Fighting capacitances lurking with malicious intent in your amp:Slew rate current, Miller, stray dogs and bandwidt by Andre Jute

[Andre Jute[_2_]]

From JUTE ON AMPS http://www.audio-talk.co.uk/fiultra/JUTE%20ON%20AMPS.htm

KISS 125
Fighting capacitances lurking with malicious intent in your amp:
Slew rate current, Miller, stray dogs and bandwidt
by Andre Jute
Just after WW II the PR man at Mahatma Gandhi's ashram was giving a
group of American and British journalists the short tour. The
Americans were interested in the two nubile virgins the loincloth-clad
old sage slept with every evening to keep him warm and to test his
willpower. The British, still hungry after war shortages, wanted to
know what Gandhi ate.

'Only the simplest food,' the PR man said. 'A little lemon and honey
with soda water.'

'Wow,' said the Brits and moved on. (This was before their tabloids
became the scummiest in the world.)

Afterwards the Time-Life reporter hung back. 'Aren't lemon and honey
and soda water very great luxuries in India?' he asked cynically. 'How
can you be so hypocritical as to describe that as simplicity?'

The PR man, who had been an Inner Temple barrister with Gandhi in
London a few decades before, knew when he was caught out. 'Quite,' he
said smoothly, 'but I never said it didn't cost a fortune to keep
Ghandiji in simplicity.'

***

Ultrafi tube amps are like that. The simpler they appear, the greater
the mental effort required to win the benefits of that simplicity.
There hangs about the apparent simplicity of The KISS Amp 300B a whole
miasma of invisible capacitances which can test one's willpower far
more than two naked nubile virgins.

Although we commonly say that 'a Class A power tube draws no current
on its grid', a power tube requires current from the driver to
overcome various capacitance which loiter with malicious intent in
your amp. A useful shortcut to half a good answer is slew rate current
(which tells us how fast a capacitance is charged up and discharged)
and as usual experience provides the other half of the answer. The
first half of the answer can be calculated with slew rate formulae and
one empirical assumption generously suggested to me by Gordon Rankin
(who isn't responsible for what I do with the information) on 15
October 1996 while I was working on an 845 amp:

***

The Slew Rate
SR = 2*Pi*Bandwith*Vmax
Where
Bandwidth = 20kHz or whatever higher frequency your design will pass
Vmax is the maximum voltage the stage will deliver

The input capacitance
Cin = (A + 1) Cgp + Cgc
Where
A = the gain of stage for which the input capacitance is being
determined
Cgp is the Capacitance of the Grid to Plate
Cgc is the Capacitance of the Grid to Cathode

The Slew Rate Current
SRC = Cin*SR

Erno Borbely, Jung and Ron Gunzler (according to Gordon Rankin)
suggest that the stage current should be 5 times the slew rate current
to overcome the input capacitance of the next stage.

The Stage Current
Scur = SRC * K
where
K = 5

***

Bear with me while I put in some numbers:

Desired Stage Current is therefore 5*Cin*SR.

For a 300B,
Cin =9 + (1 + 3.85)*15 = 81.85pf

For a fullrange amp with 80V signal voltage,
Slew Rate = 2*3.14*20000*80 = 10,048,000
(note that to arrive at an irreducible minimum we enter only the audio
range)

therefore

Desired Stage Current = 5*81.85*10,048,000

and after moving the decimal we get

4.1mA

***

Mmm. That is of course an absolute minimum requirement. Notice
something above? No one has yet counted stray capacitances. In a
simple 300B amp strays usually amount to between 15 and 25pf. They too
have to dealt with, which is why the constant is a multiple.

We know from experience that a 300B likes more current on the plate of
the driver than 4mA. It was the excellent Steve Bench who first
suggested to me that 8mA is the right minimum number for a driver for
300B. In fact, this is less of discrepancy between theory and practice
than at first appears. We would normally design the amp out to at
least 40kHz, not just 20kHz, so the calculated current requirement
would match the 8mA from experience without altering the famous
constant of 5.

Do the calculations for an 845, which has a difficult signal
requirement of around 150V in the more common designs, and you will
discover that 20mA is just about a minimum on the driver, which
accounts for why some of us laugh when we see designs for 12AX7
driving 845. It also accounts for why I like to drive kilovolt
transmitting tubes with a 300B booster amp, or at least a power tube
for a driver. The KISS Amp 300B is just such a booster design,
requiring only a switch to turn the primary of the output transformer
into a choke load on the plate of the 300B and a polyprop cap in
series on the output to the grid of the main kilovolt amp. In other
words, the 300B is used as a preamp (control amp) tube and as a weapon
to absolutely murder Miller.

***

Now we come to the Miller Effect, because we must. Even the exemplary
RDH treats this distasteful subject, somewhat akin to removing the
nightsoil, less thoroughly than I do here, and believe me, I'm doing
the minimum I can get away with and still claim a modest electronic
respectability.

Miller is a multiplicatory part of one of three shunt (read "unwanted
but unavoidable") capacitances in an amplifier:
Cpc, the output capacitance of the first stage (from plate to cathode)

Cin, the input capacitance of the second stage, which is Cgc (from
grid to cathode) augmented by the Miller Effect

Cstray, stray wiring capacitance
These shunt capacitances add up, with the numerals indicating the
first and second tubes:
Cs = Cpc1 + Cin2 + Cstray
where
Cin = Ggc2 + Cgp2(A2 +1)
A2 is the amplification factor of the second tube
Cgp2 is the grid to plate capacitance of the second tube.

The Miller Effect is the Cgp2(A2 +1) part of Cin. It is clearly
important because the grid to plate capacitance is multiplied by the
amplification factor of the tube.

