Patrick Turner wrote:
Snip a bit,


The Williamson Theory

In 1950, Hi Fi bugs didn't know what "feedback" was.

But many ppl did know what it was, and how to use it in various ways.

Today, many ppl still don't know what NFB is, or how to use it.

Those people by definition are not schooled in the electronic art.

Any old theory of
feedback was an improvement over nothing. Williamson had phase
diagrams and curves to show that in order to place a stable feedback
loop around an amplifier, the problem was phase margins. In order to
assure stability, the output transformer should have a response from 2
to 200,000 Hz. This assured that the amplifier was stable in the audio
range of 20 to 20,000 Hz. This criteria had merit. Unfortunately,
Williamson did not carry his analysis far enough. Seemingly, he never
actually built a model of his amplifier. If he had, he would have
realized that there was more to it than an output transformer and
feedback loop. The only component in his design that met the 2-200,000
Hz. criteria was the output transformer.

This isn't correct. The original 1947 design had a 6SN7 3 stage voltage a=
mp
which
did have a bw exceeding 2 - 200 kHz. I know, I have built williamsons.

The other thing about feedback is that many folks used it with OPTs
and tube stages with maybe only 20 kHz of bw.
The simply used slightly less FB and used phase tweaker networks.

Hardly anyone in Oz purchased partridge transformers,
and nearly everything made in 1950 was crap with poor bw, but you could b=
uy
it,
use it, and it was cheap.




As presented, the Williamson
amplifier had a number of serious flaws. It was a good start, but it
was not thought out properly. When the author built a Williamson with
the Partridge transformer, he soon discovered these flaws. The
Williamson circuit and a critique will now be discussed.
The Williamson Circuit
We will discuss the circuit, block by block.

The Power Supply:

The Williamson power supply was built on a separate chassis. It had a
high voltage cord that connected it to the audio amplifier. Such
configurations aren't built any more.

Yes they are.

They are dangerous.

Not if its done properly.
Agreed.


Running a
400-volt DC power line for domestic use is against UL standards. The
seeming purpose of this set up was to isolate the power transformer
from the audio circuit (and output transformer). It has been shown
that there is no purpose in such a setup.

But its a good idea to have a separate power supply.
It makes the amps able to be lifted around easily.
A remote PS is good for noise.

And the two chassis can mount in a frame, PS on the bottom,
so its all one unit when set up.



The circuit contained a 5U4 directly heated cathode as a rectifier.
The circuit was of the capacitor-input type and contained two chokes, a
5 Henry power choke and a 30 Henry smoothing choke. The filter
capacitors were oil filled 4 mf.400 volt units.

This supply had serious deficiencies. The audio circuit had a
direct-coupled first stage. On warm-up, the B+ supply initiated high
voltage to the bus before the output tubes warmed up. This caused a
very positive voltage to appear on the grid of the second stage during
warm-up. This caused current to flow from the second stage cathode to
the second stage grid. The likelihood of grid damage was high during
warm-up as grid wires are very small and won't carry much current.

But the grid current initially flowing in the CPI stage was limited by the
cathode R of the stage. And it wasn't for a long period.
I have used SS rectifiers, the voltage comes up high within 3 seconds,
and sure that direct coupled grid goes high, bit it comes down
to about 100v ok after 15 seconds.


It's not challenging at all today to design a power supply without
tube rectifiers, that cycles the voltages needed for operation of the
amplifier in the proper order and moreover ramps each up in a
controlled, benign fashion. It is simple cookbookery, no engineering
per se is really needed. Most tube builders are electronically below
any reasonable fitness standard, however, and so it does not happen.

Such a power supply would be good for running slightly modified
vintage amps, test bench work, or running radio apparatus as well as
hi-fi.



When the Williamson was turned off, it fed a large very low frequency
power pulse to the speaker.

This slow pulse from the collapsing B+ voltage didn't last long.

Many Hi Fi speakers couldn't take this
very low frequency pulse and blew out.

Its no worse than running the amp to clipping with a signal at LF.


History records few if any such blowouts.


