Marcin Slawicz wrote:
Uzytkownik "Patrick Turner" napisal w wiadomosci
...
*** Snip ***
Williamson had a lot of very bright ideas, many of which were ignored
by the makers of so many compromised amps after 1950.
Patrick Turner.
Thanks Patrick for your great explanation. I was familiar with White's
article and really resisted some statements I found there. There is quite a
number of misguiding articles regarding the Williamson amplifier on the web,
and because of my nearly no-experience in tube electronics, I am not always
sure what true and what false is.
At all times in the past, nearly all men with a slightly different way of
building an amp
would routinely say how bad the other guy's amp was.
The same BS goes on today.
I have built just about all types of amplifier, and all are derived from what my
father's
generation thought up.
For example I like the Quad II idea of CFB from the OPT, but I think the way
Quad
implement their idea is very lack lustre, and its a chic amp which has been
dumbed down by accountants.
It still sounds ok though, at a few watts.
I criticise many of the sacred cow brands such as Leak, Quad, Dynaco,
and I don't care whose ego I bruise; if they can't see that these old bits of
junk
**could have** been a lot better pieces of engineering had the makers used a
little more labour
and material instead of buying that new Mercedes for the boss, then they'll
never see anything.
Anyway, I use the Quad II idea in preference to the normal screen taps of UL.
But I use much more CFB, since the amount Quad use doesn't do enough to reduce
Ra-a and thd. Same goes for UL; one still has an output stage with Ro = approx
RL,
but its better than nothing.
If you don't mind, please take a look at my first tube project I am
assembling these days
(http://www.echostar.pl/~slawicz/conc...oncertino8.htm). The site is in
polish and it's not completed yet, but you'll find the schematic there.
The biggest difference between Williamson's amplifier and my one are
(despite the UL output stage) the constant current draw first two stages in
Aikido style described many times by John Broskie in TubeCad Journal
(http://www.tubecad.com/april99/page6.html). I hope it will work properly.
The simulation shows about 14 dB better PS noise and ripple rejection
comparing to typical Williamson front end.
Best regards,
Marcin
There is always a constant current drawn from the PS to any kind of SET
V1 and CPI V2 stages, simply because class A stages do not have a changing
supply power, other than that due to the 2H produced in signal, and in your case
that
will be below 0.1% evan at 2vrms from V1, so the DC voltages at
all electodes won't change much.
The use of the R19 bypassed with C5 and R24 is a step in the right direction
because it makes the set up conditions of
V1 and and V2 easier to arrange so that the anode voltage at V1 is best for
linearity,
and the grid voltage at V2 is best so Ek isn't too high above the heater supply.
Also at very LF, where the williamson margin of stability becomes
much less than at 1 kHz, the R19 and R24 become a voltage divider reducing open
loop gain
by 6 dB below 20 Hz, because C5 graduallly becomes a large and open impedance at
LF.
You could further stabilise the amp by using more LF attenuation in the open
loop response than you have.
Go to
http://www.turneraudio.com.au/htmlwe...0ulabinteg.htm
There you will see a schematic for 50 watt class AB1 UL channel using KT88/6550
and including an integrated input preamp stage.
It isn't a williamson, but is a derived topology from what Mullard used
back in the 50's, and which has been addopted by 1,001 makers since then.
However, I have a few secret things in there that would NEVER have been done in
1955, such as the gain&phase shift network between the first tube in the power
amp, and the
second, which is 1/2 of the longtail pair.
This network should be of great interest to anyone building their
PP amp especially if the OPT isn't up to the **best** of the Partridge models.
About 99.99% of OPTs sold on this planet are a lot worse than
Partridge OPTs.
There is a network containing 0.47 uF, 0.047 uF, 1M, 220k
which acts to reduce LF gain a maximum of 15 dB, starting at 15 Hz.
By 1.5 Hz, at which many amps would oscillate badly without this network,
the LF gain has become about 12 dB less, and as stability is affected by the
amount
of open loop gain and the amount of FB applied, the amp will have far less NFB
applied
at 1.5 Hz than at say 100Hz, so the amp will be rock stable at *all* LF.
The amount of FB that *can* be applied safely without oscillations
occuring is also dependant on the phase shift of an amp, and stablity
is threatened as phase shift approaches 180 degrees betaween input and output.
At LF, the CR couplings between stages and the primary OPT
inductance will produce a max of 90 degrees each.
So its not uncommon for a tube amp to have maybe 180 degrees of phase shift
at 5 Hz, depending on the OPT and CR couplings, because the phase shift
of each reactively affected stage is acumulative.
So 3 lots of 60 degrees adds to 180 degrees.
FB applied, in dB = 20 x log of A / ( 1 + [ A x ß ] ), where A = open loop
voltage gain, ß = fraction of the output voltage fed back to the input,
and 1 is a constant required for all equations to work.
If mid F gain, or 1 kHz open gain A = 50, and ß = 0.1, then FB applied at 1 kHz
=
20 x log of 50 / ( 1 + [ 50 x 0.1 ] ) = 20 x 1.021 = 18.4 dB.
A is the ouput voltage from where the NFB is taken divided by the input voltage
without any loop FB applied.
You need to be totally and utterly familiar with this formula
and how to derive ß or else you just won't know what you are doing
when you are building an amp, like most ppl when they start.
The smoky and lousy sounding amps are a testament to their ignorance.
You can also measure and calculate the amount of FB you have applied.
