[Patrick Turner]

On Jun 26, 2:25*am, John Byrns wrote:
In article ,
*Patrick Turner wrote:





On Jun 24, 3:34*pm, John Byrns wrote:

I want to see a schematic with all test results before I make up my
mind on Alex's FB "trick." It could be a clever trick, or a swindle.

However the network introduces both a zero and pole into the response, with
the
zero at a higher frequency than the pole. *Remember this network is just
another
tool in your toolbox; it is not a cure all and requires some sophistication
in
its application. *Now the one thing I know about stabilizing the low
frequency
response of a feedback system is that it is all about correctly placing the
poles, and zeros if there are any.

In any amp where there are say 2 CR coupled stages and a final stage
with LR then you have a recipe for LF instability and a poor margin of
stability at LF.

OK, I have worked through some of the math and understand more fully what is
going on with Alex's feedback network. *As I said the network introduces both a
zero and a pole in the loop gain.

Ignoring the added pole for a moment, the zero can be placed so that it exactly
cancels the effect of one of the three poles in the amplifier you describe
above, with 2 CR coupled stages and a final stage with LR, effectively reducing
the number of low frequency poles by 1, making the LF stability problem easier
to deal with.

I wish I could be so sure. Just because you have in theory "cancelled
out" one devil it doesn't mean you've done much, especially
considering that the Laex FB path renders the FB amp to be a cathode
follower at DC, when you have no ability at all for the V0 to follow
and input at DC. The circuit needs to be built and tested. But notice
how silent and bone lazy everyone is about the issue. They just don't
want to get out of the their arm chairs. How can I have respect for
laziness?

*The zero effectively cancels both the phase shift and amplitude
roll off caused by the pole that is being canceled. *Unfortunately it is
impossible, at least so far as I know, to build a network with an isolated zero
such as I have described, so an actual network, such as Alex's, must include a
pole at a lower frequency. *Hopefully this new pole won't cause us too much
trouble if we place it at a very low frequency where the loop gain has already
fallen well below 1.0 as a result of the other two remaining poles that weren't
canceled.

Now the obvious question is, why bother with this extra complexity when we could
simply directly move one of the 3 poles to a very low frequency, as would
probably be part of the normal pole staggering process anyway? *I will leave
that for others to comment on as I have not personally mucked about in my
workshop with amplifiers that have 3 LF poles. *I suspect that one reason may
have to do with LF overload when using a Bean Counter approved OPT.

Its not always convenient to move poles down, and often it merely
moves the oscillaton F lower. It can be exasperating to find that even
if you increase coupling caps from say 0.22uF to 2.2uF, the trace on
the CRO still slowly rises and falls because Fo has just gone lower.

But you don't need a bean counter designed OPT to give a pole that is
at an F too high. You could have a well designed OPT for an SE amp
which will is designed to saturate at 20Hz at full PO and yet it
oscillates at LF. This is because the permeability of the core has
been much reduced by the air gap from the maximum one might find in a
PP OPT with fully interleaved laminations. So I have found my shelving
network especially effective in SE amps. The Williamson PP amp
required Lp = 100H minimum with triode Ra-a of 3k2 plus RLa-a = 10k in
parallel, ie, the ratio of RAA to Lp = 2k4:100H, giving a pole at
3.8Hz. With say 4 x EL34 in parallel SET, Ra = 310 ohms, RL might be
1k0, so RA = 236 ohms. Lp might be 8H to give XLp = 1k0 at 20Hz, so
the -3dB response pole is at 236 / ( 6.28 x 8 ) = 4.7Hz. The SE amp
is then in theory slightly worse off with regard to phase shift at
LF.

My shelving network is especially effective where the OPT has far less
inductance than it should and usually the Fsat is also way too high,
eg, in most old radios the OPTs are allowed to saturate at 70Hz. This
is why they are so small in size; Afe is 1/4 of the size needed for hi-
fi, for the number of turns used.

In the case of a pentode or beam tetrode SE OP tube, the open loop LF
pole is determined by the RLa and the Lp, and Ra has little effect
because Ra is such a high R.
So say you have 4 x EL34 in parallel pentode then RLa = 1k2, and Lp
needs to be 9.6H and the response is down 3dB at 20Hz. If the speaker
load is high, or there is no speaker connected, then the LF open loop
pole is between Ra and Lp, and if Ra for the 4 x EL34 = 4k0, then pole
rises to 66Hz. One cannot easily move the pole.

So, just how Alexe's brainchild works in all conditions remains to be
established. Unconditional stability is mandatory in all my amp
designs as it is in all other reputable brand names.


I can see how Alex's network has the potential to resolve a problem I have
encountered when mucking about with simpler amplifiers having only 2 poles. *
When using OPTs designed by Bean Counters, especially SE OPTs, there is a
tendency towards LF overload in the OPT and final tube(s). *

Nearly all the SE amps from brand names have Fsat too high and Lp too
low. And so do many PP amps. amps with only 2 stages like Quad-II have
better LF stability with shelving networks. They are inherently stable
because of their single CR coupling and OPT LR but the trace wobbles
after going from high level to zero level. It remains to be seen if
they'd enjoy a cap in the FB network.


I have attempted to
mitigate this problem by choosing a relatively high pole frequency for the
interstage coupling network to keep LF signals out of the OPT and final tube(s). *
This puts the interstage pole too close to the pole caused by the OPT which then
causes a bump in the ³CLG² low frequency response, plus of course it isn't
really a very good solution to the LF overload problem. *It occurs to me that
Alex's feedback network might also offer a solution to the OPT saturation
problem in Bean Counter designed OPTs, just as it offers a solution to the input
stage problem.

Mere postulations John. The only real truth is known when theory is
applied.

Patrick Turner.