[Patrick Turner]

Phil wrote:

Andre Jute wrote:

In an effort to be agreeable, I tried hard to give you negative
feedback inside the tube as an explanation of the overwhelming
superiority of triodes (or trioded pentodes) for audio reproduction,
among other reasons because NFB is accessible to many who belong on RAT
and is a genetic deformity of the silicon scum whose only purpose on
RAT is dissension. NFB is what the silicon slime abuse to make their
inadequate components sound passable, and what even tubies inspired by
the age of sophisters and cost-accountants use to linearize pentodes.
NFB thus has a base level of familiarity which gives it a head start in
any black box model intended to explain something to diplomaed
quarterwits among the silicon slime as well as the better-educated
kibbitzers in my own camp.

But fine, you want to reject my explanation, then you must offer a
better reason to explain why triodes are such superior amplification
devices to anything else, so much more pleasing to the ear, so much
more accurate to the cultivated taste.

Despite my cracks about the metaphysics of tubes, there *has* to be an
electrical reason for the superiority of triodes. But sure, all kinds
of input is welcome.

Andre Jute

Andre, I think you are too quick to dismiss the idea that feedback
transforms amplitude distortions into phase-smearing, like Otala claimed
in his talk/paper (assuming a paper ever followed the talk!). When
Patrick was defending feedback, he mentioned that poorly designed amps
sound like crap when they have lots of feedback, but that if you fix
them up a bit, meaning get rid of much of their excessive nonlinearities
-- read, *amplitude* nonlinearities -- then adding feedback sounds okay.
Now, if we eliminate the idea that feedback produces random noise -- and
we *know* that it reduces amplitude non-linearities -- then the only
distortion mechanism left, I believe, is phase-shifting. However, I want
to describe the usual "constant 20 degrees phase lag at 40 KHz" as
"phase-shifting," and a dynamic, microsecond to microsecond shifting
back and forth of one frequency relative to another as "phase-smearing,"
or "time-smearing."

What Otala was saying is that applying feedback to circuits with lots of
open-loop distortions, which are (I believe) almost always amplitude
distortions, converts these distortions into a back and forth smearing
of the high frequencies relative to the low frequencies (and it may
smear both in terms of the delay through the amp). Nor can we assume
that this time-smearing is a simple function of the low frequency
amplitude, because it is probably proportional to the magnitude of the
distortions, as well as the LF amplitude: VERY non-musical. When enough
feedback is applied to badly designed amps, the amplitude distortions
become quite small, so why did Patrick find that they sounded much worse
than the same amp sounds when touched up enough to reduce the larger
open-loop amplitude distortions? As you say, *something* is wrong, and
if it isn't high amplitude distortions (it can't be), and if feedback
doesn't produce spurious noises, then the only thing left is exactly
what Otala said, time-smearing.

In essence, feedback *connects* two things that are normally separate in
an amp, namely amplitude distortions, and phase-smearing. It achieves a
balance between these two, a balance which is determined by the speed of
the amp, the amount of feedback, and the amount of amplitude distortion.
Contrary to what you say, Otala was *not* referring to TIM, and
transistors did not become so much faster after 1980 than the ones used
in his '73 article to make the problem he described go away. Yes, the
amps had to be designed well enough to avoid TIM, but that was not a
real problem even in '73 *if* you knew what you were doing. The problem
he described in '80 was quite different, and even high MHz tubes are
subject to it.

The interesting things, assuming that feedback problems are indeed
time-smearing (regardless of whether this comes from the conversion of
amplitude distortions), are one, a single tone will reveal nothing of
this, giving very low THD numbers, and two, multiple tones should show
something, although looking at it in the amplitude realm will only show
higher than expected IMD. There should be a fairly easy way to test to
see if this really produces time-smearing. In general, we put a 4 volt
60 Hz signal and a 10 mV 20 KHz signal into an amp with lots of
feedback, preferably using non-linear sections of the amplifying devices
(say, use two 12AX7 type triodes where the plate curves vary from widely
spaced to closely spaced, and use about 60 to 80 dB of feedback to get
the overall gain down to 1). Use a high pass filter to see only the 20
KHz signal, and use the 20 KHz signal from the generator to trigger the
'scope. If time-smearing exists, then the 20 KHz signal will appear
"fuzzy" on the scope, since it is being shifted back and forth. Repeat
without the 60 Hz signal to make sure the time-smearing isn't coming
from somewhere else, and then also divide the output from each tube down
so that you get the same amplitude output *without* using feedback, and
see if the time-smearing goes away. Finally, if possible test over a
region where the curves are fairly linear, to see if that also reduces
the time-smearing. If so, you have proof that feedback transformed the
amplitude distortion into time-smearing. I will try to do this, but I
have little time, less energy, and not the best test bench in the world,
so it may be a while! If you or someone else both can and wants to try
this, I suspect we will get an answer much faster than if I do it. Note
that we can use pentodes or transistors, too, but using solid state has
the disadvantage of possibly introducing other complications due to the
poor quality silicon parasitic capacitances. Also, I am suggesting the
use of very different magnitudes as well as frequencies for the two
signals, in case equal magnitudes somehow avoids this problem, or at
least masks it from this test. And again, contrary to what Patrick said,
the idea that *if* this were true, then by now someone would have
already tries it, and the results would have become widely know, is just
naive. The human race simply isn't that intelligent, at least not yet.