Applied FB in Db = 20 log ( output voltage without FB / output voltage with FB
).
A typical Williamson amp may need 0.2vrms input for 10vrms output without
loop FB.
Hence open loop gain = 10 / 0.2 = 50.
Now with this LF gain reducing network, or step network, or shelving network
which some folks call it, open gain at LF is reduced about say 4 times at 5Hz,
and then the OPT and other CR couplings will have reduced the gain in the output
stage about 6 dB,
so total open loop gain at 5 Hz might be 50 / 4 x 2 = 6.25.
So the amount of FB in Db = 20 log 6.25 / ( 1 + [ 6.25 x 0.1] ) = 11.7 dB.
This is a considerable reduction in the amount of applied NFB.
The open loop phase shift at below 5 Hz won't be any worse than you'd have with
all CR couplings, so stability is far better with such a network in place than
without,
and I use such a network on all my amps.
Many amps with FB will oscillate at say 1.5Hz, but not with this network
when it is fitted, because open loop gain continues to fall.
The values can be fiddled with make it more effective with crummy low inductance
OPTs,
so that 0.022 uF or even 0.01 uF will begin to reduce the open loop gain
at a higher F.
Such networks do not substantially reduce the bass response at 20 Hz or increase
the
bass F output impedance if they are simply designed to do their dirty work below
20 Hz.
Who cares if the Ro of an amp becomes higher at F outside of the audio band?
The same idea of gain stepping applies to the HF response.
In my schematic there is a network of 3.3k and 330pF which applies itself
effectively
too the anode of the input triode.
The input triode in my case has a bypassed Rk, so its Ra is a low 10k
which is in parallel with the 75k DC RL + the bias R of 220k so the output
impedance of the set input tube is 10k //75k//220k = effectively 8.4k.
Gain for all tubes = µ x RL / ( Ra + RL ).
Now as F rises, the RL seen by V1 anode falls because the 330pF
begins to have a low impedance, and by 500 kHz, Z330pF = 1k.
Thus the load the V1 anode becomes approximately 4k,
and thus gain of V1 is halved at such a HF.
The phase shift effect of any stray C or Miller C at such F is reduced
because the R component of what is a low pass filter circuit has become lower.
Therefore the amp is more stable because at HF the phase shift and open gain has
been slightly reduced at an F where
the amp is otherwise likely to oscillate, especially if a capacitance load is
ever used with the amp.
meanwhile, anything going on at up to 20 khz is totally unaffected by the
gain/phase tweak HF network.
To get the best value for the CR HF step networks, don't just calculate.
You won't succeed. There are far too many unknown C quantities, not to
mention the leakage inductance of the OPT which all affect the signal through
the amp.
I set up the amp to be stable with a 0.22 uF across the output without a load R.
Many amps oscillate violently in such a condition.
But first you have to apply just enough C across the global feedback resistor
to **advance the phase at HF** of the HF signals being fed back, to compensate
for the phase lag the HF signals suffer as they pass by all the R-C and R-L
interfaces.
If the OPT is just good enough, without too much leakage L then you should be
able to just stabilise
the amp with the C across the FBR.
Once the oscillations have stopped with the 0.22 uF,
you then hook up a variable radio tuning cap of 20pF - 365 pF,
and a 25k pot in series with a couple of jumper leads,
and adjust the pot and cap for the least ringing on a 5 kHz square wave with the
0.22 uF cap in place as the sole load on the amp.
If the ringing is only a couple of cycles, and the overshoot is only 3 dB more
than
the peak rise voltage, you have done well, but the sine wave response at full
power
into a resistance load should still be 65 kHz, - 3dB.
Its easier said than done, because you have to consider about 10 interactive
things at once,
and the God Of Triodes arranged the laws of electronics to be dammned difficult,
and bloody frustrating, to make sure only those with a modicum of intelligence
could
work out what exactly has to be done with the bits and pieces in an amp to make
sure they will
always be stable, since idiots who will never understand or have a natural feel
for
these things give up and go away, and become politicians or parking inspectors.
It is vitally important to consider all amplifiers as bandpass filters,
with poles affecting their bandwidth, and in the case of a tube amp.
there are quite a few poles because of all the Miller C in different stages
and the effects of primary and leakage inductances.
Solid state amps also have to be considered similarly,
and in their case the amount of NFB applied is usually
vastly more than is applied in a tube amp because the
poles of the roll offs are further away from the boundaries of the 20Hz to 20
kHz audio band.
But they too will oscillate badly at HF especially without correct HF
compensation
networks placed to control HF open loop gain.
They also have a zobel L&R and C&R network to make sure there is a load on the
amp
at HF, and that the output stage is never directly connected to an extremely low
load value
which would cause the devices to fail easily ( which they do anyway, judging
from the
2 or 3 amps I get to fix each week with fused bjts ).
Such zobel networks can also be applied to tube amp output stages, but
experience and practice
is needed to get the values right to make sure they don't make the amp less
stable,
and don't help overload any stage of the amp by causing too large an error
signal at HF.
With some intelligent use of R&C parts, your Williamson will sing.
But every OPT has a different amount of leakage and stray C so there is never
going to be an easy way to make your amp unconditionally stable,
ie, any load whatever can be connected including any L, or C, or no load at all,
and oscillations will not occur, ever.
When all the conditions above have been addressed properly,
usually tube amps sound just fabulous.
Patrick Turner.