Post by Chris HornbeckOn Mon, 28 Feb 2005 11:28:06 +1100, Patrick Turner
Post by Patrick TurnerThe time constants are large in the couplings.
One could perhaps use 0.1 uF, and still get LF stability,
depending on the OPT inductance.
In the particular case of low frequency stability and
overload recovery, I tend to fall into the DTN Williamson
camp and favor (what's nowdays called) a single dominant
pole compensation. It's not necessarily optimum for
extended response, but it is optimum for overload
recovery, and, as the old saying goes "If it don't sound
good loud, it don't sound good.".
Williamson's amp had more RC couplings than the Mullard 520 et all.
Therefore W needed at least 100H in the OPT even at low levels
to get a low enough pole between the load and Ra of the triodes in
parallel.
Therefore he use 4,400 P turns on a core of 32 mm tongue x 44 m stack.
Bean counters universally hated Williamson because he said whatcha have
to do, and nobody wanted to do it.
GE knew that nobody took W seriously, so they even provided notes in the
back of a book in 1957 called '17 amplifier circuits from 5 watts to 1,100
watts',
so that ppl would be able to use OPTs with only 20H of minimal P
inductance,
and still get good bass, **and stability** with a lot of NFB used.
Many Williamson amps oscillate badly since the OPT ppl have tried to use
isn't
anywhere near what W said it had to be.
Post by Chris HornbeckPost by Patrick TurnerThese networks have to be worked out for optimal
critical damping and will vary with the OPT chosen for the job.
One of the issues when DIYing an audio amp is that
any given schematic will not give the same stablity margins
with different OPTs since the shunt C, leakage inductance and primary
inductance all may be different.
And in the case of high frequency stability, neither of us
lean toward the op-amp model of a single dominant pole. A
good, finished design that includes loop feedback *will
neccesarily* require some cut-and-try both in the forward
and the feedback paths. As your circuit elegantly shows.
Its impossible to define the nature of the OPT shunt C and LL until
the OPT is in the amp.
The aim of a good amp dsign with tubes is to have the open loop
BW without the zobel compo networks of phase tweakers
as the same as closed loop BW with all the tweakers in place.
In my 300 watt amps I got 300 kHz of OPT BW due to
the 5P x 6S interleaving, and the fact that since RL is only 1k a-a,
and the core so large, only 1,100 primary turns are needed
for good bass, and the Lp and LL is mainly determined by the Np squared,
so LL is below 1 mH.
But I cannot allow the amp to have a final BW with NFB of 300 kHz.
That would only work if the load was an R.
I like to get 50 kHz of BW even with 2 uF and no other load,
although one of course cannot get full mid frequency voltage
of around 45 vrms across a 2 uF cap at 50 kHz.
But at 4.5 volts of output, its easy.
In fact at low levels any value of cap between 0.1 and 10 uF shouldn't
cause
more than 3 dB of peaking in the sine wave response anywhere.
If the peaking is above 20 khz, well and good, as it should bw with up to
5 uF, but with
10 uF the peak is below 20 khz.
The tweaking of the open loop phase and gain reduction above 20 kHz is a
must
even if the open loop BW is 300 kHz.
That's because there is still leakage L and some HF poles from the miller
effect,
and so a wide band amp with OPT will simply oscillate
at a higher F than one with more LL and lower miller poles.
The trick is to use a **minimum** of phase tweaking to control both the HF
gain and
HF phase shifts caused by miller, so that a BW of 65 kHz can still be
obtained
with a pure R load, yet no pure C load will cause more than 1.5 dB of
peaking in the
sine wave response at any F below 20 khz.
Usually ESL have some R in series with their dominant C component/s
and some R in parallel with all that, and then the amp
set up the way I do it will have a flat response into any real world load,
and act better than many SS amps I have tested.
The compo C across the global FB resistor can also be minimised
by placing a small L, maybe only a couple of uH
in the speaker return wire in the amp and taking the
low value R used in the cathode of V1 to a point where
negative current FB is generated as F rises above 20 khz.
The current in the L tweaks the phase 90 degrees, and so "fits"
the delays in the rest of the amp, and the CFB allows the Ro of the amp
to become higher above 20 khz where low Ro is a pest.
Yesterday I added such an L into a leak TL12, a truly HORRIBLY
unstable amp compared to mine, and was still able to get 50kHz of BW
into 8 ohms, and LF stability without a load and HF stability with any
sort
of pure capacitance, and a far better ability to drive a speaker such as a
Quad ESL
which is like 15 ohms strapped with 2uF and 1.6 ohms in series.
I was able to not have to use any phase tweaker zobel at V1 to
control the miller C between the EF36 and following 6SN7 LTP.
The generation of some CFB at HF is actually more effective
with low value reactive loads at HF than trying to rely
solely on caps across the global voltage NFB R.
But zobel across the output, usually 10 ohms plus say
0.22 uF are a necessity with amps with poor stability margins or no margin
at all.
Post by Chris HornbeckNote 1: If conventional solid-state amplifiers were paid the same
level of attention to these same issues, they'd be happier
campers.
Many SS amps have passive zobels with LR and CR after their outputs
which really is back at the emitters of the output stage,
also where the NFB is sent back to the diff pair.