An interesting conclusion, assuming this feedback time-smear mechanism
exists, is that the output of a feedback amp does *not* match the input
when multiple signals exist! However, our normal tests cannot see this,
since they tend to focus on one frequency at a time. I believe that
Patrick, as well as Otala, said that using local feedback to achieve
good open loop linearity, combined with some global feedback, tends to
sound pretty good. This makes sense for two reasons, first because
although the local feedback will produce phase-smearing, it does so at
*very* high speed, which I believe directly reduces the amount of
smearing (common sense says that as the speed of devices approaches
infinity, feedback becomes "perfect"). Second, the result does contain
some mis-match between input and output, even given the high speed of
local feedback, but now the global feedback will not only reduce the
remaining "normal" amplitude distortions, it should also reduce the
time-smearing produced by the local feedback, since this time-smearing
will still produce an error signal for the global feedback. Of course,
this reduction in time-smearing will itself produce more time-smearing,
but it should be a case of 0.1 x 0.1 = 0.01, so the final degree of
time-smearing is reduced by "dividing up" the total feedback into local
plus global.

Finally, you saw the review of Otala that Phil Allison gave, and I think
you were as impressed by it as I was (although it looked bad to PA).
When a man who was as talented, knowledgeable, and honest as Otala
produces a PROOF that negative feedback *always* transforms the
amplitude distortions of the open loop into phase smears of the closed
loop, we should not dismiss it as simply a problem of "old devices," a
problem that a slight increase in speed can make go away, especially
when the discussion that led to his analysis (the Audio Critic BS
session) included a lot of talk about vacuum tubes, and how their speed
advantage made it easier to use them with feedback. We really should do
at least one or two tests before dismissing his conclusions. Again, I
will try to do so, but if you want to know anytime soon, you should
probably rely on someone else.

Phil

I have seen no evidence of the time smearing you are talking about.

The essence of time smearing you speak of when testing say 5kHz
with 60Hz present as a larger signal is that the 60Hz affects the devices as
a changing
reactance load on the devices so that the phase of the 5kHz waves are phase
advanced
and phase lagged alternatively 60 times persecond.

Phase modulation and FM modulation was gained deliberately in
reactance tube modulators which exploited the change in gm with Ia in
a tube thus shifting the F of an oscillator or the phase of an RF carrier.
There is much about thei is old books.

But dynamic phase shift of a fraction of a 5kHz wave I have not seen due to
dynamic action by a lower F.
The NFB reduces ALL artifacts and such phase shifting is reduced in the open
loop
character of an amp where it allegedly should exist.

I repeatedly gave the conditions needed where the application of NFB didn't
make any improvement
to the sound, and made little difference to the measured THD artifacts
especially when weighted
for audibibilty, and increased the number of artifacts significantly.
Far greater minds than I have spelled it all out in Wireless World years ago.

I suggest you read all your local university library archives containing the
magazine with its brilliant audio
articles between when it first appeared in 1917 to now.
( I assume your local uni isn't full of football magazines in the archives ).

I have only read and copied out the audio stuff up to about 1996.
In all of this literature on the effects NFB there wasn't much about
FB causing dynamic phase shift that I can recall.
Perhaps if you read what I read you'll find something I missed.
Seriously, methinks you NEED to do some real study.

Perhaps you'd like to re-iterate what I said to all about open loop bw,
phase shift, and the amount of applied NFB and the amount of open loop THD.

The fuzziness you say is observed when viewing the HF wave in a cascaded
12AX7
amp where the gain has been reduced to 1 is hard to believe.
A cascaded pair of 12AX7 would have an open loop gain = at least 3,000,
and applying FB to reduce gain to 1.0 = 70dB of applied NFB.

Any dynamic phase shifting in any signal before FB is applied should be
easily
visible / measurable / quantifiable before NFB is connected,
but after a reduction of 70dB it would most definately be invisible on a CRO.

Do offer my sincerest respects to Mr Otala and ask him to
tell you all about what you appear to maybe not understand as well as he may.

I suspect you instead remain delighted
by the jargon and terminology around the subject rather than staying
with the cold hard facts about applied NFB and its effects and the conditions
under which it is applied,
all of which cannot easily be dealt with without being
utterly precise at all times, which seems right considering the books and
magazine articles
which have been written about NFB so far.

Patrick Turner.