Any speaker that blew after one pulse from the amp
wasn't a hi-fi speaker; maybe an ex radio speaker.
I have had a real hi-fi 12" woofer get +65 v from the rail of an SS amp
when a mosfet shorted without a protection circuit.
I heard a bang, and a hum, and instinctively reached across and turned off
the amp.
The 100,000 uF rail cap discharged through the speaker.
No damage was done.

The qualification that a speaker is hi-fi includes the ability to take the
full power
of the amp for some specified time.
The williamson could only make 10 watts, and with a 37:1 OPT ratio, the
full 450 v B+ rail voltage translates to a max 12 peak volts that is
possible at the
speaker. It cannot be a long lasting voltage, as it can be with an errant
direct coupled SS amp.

Its a very low quality speaker than can be fused by a williamson turning
off.

The ones I built did not display the problems you speak of.


This turn off power pulse showed
that the power supply was unstable, poorly decoupled, poorly regulated
and prone to motor boating, a very nasty instability problem.

The williamson was only barely stable as originally presented.
And with a preamp with bass boost, and a phono stage
and with all the power taken from the power amp,
indeed the amp could oscillate at LF.

But where the preamp had its own supply, this didn't occur.


The
capacitor-input system rendered marginal the use of the first choke as
a filter element. Capacitor input systems defeat the advantage of using
chokes in a power supply.

Rubbish.

Cap inputs with chokes, then more caps make brilliantly good PS.


When the Williamson circuit was published, inexpensive electrolytic
capacitors were already available in the post war market.

They were regarded as unreliable.

The oiler caps were bullet proof.
High values were not needed, because the output stage stayed in class A.

Quad even used only 16 uF to anchor down the CT of their
class A output stage.

Oil
capacitors are very expensive per mf. and as obsolete as copper oxide
rectifiers. It was a poor choice for home use. Oil capacitors make
very poor power supply filters, for they lack enough capacity to do the
job. Electrolytic capacitors are much preferred in good designs, being
available in sizes to 500 mf. or more.

Today that all is the case and I routinely use 470 uF caps where once the
same space
was taken up by a 4 uF paper in oil cap.
I still use chokes in my designs, but they don't need to be such large
values as used
in 1950.

The oil caps will last far longer,as proven by the continuing
operation of much WWII and Korean military gear. However, the hobbyist
can always replace them himself cheaply, which is why photoflash caps
enjoyed such popularity in the Early Audio Amateur Era. Amps so modded
in the 80s should be recapped now.



Continuing our critique of the Williamson amplifier, we turn now to the
amplifier circuit. The amplifier is composed of two sections; the
"front end" or voltage amplification and phase inversion section and
the power output section.
The voltage amplifier and phase inverter:

This section or block is composed of a voltage amplifier, phase
inverter and driver. The amplifier is a double triode, a 6SN7, one
section direct coupled to the phase inverter. Most of the front ends
at the time used pentode tubes like the 6SJ7as amplifiers. These tubes
had good amplification, but high distortion. Williamson's all triode
amplifier set a new standard for the industry (unfortunately not
followed by the Dyna). There are really no problems with Williamson's
triode input. It was a clean amplifier.

Not utterly clean, some 2H was made by the SE input tube.
Pentodes usually make at least as much thd, but have a broader spectra.

It didn't matter because 90% of the thd produced by most tube power amps
is produced by the output tubes.
There is 20 dB of NFB.
This reduces **ALL** of the thd including the 1/10 of it from the input
stages
by a factor of about 0.1.
And because the output stage was a triode satge thd at 10 watts
was 1% with no loop FB and 0.1% with FB.
Dyancos and many other amps with UL, pentode, tetrode stages,
or class AB1 operation had thd at up to 5% without loop FB
so with FB they could only get down to around 0.5%.




