around the circuit.

When we listen to these three amps we are not just listening to the "sound" of the particular active device used, but also the "sound" of all the other circuit elements put in place to linearize them sufficiently to make our "well designed" amp.

Unfortunately, it's even more complex than this. The simple example assumes that each linearizing circuit element can be made from the chosen active device for that amp design. But, we have for decades mixed and matched. Tube amps commonly use CCS loads made from BJTs or FETs. BJTs and Mosfets are often used in combination. Sometimes all three types are in the same design. In these instances, the behavior of the active devices is mixed together. What is the sound of this??

In every instance, including so-called "money is no object" amps, there are always a set of design choices and trade-offs. Always. There is always a point where the designer has to say, ok this is good enough to achieve what I intended to achieve.

So, when asking what is it about tube amps vs SS amps, it seems to me you have be more specific. Like, what is it about this particular tube amp and its internal topology that is (apparently) more pleasing than this SS amp with its particular internal topology? If we ask that question, we might have a chance.

Or, you can follow this link in this post from cnpop e on diyaudio:

"For many years (and maybe still?), an audio engineer called Richard Clark ran an "Amplifier Challenge," where he offered a prize of $10,000 to anyone who could demonstrate in double-blind testing that they could reliably discriminate between any two amplifiers, subject to certain fairly basic conditions. The "challenging" amplifier and the comparison had both to have less than 2% THD at the levels at which they were operated; the signal levels would be carefully matched to 0.05dB accuracy; and if either had a significant deviation from flat frequency response in the audio range, then an equaliser would be used on one or other amplifier (challenger's choice) to match them. Nobody ever won the prize, although of order 1000 have tried.

Typically, as I understand it, if the challenging amplifier were a tube amplifier then Clark would make a little R/C equaliser to match the solid-state comparison amplifier's frequency response to the tube amplifier, and maybe add a series resistor at the output of the SS amplifier to match the output impedance of the tube amplifier. So effectively, by this means, he would mock up the performance of the tube amplifier at the cost of about $5 worth of components, he reckoned.

Different people will place different interpretations on what this shows, I'm sure. To me, it would seem that if no one can hear the difference between the real tube amplifier and the mocked-up one in blind listening tests, then the sounds are the same.

Chris"

More details here: http://tom-morrow-land.com/tests/ampchall/rcrules.htm



Many things have been left out so this post doesn't get too long.

Because it seems that when we try to just treat the device as a "black box", they more or less don't reveal the differences that we're hearing. Or at least within what many consider to be audible thresholds.

Nah, under certain conditions I believe those "black box"-es will reveal their strengths and weaknesses... In the case of "tube sound" however, one has to be specific because more than likely different "tube" amps along with their circuit topology will likely sound different form the next... In other words, one should not generalize something so diverse.

Right now, it seems the best we have to go on is whatever design quirk the amplifier seems to have, specifically with regards to the output stage. SET, DHT, Class-A, Class-AB, Class-D, BJT, MOSFET, OTC, OTL, cap coupled, whatever. This makes loose sense to me I guess, because it means putting different loads on that output topology can yield different results in how the load interacts with what's driving it. But what's actually happening and how can it be characterized? I don't know. And I fully accept the possibility that we're all just hearing shit.

Yup, I guess that is the question.

@runeight

AWESOME!!!

It makes sense that different active elements (TUBES, BJT, FET, MOSFET... - and among them different sub-classes - and combination of all of the above) will require different circuit topology approaches given their distinctive characteristics. It may follow that depending on the approach selected, each solution will exhibit different characteristics (some more audible than others).

Among the many goals behind these designs, amplifier "transparency" (relative to the recording) has been among the most sought after by some (but not always). Based on what I've heard from a simple Valhalla + HD600, S7 + HD800, CTH + Paradox, among others, I feel tube amps can fulfill this "transparency" goal to great success.

However, like you said, there are always trade-offs (NSTAFL) and while it may be possible to design an amp that will perform great with a healthy range of headphones, it may not necessarily be the best for each and everyone. I believe this is the case for the Valhalla, the fabled O2, and pretty much everything else "tube" or not. Furthermore, other requirements may come into play (SWAP, $, manufacturability, portability, ...) which may narrow down the choices.

Back to no-feedback, non-inverting, Class-A architectures (which I think the Valhalla is a good quality example off), Solderdude mentioned that linear behavior is difficult given the lack of FB. However, the Valhalla seems to have pulled a decent job with the HD600 which has a r elatively large impedance. What is it about such architecture that seems to perform well with some high impedance/hi-fi cans, but no so with lower impedance/lo-fi ones?

Thanks Marv!

The answer to this question is already given, but it could easily be missed or dismissed on account of 'that can't be it'.
It has to do with the type of distortion and how our brain interprets the added harmonics (perception).

The perception bit is responsible for the complete dissonance (below certain levels) between measurements and perceived differences.
I am fully aware some will say, ah but we all hear the same (or use the argument we do not) and it reproduces so there must be something we cannot measure or do not know how to yet. 

I agree that is sometimes perception is indeed the answer. There are such things as frequency and temporal masking that are sometimes exploited in audio compression with success. I believe there is also a limit as to how many frequencies we can perceive at the same time and so forth.

However, in the specific case of the Valhalla, the distortion data seemed to back up the high-quality audio quality perception when using a relatively high impedance headphone such as the HD600. Similarly, measurements showed that perceived audio quality degradation when using a moderate impedance headphone (HD558) correlated with a significant increase in THD distortion across the whole frequency range.

In fact, driving the HD600 with a Valhalla amp was perceived as better sounding than an HD558 driven by a lower impedance amp such as a Focusrite 2i2. Distortion results showed that the HD558 driven by the 2i2 had significantly more distortion than the HD600 + Valhalla combination as well.

So while it is true that psycho-acoustics are responsible to some degree for audio quality perception, in this case distortion numbers seemed to have correlated well with audio fidelity perception.