The average SS amp couldn't have the 60 dB of global NFB used
if the L component of the LR zobel was included,
especially without the R strapped across the L.
The tube amp has the L in there where we like it or not,
and it is the leakage L.
if the impedance ration of the opt is say 6k to 8 ohms,
or 750:1, as it is in an old Leak set to 4 ohms,
and the LL is 50 mH like it is with a Leak at the primary, then
we have a massive 66 uH in series with the output.
and in a tube amp its impossible to have *effective* global FB
unless we take it from the output.
Since the LL is so high, only a limited amount of NFB can be used.
But if LL was say 4 uH then the pole caused by LL and the load is
at a far higher F, and much more FB can be used.
4 to 10 uH is commonly used at the output of SS amps
with an R of between 4 and 8 ohms, and then there is a following
series RC and perhaps a series RC from the output to 0V.
The first RC acts to have the output emitters see an R load at above say
100 khz, and thus the output device gain is limited and stability assured.
the rest of the L and C and R hanging off the SS outputs make sure
C loads of any value never cause the amp to actually see the low impedance
of a C
type of load.
Square wave tests with SS amps can show serious over shoots
with cap or RC loads, and most of this over shoot is merely a
function of the passive filter components of the "outpboard" network.
A CRO on the output emitters usually show a good square wave
The square wave at the output of the diffamp, ie, the early error signal
should show a smooth curved line of the horizontals and thus show
that the amp is ever so easily compensating for the droop in the open loop
response regardless of load or oscillations in the load current.
Good tube amps should show minimal oscillations in their
error signals with a square wave.
Post by Chris HornbeckNote 2: Maybe a good topic for a coming-soon-to-a-newsgroup-
near-you thread might be "Optimizing the open loop and closed
loop performance of fed-back amplifiers", concentrating on
measurement strategies.
The theme has been done to death.
A search of rat archives under my name and 'NFB' and 'critical damping'
should give you at least a book or two to read.
Post by Chris HornbeckNote 3: The mid-1990's craze for homebrew SET's, and, importantly,
zero loop feedback, may be a reflection of how *wrong* this issue
is typically treated among us builders.
Not really.
There **is NFB in triodes**.
There are at least 2 ways to get to an amp with
low thd of less than 1% at any full power, and with Ro = less than 1 ohm.
One is to start with a Horror like a Leak amp, and place the
original 26 dB of NFb around the circuit and that'll give
Ro less than 1 ohm and thd a lot less than 1 %.
If we take the FB away, and set the Leak amp up for 9 ohms, we
will find that Ro = 12 ohms, very high for a triode amp but it is due to
the effective
winding resistance and transformed plate resistance being equal to
the sum of 2.7 and 9.3 ohms respectively.
the "9 ohm" load match gives the PP KT66
a load of 2.9 k RLa-a plus the 850 ohms of winding R a-a.
Without NFB, the Leak would be an attrocious performer.
BUT, if we then said OK, we will use a quad of
KT66 in triode, and then use a low winding loss transformer
that has 5% total losses, so that Rw at the output was only 0.4 ohms,
and then we will have an impedance ratio of 2,666:1,
then the Ra-a of 1.6k of the 4 x KT66 will ap[pear at the output as 0.6
ohms.
The total Ro will be 1.0 ohms.
The load seen by the 4 tubes with 8 ohms connected will be 2,666 x 8 =
21,333 ohms
With Ea of 450v, we should get a peak swing down to
say 100v, so Va-a = 500 vrms so Po = 11.7 watts,
and it will definitely be all class A, and remain class A even if
4 ohms is connected, and po will increase lots.
The thd of KT66 in triode with such loads is around the 1% at 12 watts,
but then even unloaded, at 500 vrms a-a, the thd won't be much less;
we have reached the residual level of thd at high V swings.
But the thd will decline with Vo, so that at 2 watts, which gives 93 dB
SPL
into todays insensitive speakers of about 90 dB/W/M,
the thd should be arond 0.4 %, and mainly only 3H.
Tubes such as 300B should do a little better.
So we have out 12 watts at 1% and 1 ohm, and the onlt NFB we have allowed
is the
NFB put there by the God of Triodes when he said "Let There Be Triodes!",
in about 1903.
The drive amp driving the output stage has to be designed carefully
but getting 70 vrms grid to grid at 0.1% thd without NFB is easy.
With similar logic, 4 parallel KT66 in SE will produce
low Ro, but have maybe 3% thd at 12 watts, but its mainly 2H
instead of 3H, which will be nearly absent, since the 3H appears
in output class A stages because of the imbalances of the difference
between Gm
with PP class A tubes turning on or off.
Few devices have a very pure symmetry of turn off/turn on Gm with respect
to output voltages.
The well set up triode amp with no loop FB can sound utterly delightful,
and it is hard to build a bad sounding one in either PP or SE
providing the basic rules are followed, thd below 1% at
normal levels used, Ro less than 1 ohm, and enough damn power
to give yourself a ceiling.
A quad of KT88 or KT90 in triode are rather better than KT66, 807, 6L6 or
EL34 imho.
Patrick Turner.