The phase inverter is another matter. Because good design demands a
"push-pull" power stage, the output tubes must be fed by phase
inversion of the driver. Good design mandates that the driver has
certain characteristics. The drive should be balanced amplitude wise
and phase wise. The careful phase inversion is the most difficult to
achieve. The Williamson phase inverter was a split load phase
inverter. The plate and cathode resisters of the second section of the
6SN7 were matched at 47 Kohm resisters. This balances the amplitude of
the inverted signal (as long as the load resisters don't drift with
time), but there is a hidden serious flaw, not dealt with by producers
of the Williamson amplifier.

The Williamson CPI works fine.


The plate impedance and the cathode impedance are not of the same value
even though the load resisters are the same. This means that at high
frequency, the output of the phase inverter is no longer balanced.

But it only becomes unbalanced at above 20 kHz.

If you wish to extend the balanced bw, a 15 pF across the Rk of the CPI is
about
all you'll need.

The anode output of the CPI sags as F rises before the low Ro of the
CPI cathode sags.
A little compensation to the Rk thus boosts the output from the CPI
cathode.

Ppl didn't need to know this because W amps work without the compensation
at this point.


A
scope sampling the signal between the two driver signals shows the
discrepancy. This unbalance causes distortion. This distortion is
amplified by a negative feedback loop around the amplifier.

And the distortion gets reduced by a factor of about 0.1.

Its only important to reduce distortions below 20 kHz.

Above 20 kHz, it matters less, since the first possible harmonic
of say a 20 khz tone is 40 kHz, and we cannot hear that.

Some would say distortion is OK of signals above 10 khz.

If you have 5% thd of a 10 khz tone, and switch it on and off while
listening to a 10 khz squeal, you will hear no change.


If the
driver signals are tested with the feedback loop in place, the
unbalance is seen to be objectionably high, particularly at high
frequency. The poor phase inverter was the Achilles heel of the
Williamson circuit. Transient response, due to this defective phase
inverter is also poor.

I dispute this as well.
The open loop bw of the W amp was about 80 kHz.
With NFB applied, the bw could be 100 kHz.
But many W amps are now trimmed to go to 65 kHz,
and at full power into the rated load, and without any stage saturating
with
grid current.

The resulting HF response is fabulous, and a reason why tube amps are so
detailed and
fine sounding; they have no problems with musical transients.
It does depend on clipping never occuring, but ppl with hi-fi systems
never go near clipping.

Maybe they -shouldn't-. They do!



This type of distortion was not tested for at
the time the Williamson appeared. It is one of the reasons some claim
that negative feedback is "bad". Negative feedback is not bad if it is
around a clean amplifier. Negative feedback around a Williamson is a
mixed bag because of the flawed phase inverter.

Wrong, the W amp has a fast drive amp including the CPI.
Its bandwidth with 6SN7 was over 200 kHz.

Many other designs were slower, such as a mullard 520, which has a
****ant EF86 driving a 1/2 of a 12AX7 of an LTP,
and the response at the output of the 520 12AX7 anodes was much
reduced below what W did.
The CPI acts as a buffer between the SE input stage and the balanced drive
amp.
the CPI has its own large local amount of current FB.
To make a maximum of 30 vrms at the balanced amp output to drive the outp=
ut
triode grids,
only about 3 vrms needs to be applied to each grid of the balanced amp.
This is really easy for the CPI to do, even at HF, since the current need=
ed
to
charge and discharge stray and miller C is low, since the signal voltages
are low.


There is another flaw in the front end. Examination shows that there
are two sets of coupling capacitors in the front end. This means that
when negative feedback is applied, the amplifier becomes unstable at
very low frequency because of the time constants of the capacitors. At
very low frequency a phase shift occurs of over 180 degrees around the
loop and oscillation can occur. This is aggravated because of the
poorly regulated power supply. At the time Williamson wrote his
article, capacitors had inductance. This limited the high frequency
response of the front end.

Many W amps were made and worked as W said they would.

Many were. Many more were deviated from and still usually worked. In
the US no amateur wound his own output transformer. Most had a pet
builder whom they attributed Godlike powers to, such as Ercel Harrison
at Peerless, and would use their product over all others based on some
mantra. Much of the "secret wisdom" of these "gurus" is quietly sitting
on library shelves if one knows where to find it-a hint: The U.S. Navy
is the most ardent audiophile organization in history.

