LYNN OLSON:
THE GEORGE ZEN WASHINGTON
OF PUSH/PULL CIRCUITS
A COMPASS IN THE THERMIONIC STORM
Dr. Gizmo's Secret Agenda Revealed
You have been reading my articles and many of you are scratching your butts wondering if I am totally nuts because my views are so anti-orthodoxy. So I can think of no better way to make you comfortable doing the most difficult thing possible for a man which is to change your mind then to enlist the help of one of America's most esteemed thermionic savant who I have absolutely no influence over .who is three thousand miles away .and whose unique vision of the audio arts, which is expressed in his new Amity amplifier, will aggravate you make your eye twitch. This is a good thing, because there is too much mealy minded information being spewed forth about tube amplifiers. Study what follows very closely. This is not only a fascinating exploration of Lynn's amplifier, in his own words, but the best tutorial in the righteous future of push/pull amplifier design.
I asked Lynn to describe the features and benefits of his Amity amplifier. Of course you are going to surf to his web site for more information aloha-audio.com
Do you have any friends who love their push pull amplifiers? Be a friend and email this article. Print this article and study it in the potty, in your office, and in bed. You are gazing at the PRIMAL BIG BANG of a new aural matrix a new form of musical beauty.
1) Topology and active devices selected for absolute minimum
distortion, particularly with attention to reducing upper harmonics.
No pentodes, no beam tetrodes, no transistors, no MOSFETs, no IGBT's,
no nonlinear triodes such as 12AX7, 12AU7, 12AT7, or 6DJ8. All
devices selected for absolute minimum upper harmonics and
well-behaved and moderate proportions of 2nd harmonic. This tends to
favor mid-Thirties DHT's and early-Sixties high-transconductance
triodes.
The mid-Fifties favorites such as 12AX7 and 12AU7 replaced the
superior octal equivalents 6SL7 and 6SN7 during a period of universal
feedback, so the threefold degradation in THD went unnoticed. The
12AT7 has a kinked set of curves that were OK for TV video amplifiers
(which it is designed for), not so good for audio. The 6DJ8 is a
color-TV cascode RF amplifier and was never intended for audio. The
so-called subjective "detail" of this tube is simply a high
proportion of 3rd harmonic, which is very undesirable in a triode and
even more undesirable in a PP circuit, which cannot cancel even
harmonics.
2) Circuits that optimize the linearity of each active device. For
triodes, this means active, choke, or transformer loads, no
RC-coupling anywhere. RC-coupling, although essential for the
extended HF bandwidth required by global feedback, also degrades the
distortion of triodes anywhere from 2 to 4 times relative to active,
choke, transformer loads.
I personally favor transformer or choke loads in favor of active
loads due to simplicity, reliability, and avoidance of active-load
coloration. I have not had good luck mixing solid-state with
high-voltage vacuum tubes, and prefer the reliability and
predictability of all-tube circuits. In an all-tube circuit, you can
tell if it's working just by looking if it's lit up, and you can
debug just by swapping parts.
An additional advantage of no RC coupling between the driver and
output tubes is immediate recovery from overload. RC coupling usually
requires hundreds of milliseconds to recover the correct bias point
for the output tubes. This is great for guitar amps, not so good for
hifi applications, where instantaneous recovery is much more
desirable.
3) Zero feedback in either local or global circuits. Some would say
all triodes have large amounts of local feedback, but I would say in
response that the proportion of upper vs. lower harmonics undercuts
this. Feedback of any type has no effect on the proportion of the
harmonics, and simply changes their magnitude. The unique hallmark of
DHT's and good IDHT's is their absence of upper harmonics. This is
*not* true for other devices, such as the types mentioned in (1)
above. Feedback does reduce the absolute amount, but has no effect on
the *proportions* of the remaining harmonics.
Since the circuit is intrinsically linear, feedback is not required,
which sidesteps problems with phase margin and stability with complex
and nonlinear loads ... which pretty much describes all loudspeakers.
In particular, complex back-EMF currents from the drivers stop at the
plates of the output tubes, instead of being mixed at the summing
node with the incoming signal. This avoids load-dependent distortion
terms being generated in the feedback circuit.
