Yeah, I did not like the Vahalla with any low Z headphone. I even tried Grados, ATs, etc. High Z or into a power amp

I don't think tubes are more linear for smaller signals than transistors or other SS devices.

Both are very linear for small signals.
What one has to keep in mind though is that these small 'plankton' signals are superimposed on all the other signals (they are literally an addition of all the recorded signals together) thus when a large bass signal swings in a substantial amplitude that small plankton signal goes up and down in that high amplitude continuously in that low frequency.
Going from the 'top' (or high point) of the transfer characteristics to a lower point, or the 'bottom' even when nearing clipping levels.
That small signal may (will) have a different amplitude at the bottom, top or in the middle of that transfer characteristic and is thus somewhat modulated in amplitude by that larger signal.
The more linear the amp (more feedback and overall gain) the more linear that behaviour the less distortion as it were.
Triodes without overall feedback are anything BUT linear which kind of 'reduces' the theory of better linearity for smaller signals in validity at this point IMO. Also the transducer swings up and down and needs to 'follow' those small variations to make them audible. This takes a good transducer, certainly at higher amplitudes.

Also it highly depends on the used voltages. Tubes perform better at higher plate voltages (especially triodes) than in low voltage designs, certainly with larger voltage swings (which can occur at the same time huge ones exist).
This thus depends on the type of tube, operating point as well as voltage swing and amount of feedback (if used) and topology.
Grounded grid will behave vastly different than grounded cathode or cathode follower circuits with the same tube.

Tube circuits in general have higher noise levels because of the relative high resistances used (low currents) where as SS works with relatively low resistances in the circuits.

Let's all keep in mind that triodes and transistors (NPN at least) are fundamentally similar devices.

Tubes and transistors are not fundamentally similar at all.
The only thing they have in common is they are non-linear and can be used for amplification.
Tubes (and FET's) need an input voltage and have an output current while bipolar transistors have input currents and output currents which are fundamentally different and require different biasing techniques and different value components around them.
Both need biasing by the way and one is not more complex than the other just different as they work 'opposite'.
A tube and FET will conduct less current when a positive voltage is applied to the input (gate/grid) while the NPN wil conduct more current when a positive voltage(current actually) is applied to the base.

The 'plankton' and small nuances that are often mentioned are not very small signals by the way but  are FAR above the noise floor. The dynamic range of human hearing is much smaller than that of a DAC/amplifier for instance. For this reason the plankton (as it is 'clearly' perceived with a good set of transducers) is much higher in amplitude than most people realize.
The reason it becomes more audible might be due to compression (which happens in non feedback designs, especially tube designs) or because of high amounts of headroom. When there is little amount of (annoying) distortions around we actually tend to play music louder and thus also hear more things that otherwise would be below the audible threshold. Also transducers need to act linear and 'fast' to show this.

Could you expand on your comment that tubes are not linear without feedback?  I assume you mean as a general sense about a triode apart from a well implemented circuit as any number of valve amps are very linear with little to no negative feedback.  Or would 10hz to 65khz @ -1dB not count as 'linear' enough in the respect you mean?

We both are thinking of other linearities.
I am thinking of gain versus operating point/output voltage while you are thinking of frequency range 

10Hz to 100kHz would have to be no problem for tubes as well transistors (depending on the circuit around it)

What I mean is when you look at the V/I graphs of a tube you will notice the 'gain' varies with the DC operating point (as do transistors) even within their 'linear range'.
The closer you come to the extremes of that range the bigger the difference in gain opposite the gain around middle of that 'line'.
How linear it behaves depends on the supply voltage, voltage swing of the signal (that moves up and down around the middle (bias point) of that 'line' and topology and circuit around it.

FR is usually not a problem.

Feedback is used for a couple of reasons.
Two main important ones are better(wider) frequency range and lower distortion (less deviation from the original applied waveform in amplitude).
Stability, temperature dependency e.t.c. are other reasons.