6SN7's have poor high frequency response
further limiting the high frequency response of the front end.

Wrong, a 6SN7 has a fairly decent HF capability; typical of many medium =
=B5
triodes.
You are confused; its the high gain high Ra types that have the limited b=
w=2E



The
front end began to roll above 30 Hz.

??

The point is that Williamson's
Partridge transformer was not of much use in this kind of amplifier.
200,000 Hz is well beyond the response of the rest of the system.

The Partridge was an item better had, than not had;
the OPT market was riddled with crap OPTs.

the OPT market is still riddled with crap OPTs, and nearly every major
maker
between 1950 and 1970 tried to cheapen and ruin OPT construction down to
crap, eg, Leak, QuadII, etc....



The output stage:

In the American version of the Williamson, two 807's, triode connected
were connected in push pull and fed into the primary of the Partridge
output transformer. The pair of tubes were cathode biased (together)
with an unbypassed common cathode resister. This arrangement cost
output power and high frequency response.

Wrong.
Correct! He's on crack.

The signal voltage across the common Rk of the W amp wastes almost
no power at all because this common cathode signal is so small.
Ever measured the signal power generated in the Rk?

While in class A the common Rk does not spoil the bw.

One can even use a CCS a s a common Rk, and the amp will work
as well as ever; but only in class A, as it it was meant to.


It is somewhat strange that
an engineer working for a tube manufacturer would recommend such an
output circuit. His company made KT66's, a beam power tube, ill suited
to be used as a triode. The 2A3, a power triode available at the time
was not used in the Williamson.

RTFM, Stan. The source document, available from UK and US sources,
specifies that the KT-66, 807 or 6L6 may all be used. US builders
usually followed Sarser and Sprinkle, who used the 807 strictly because
WWII surplus ones were cheap. Later surplus ones, by NOS standards,
still are. Triodes were not used for reasons clearly stated in the
source document, not for lack of them. The Brits had such jewels as the
PX series and the DA100.

But KT66, and 6L6 and 6V6 were very common tubes; they still are.

Brook began making an amplifier with
the 2A3 shortly after the advent of the Williamson. The Miller effect
(grid to cathode capacitance)

The Miller effect is the gain x capacitance between grid and anode, not
including to the cathode.
Triode gain for KT66 was low, so Miller C was low

reduced the high frequency response of
the Williamson amplifier as the 6SN7's had fairly high plate impedance
for a triode.

No, the 6SN7 was and is regarded as having a low Ra.
The data suggests 7.7k at 10mA of Ia, but at 3 mA,
its more like 10k, but that's low enough to allow the W amp to work ok.

12AX7 with Ra =3D 65 k was regarded as high Ra.


The 807's were fed with 400 volts on the plates, which allowed them to
put out 10 watts RMS. This was in line with what some others were
doing in Hi Fi amplifiers, but less than the 20 watts output generally
available with 6L6's pentode connected. The distortion in the 6L6's
was higher, but with feedback, the distortion was acceptable. The real
advance made by Williamson was the design of an all triode amplifier.
It inspired others to meet its distortion performance, (with a
resistive load) poor though the Williamson was.

The input to the output stage had 1000-ohm suppressor resister in the
grid circuit. It also had a resister in the screen circuit, but the
screens were not regulated, and tied to t he plates.

This brings us to the operating characteristics of the 807's, triode
connected. Power triodes are voltage amplification devices. They try
to amplify voltage. With an output resistive load, this presents no
problem to the load line.

By contract, power pentodes or beam power
tubes try to present a constant current to an output load. With a
resistive load, this also presents no problem to the load line.

The problem is that loudspeakers (the intended load of the output
transformer) are not a resistive load at most of the used frequencies
of a loudspeaker. When a loudspeaker is attached to an output
transformer instead of a resistive load, the load line of the output
tubes goes crazy, whether the tubes are triode or pentode connected.
Neither triode or pentode mode operate well with loudspeakers. This is
why all performance tests are carried out with resistive loads.