4) Adequate driver design. Most commercial tube amps only have 1 or
2dB of headroom in the driver circuit, so the entire amp clips at
once. This leads to longer recovery times and exaggerating the
audibility of clipping. I prefer 3 to 6dB of headroom, so the driver
can retain its (voltage) linearity even when the output stage is deep
into clipping.
Perhaps more important is adequate current and low output impedance
in the driver. Much of the amplifier coloration is actually in the
driver, and is a result of not enough current to properly charge the
grids of the output tubes. I give Arthur Loesch credit for pointing
out that the "sound" of different DHT output is greatly exaggerated
by not enough current in the driver. With enough current, they become
more transparent sounding and begin to lose the characteristic
colorations they are known for. This would imply that grid current is
present during much of the duty cycle and is quite nonlinear. The
more current available and the lower the source impedance, the less
important this grid-current nonlinearity will be. (The same
considerations of linear current deliver apply with greater force to
transistor-amp design, where driver sections tend to limit the
performance of commercial high-end designs.)
5) Absence of rectifier switching noise. The noisiest circuits of all
are solid-state bridges driving large values of electrolytic
capacitors. (As found in almost all transistor gear and the DC
supplies for heaters and filaments.)
The commutation noise of the diodes shock-excites the RLC of the
stray L in the cap bank and the stray C in the power trans secondary.
The resonance of this tank circuit is typically anywhere from 4 to
20KHz and the Q's are large, anywhere from 5 to 100, depending on the
DCR of the caps. This is why paralleling large values of
electrolytics with "better", faster polypropylenes can frequently
result in worse sound. It is also the reason power cables are audible
... they act as antennas for the small Tesla coil that most power
supplies resemble. The supply radiates noise into the chassis, the
power supply B+ lines, the audio circuit, and the power cable. This
broadband noise can be filtered and shielded (at considerable
trouble), but it is much easier to eliminate the commutation
switch-noise right at the source.
Choke-fed supplies are much quieter due to the choke slowing down the
rate-of-charge of the main cap bank. I use a hybrid
choke-fed/pi-filter to minimize the shock-excitation of the main PS
choke (this tip directly out of the RDC IV).
In terms of ragged waveforms, solid-state diodes are the worst,
followed by Schottkey diodes and HEXFRED's, followed by conventional
tube rectifiers, followed by TV damping diodes, which are the
smoothest of all in terms of the AC waveform on the power trans
secondary. This, along with 2 amp peak current, is why I use them.
The 30 second warmup is just a bonus.
6) Capacitance is Everywhere. Most mid and HF colorations in
commercial amplifiers are due to unforeseen stray capacitive effects.
Circuit boards are generally made of low-quality dielectrics (poor DA
and DF), absorb moisture, heat up and dry out, and are microphonic as
well. If a circuit board is a necessity, Teflon is by far the best
dielectric, but it is very expensive and not very strong
mechanically. Point-to-point wiring has the great benefit of air
dielectric.
It should be remembered that *any* capacitor with a polarizing
voltage is a low-quality microphone ... much of the benefit of oils
are simply due to mechanical self-damping. This is why solid-state
components with no vacuum tubes at all are still somewhat
microphonic; the numerous power-supply bypass caps are microphonic,
and pick up ambient vibration (including buzz and vibration from the
internal power transformer). In addition to picking up ambient noise,
caps also self-excite, and create their own resonant noises.
The residual HF imbalance in PP transformers is due to capacitive
imbalance ... you have to get a "geometrically balanced" PP
transformer in order to get AC *and* DC balance. Even some very
expensive transformers are not balanced at 50kHz ... don't worry, the
trans in $20,000 commercial amps are a long way from being the most
expensive. They're usually made offshore to a very low price.
The Miller capacitance in power tube grids leads to various
slew-limiting mechanisms (current delivery from the driver becomes
nonlinear), and the Miller C in the driver tube unbalances the
traditional split-load inverter at high frequencies. The
common-cathode long-tail inverter has unbalanced Miller C on its own
inputs; one side is a grounded-cathode, while the other side is
actually a grounded-grid circuit. The various tube phase-splitters
need to very carefully analyzed to see if they are balanced at high
frequencies.
7) Symmetry is Everything in PP. One of the most difficult aspects of
PP, or differential, or bridge-mode, or whatever you want to call a
balanced circuit, is the cancellation of even harmonics. Note this is
cancellation of internally generated distortion, not the music
itself, as a writer asserted in a magazine some time ago. It would
take a *very* sophisticated DSP system to cancel the harmonics in
music. Instead, what analog amplification does is add *additional*
harmonics that don't belong there.