Non feedback designs may have good frequency ranges (for both SS and tube depending on the used amplification devices) but some local feedback or 'gain control' will be present in the form of components surrounding it.
Some tubes already will roll-off in the audible range (in both ends of the spectrum) depending on voltages and components surrounding it others will have a very wide frequency range.

Non linear behaviour, in this case meaning, the output voltage does not have exactly the same gain at every point in a voltage swing, will be present though. Depending on what type of distortion this causes it may be more or less annoying audible or pleasant audible.
For tubes there is mostly compression which means flattening of the output waveform.
This shows itself in creating a certain spectrum of harmonics. Those of tubes are closer in 'nature' to those of instruments e.t.c. where those of SS have lower distortion in general but much higher and less 'nice' sounding much higher band harmonics are present.
These will be most audible in non feedback designs.
Feedback actually 'compares' the output to the input signal or should I say feedback 'controls' the amplification components so they behave more linear.

How much (amplitude of which order harmonics) of this is audible is the general topic of on-going debate.

In most cases the THD%+Noise plot vs voltage looks worse for lower impedances (higher distortion numbers in the operating region), which sort of agrees with the HD600 vs HD558 + Valhalla results. However, this is does not seem to be always the case (i.e. Beyerdynamic A1),

I'll give a hint about the A1.... 100 Ohm output R so at any time the load to this amp will be at least 100 Ohm even when the load is a short.
So it never sees a low impedance load.

All designs measure different, that doesn't mean it is related to SQ in any way though (below certain levels)

Indeed a heavier load (certainly a complex one, those of headphones aren't as complex as speaker loads) will give more distortion.
It will be compensated by feedback in a substantial way.
This is NOT the case with non-feedback designs (feedback reducing the distortion that is) and thus non feedback designs (such as the Valhalla) will exhibit higher distortion when the load get's heavier.

It's what Purrin heard as well I guess.
It is simply not well matched/designed for low impedance cans.
As Jason already mentioned by the way.
That doesn't mean some people might still like the Valhalla on low impedance cans which is what he refers to in his statement.

Quote from: ultrabike on August 08, 2013, 09:29:04 AM

In most cases the THD%+Noise plot vs voltage looks worse for lower impedances (higher distortion numbers in the operating region), which sort of agrees with the HD600 vs HD558 + Valhalla results. However, this is does not seem to be always the case (i.e. Beyerdynamic A1),

I'll give a hint about the A1.... 100 Ohm output R so at any time the load to this amp will be at least 100 Ohm even when the load is a short.
So it never sees a low impedance load.

All designs measure different, that doesn't mean it is related to SQ in any way though (below certain levels)

Indeed a heavier load (certainly a complex one, those of headphones aren't as complex as speaker loads) will give more distortion.
It will be compensated by feedback in a substantial way.
This is NOT the case with non-feedback designs (feedback reducing the distortion that is) and thus non feedback designs (such as the Valhalla) will exhibit higher distortion when the load get's heavier.

It's what Purrin heard as well I guess.
It is simply not well matched/designed for low impedance cans.
As Jason already mentioned by the way.
That doesn't mean some people might still like the Valhalla on low impedance cans which is what he refers to in his statement.

Hey Solder, this question might be moving into amp design territory but I must ask...

How does using NFB reduce distortion into low impedance loads? I understand the basic concept of using feedback as a form of "error correction". But you still see most amplifiers (NFB or no-NFB) have distortion increasing into lower impedance. Are we just talking about a degree of magnitude here?

I had always thought that higher THD into lower impedance was more of a function of the output devices becoming more non-linear as they're required to handle more current. They'd be operating in a different segment of their V/I curve as you said. Hence why you also see higher THD when an amplifier is required to deliver more power, even into high(er) impedance. It would seem to me that amplifiers that are more capable of acting like ideal voltage sources (2x Power into 1/2x R, 0 output Z, etc.) would behave better into low impedance loads. Which would, again, require an "overbuilt" power supply and adequate biasing. Where does NFB come into this?