So what?

Triodes seem to work well into reactive loads...



Keroes and Hafler invented the tapped screen mode of operation of
output tubes. By connecting the output tube screens to a tap at an
appropriate winding location, the output tubes put out constant power
into a load, rather than either constant voltage or constant current.

So what?

The result of UL is that Ro of the output stage about equals RL.
With triode RL is a lot more than Ra, and with pentode
Ra is a lot more than RL.
UL was used to maintain pentode power but allow
enough plate signal fed back into the tube to linearize the
current flow.
Its because to get a wide plate voltage swing from a triode,
you have to operate class AB2, which was a pita and cost another 6SN7
used as a CF driver.



Distributed inductances and capacitances in the speaker circuit cause
the varying impedance of a loudspeaker over the used range (see:
Acoustical Engineering--Harry Olson, Chief Engineer, Audio, RCA. Harry
also taught acoustics at Columbia University when his book was
written.) Olson's book is the bible of the audio industry to this
day).

Where is Kroomel when you need him?

As is easily shown, inductances and capacitances are reactive in
nature. They generate what is known as reactive power. You cannot
hear reactive power. What you hear with reactive power is phase shift,
which in stereo blurs the stereo effect.

The reactive nature of speakers cause currents to flow through the amp
which are not producing audio power, but which rob the
amp of its maximum current ability.
But despite all you say about a W triode amp, used at normal levels into
almost any speaker, they manage to sound and measure well.

This BS about reactives being so evil is just a myth.
They are an inevitable part of converting electric power into sound.
If you have a speaker with an ill concieved Xover network with a
series resonance at say 500 Hz, and a 2 ohm impedance, then
when a 500Hz note plays loud, the amp can be overloaded, or clip at that =
F,

thus intermodulating all the other musical notes.
But while the music contains no 500Hz, all is well.

Speakers need to be designed well.


By operating in a constant power mode, the output REAL power from a
loudspeaker is more constant. The frequency response is more linear.
It is obvious that "ultra-linear" (constant power out) is a better mode
of tube operation than either constant voltage or constant current.
When a passive crossover network is used in conjunction with a
loudspeaker system, the quality of sound degrades more with constant
voltage or current than with constant power ("ultra-linear") mode.

There are too many variables I have not got time to discuss here.
There are no general rules that favour UL exclusively.




There are those presently practicing the art of audio tube design who
do not understand the nature of output tubes or circuits. This results
in a lot of false statements made around this subject.

Perhaps you are a leading light in this trend.....


Given
everything else held constant, no triode or pentode tube operation
equals "ultra linear", (constant power) operation. The physics is
against it.

But the power in music is ever changing, and the RL ever changing.....
Power levels vary regardless of triode, UL, ot pentode/beam.




Some Observations

As mentioned elsewhere, Williamson was a tube engineer who worked for a
tube company. Williamson never made the amplifier bearing his name. If
he had, he would have made some modifications. Using chokes for
instance, and then negating their advantages by using a capacitance
input was rather silly. Williamson used a 5U4 power rectifier, which
was a directly heated cathode rectifier tube. This meant that the
amplifier tubes saw B+ before they were warmed up; bad for cathodes and
capacitors. The B+ (without load) was higher by far than normal
operation. Williamson's capacitors weren't rated for the voltage
surge.

The PV rating of many of the capacitors I have seen in W and other
tube amps and countless radios is quite adequate.
What killed caps in PS was a saturated output tube, and the ripple current
took out the cap.





Others subsequently used a mechanical switch to keep the high voltage
from the amplifier until the tubes warmed up. A far better circuit
uses a 5V4, an indirectly heated cathode rectifier, which does not draw
current before the rest of the circuit is ready as it takes it cathode
time to warm up too.

In a recent upgrade of a Quad II with 5AR4, the B+ soars up to about 440V
before being pulled
down by the output tubes turning on.
In fact what you say isn't quite the total picture; indirectly heated
rectifiers turn on much faster than output tubes.