The problem with PP harmonic cancellation is that it isn't stable in
most PP circuits. This is a serious issue, since it means the
harmonic "signature" of the amplifier is unstable with signal level,
frequency, and power-line conditions. This is the reason for the
opaque and cloudy sound of complex PP amplifiers, and the much more
direct and immediate sound of SE.
Any time there is a summing node (as is the case with PP and
feedback) the two circuits that drive the summing node have to be
examined very carefully to see if they are drifting away from the
ideal textbook condition. With PP, this means the phase-splitter for
the two PP "halves" has to be fully stable; as far as I can tell,
only transformers meet this requirement. It also means that when two
PP circuits are cascaded, the PP output of the first stage needs to
be re-combined before passing it on to the next stage, otherwise gain
mismatch will build up. Remember, these mismatches are dynamic, and
very undesirable.
This is what happens in the traditional RC-coupled Williamson, for
example, and is a subtle advantage of the simpler Hafler/Dynaco
circuit, which does not pass on gain errors from the driver to the
output. Instead, the large impedance mismatch of the split-load
inverter is directly presented to the output tube grids. (No driver
stage at all!) Out of the frying pan into the fire.
The best solution is to use a transformer to re-combine the outputs
of the first PP stage before presenting it to the second set of PP
grids. This removes the mismatch of the first stage, preventing it
from being compounded by the second stage. The IT needs to
geometrically balanced, and the presence of an IT also means that the
first (driver) stage needs to have a low plate impedance so the trans
will have adequate bandwidth. Unfortunately, this rules out a PP 6SN7
driver, which is the obvious choice for linearity and large signal
swing. The alternative choices are NOS 5687, 7044, 7119, and the new
JJ Electronics ECC99 and Sovtek 6H30, which all have pretty good
linearity and 3 times lower plate impedances.
Another awkward aspect of PP DHT amplifiers is the difficult
requirement for very large, very linear grid drive voltages. Since
the output DHT's are so linear (less than 0.25% distortion at 16
watts for PP 300B's), the drivers need to be equally good, or
preferably better. It's poor design practice to let driver distortion
predominate over output distortion. In my circuit, I bias the VV32B's
at 85V each, which translates to a requirement for 120V rms between
the two grids for full output. If I used a step-up transformer, that
would also increase the current requirements on the driver stage in
direct proportion, which is not desirable, since the driver needs to
have high current in the first place to minimize DHT coloration and
minimize slewing. Another undesirable aspect of a step-up transformer
is that the impedances are multiplied as the *square* of the
turns/voltage ratio, so a 1:2 step-up would present the paired 300B
grids with *four* times the Rp of the single driver tube (30K for a
single section of a 6SN7).
As mentioned above, I selected PP 5687, 7044, 7119, or Sovtek 6H30
drivers to maintain linearity in the drive stage. A more over-the-top
approach would be using the new Vaic Valve DHT drivers in PP, such as
the AV 5, AV 8, or AV 20. In addition to a significantly higher cost
(similar to the output 300B's), there is the tricky requirement for
DC (or heavily filtered AC) on the driver filaments. Unfortunately,
my experiences with regulated DC heating on DHT filaments has not
been positive, although I hear that DC *current* sources have worked
well for those who have tried them.
8) Subjective Tuning. With all the restrictions described above, what
if it doesn't sound good? What's left to tune? Well, there's always
the good old power supply, along with the traditional favorite, the
wire itself. Small changes in the corner frequency of the RC and LC
networks in the cathode bypasses and the power supply (in the 3 to 4
Hz region) can have a surprisingly large effect on the sound in the
300 Hz to 1 kHz region; this can alter the entire spectral balance of
the amplifier, even though there are no obvious circuit changes, and
essentially no change in the measurements.
--
Back to The Triode Guild© Back to Meta-Gizmo
e-mail Dr. Harvey "Gizmo" Rosenberg: drgizmo@meta-gizmo.com
Copyright© 2005 Meta-Gizmo.com and Dr. Harvey "Gizmo" Rosenberg All rights reserved.
All the material contained within the above articles may not be reproduced without his express permission.