Many old amps have caps rated for 500V or more, allowing the PS to be
turned on
without the output tubes present.



The Williamson feedback loop did not respond properly with crossover
networks in the output. Then too, the only component in the amplifier
that went to 200,000 Hz. was the output transformer. The Williamson
circuit did not meet Williamson's own criteria for open circuit
bandwidth. (Operation without feedback)

The W with a Partridge was one of the very fastest amps with no global FB,
going
with more bw than many other designs relying on FB to make their bw wide.

In fact, there is simply no need to have open loop bw =3D 200 kHz, since
to have 20 dB of NFB applied at 200 kHz is a real problem
if the load becomes capacitive.

Hence W's addoption of the 470 pF + 4.7 k gain reduction phase tweaking
circuit
applied across the 47k load of V1.

Open loop bandwidth is thus *deliberately* reduced to around 15kHz
and gain is also reduced about 15 dB at 200 kHz, resulting in stability
with pure cap loads between say 0.05uF and 0.47 uF.

The sound does not suffer from such measures.



In 1947, speakers were mainly high efficiency types. This meant that
the bass resonance was high by modern standards, and the high
efficiency created a more ragged audio response curve. The electrical
impedance curve was more ragged also. However, a ten-watt amplifier
drove the speakers to acceptable levels. In today's world, ten watts
doesn't make it, as the speaker systems are no longer high efficiency.

The efficiency of any speaker varies with F, but all are designed to
make a given constant SPL at all F required providing the voltage of the
signal is held constant for all F.
Current is allowed to vary between lots, and hardly any, and whatever F.



Williamson used a two chassis system for his amplifier, believing that
magnetic coupling between transformers caused hum. Poor power supply
filtering caused Williamson's hum problems. No one produces two
chassis audio amplifiers today. There is no purpose.

I produce dual chassis amps.

There is a purpose.

And hum problems are rarely from B+ filtering.
The W amp has adequate B+ filtering.

Hum can occur from a variety of mistakes in building any amp,
not just a W.




It is curious that Williamson did not have his company design a good
triode equivalent to his triode connected tubes (KT66 or 807--U.S.)
The 2A3 was a better triode than triode connected KT66's. The Brook
amplifier that used 2A3's was a better amplifier than the Williamson.

So what of the 300B?

You don't even mention them.

But beam tetrodes and pentodes were here to stay.

They are hear to stay.

I recently tried a pair of KT90 in a Quad II amp.
They worked fine, giving about 20% more maximum power.

And in a Williamson, KT90 also do very well.


There are those who will be talked into building this "antique". I
would suggest that if they build one, put it on the shelf and just look
at it. Williamsons and buggy whips don't have a use in the 21st.
Century. Also, as far as is known, no one is making the Partridge. In
today's world, it would cost too much for 10 watts out.

The OPTs I make go from 2Hz to 200khz when i want them to.
Its easy to make something equal or better than Partridge.

The 300 watt OPTs I made went 270kHz.

Partridge wasn't the only maker who knew how to get wide bw.

The only reason why 99.99999999999999999999999999% of ppl
didn't buy Partridge is that the costs of raising a family did not permit
Partridge luxury. Hi-fi was seen as pretentious BS activity
for the layabouts with too much time and money.

These do littles soldered their amps together, and they still do,
why its better than doing a whole range of other silly things.

But a well done Williamson isn't such a bad amp.

The OPT doesn't need to be quite as W specified.

A bigger core, and less turns per volt dramatically improve the outcome.

The first large amp I made was a W with a quad of EL34 in triode.
It was terrific, 30 good watts, and later I went for UL
when i got better at OPT winding.
Nobody could tell me there was any sound change between UL
and triode.

Williamson had a lot of very bright ideas, many of which were ignored
by the makers of so many compromised amps after 1950.

Generally in the US if you homebrewed audio you were either retired or
young and technically curious-same as ham radio. But whereas hams were
generally cheap, audio guys went first rate. Dynaco and other cheap
kits were the beginning of the end for scratchbuilders.