The high frequency rolloff of a stage due to shunt capacitance is
f =1/(2*pi*R’eq*Cs)
where the effective load on the plate is
R’eq = Rg+((Rp*RL)/(Rp+RL))
You can see the important part played by the plate resistance and it
isn't rocket science to see that the you should wire tidily and keep
leads short to avoid stray capacitances.

***

You can use the capacitances loitering in your computer constructively
if you think laterally. For instance, the bandwidth of an amp should
be balanced. Any extension below a rather high level, say 50Hz, should
be matched by an extension at the top end. Conversely, if the bottom
end is being deliberately sloped off early to protect horn drivers
(which rapidly become unloaded below Fs), the HF should not be
designed out to infinity, whatever the iron may be capable of; it
should instead be balanced with the LF you are plotting. One of the
tools under your control is the amount of current you put on the
driver, and the slew rate concept is your tool to calculate it. That
ensures that your amp reaches the bandwidth you design to.

Miller gives you the tools to calculate, with data off the spec sheets
for the two tubes and off the schematic of the circuit you have
designed, and perhaps to ensure that the amp will not roll off under
the bandwidth you are trying to arrange by providing slew current. The
calculations are exceedingly tiresome but you can't afford to skip
them. Later we will see how to automate all this with a spreadsheet.

***

The question arises, why do you want an amp that has HF extension so
far beyond your hearing? What is this 'balance' good for? Does
something lie beyond the achievement of 'balance'? The answer is that
there is a subjective effect, which ultrafidelista sometimes refer to
as 'speed' or the 'fast amp' syndrome, and which novices think is only
about undersizing power supply caps. Unfortunately building a 'fast'
amp is far, far more complicated than that and definitely requires
attention at both ends of frequency scale.

The interrelationships of these factors are not well understood but
they are likely to be complicated by the usual difficulties of
psychoacoustics. That is one reason I somewhat dislike the sound bite
of 'a fast amp' and prefer the more sober phrase 'a responsive amp'.

The implication of this caution is that net gossip that "mo' driva
current is betta current" is not necessarily true. There is a correct
bandwidth for every speaker/amp combination, and particular correct
lower and related correct upper frequencies for each application. If
you just throw current at the driver because you have it to spare and
the iron can make the frequency, that can easily be as harmful to the
sound of your finished amp as not enough current.

Don't skimp. Don't overdo it. Calculate thrice, solder once. An
elegant sufficiency will give you the right sound.


THE VOLTAGES IN THIS AMP WILL KILL YOU.
GET EXPERIENCED SUPERVISION IF IT IS YOUR FIRST TUBE AMP

All text and illustration is Copyright © Andre Jute 1996, 2005
and may not be reproduced except in the thread KISS xxx on
rec.audio.tubes


From JUTE ON AMPS http://www.audio-talk.co.uk/fiultra/JUTE%20ON%20AMPS.htm

[Patrick Turner]

On Feb 22, 8:47*am, Andre Jute > wrote:
> From JUTE ON AMPShttp://www.audio-talk.co.uk/fiultra/JUTE%20ON%20AMPS.htm
>
> KISS 125


And then Mr J went on to talk about the expense of keeping Ghandi's
diet suitably simple and nutricious but not much was said about eating
boiled or fried whole grain rice which was probably what G ate when
not eating lemons, honey, and washing it down with soda water.

I sympathize entirely with Mr J's ideas about Miller C and how it does
audio not one bit of good.

I recently had the displeasure to upgrade the circuit of a pair of
Ming Da power amps.
These had V1 as a 1/2 6SN7 a gain triode with the cathode used as the
port for 9dB of GNFB.
V1 current = about 2mA.
V2 was a cathode follower to buffer between V1 and the next stage, V3
and V4 which was another 6SN7 cathode coupled LTP with signal applied
to one side so gain was about 8 and thus Miller C is halved. V2
current = about 2mA.

Next stage V4 and V6 was a pair of 300B of set up as SE gain tubes to
amplifiy each phase of signal about 3.7 times. Each 300B was loaded
with 50k taken to +480Vdc. Ia was about 5mA only.
The output stage V7 and V8 was a fixed bias PP stage with 2 x 845. Ia
was about 80mA with Ea at 1,200Vdc, and they were running damned hot!

The OPT was a crudely designed item designed to saturate at 30Hz+ and
with excessive leakage inductance.

If there was no load connected, just 0.05uF was enough to cause much
HF oscillation. The LF stability was none too hot with wobbly settling
when levels of signals were rapidly changed.

I got rid of the V2 buffer and made V1 a paralled 6SN7 with 7mA idle
current.
a zobel was placed to reduce the effects of HF phase shift Miller C
between V1 and the next V2/V3 LTP with 6SN7. The supply voltage was
arised considerably to increase Ia by 60%.
The 300B stage was converted to LTP operation set up slung between the
bias supply of -200V and its B+ of +470V s that while keeping the 50k
anode loads the Ia in each 300B was 7mA. PP operation of these in LTP
mode reduced distortions.


I would have much liked to add a CT choke with 5k anode load RL so
that there could have been 20mA Ia in each 300B and Ea = +300V, thus
halving THD at all levels up to the maximum of 300Vrms anode to anode
this stage has to make to drive the 845 pair. But there is only so
much the price of the fix can include.

The OPT in many tube amps can be a nightmare where C between P and S
windings shunts the anode to anode signal and does a lot more damage
than the Miller C which happens at slightly higher F.
The leakage inductance also helps phase shift the signal as F rises so
that the LL and Csh get together to form what is effectively a low
pass second order filter with a resonance because the available
loadings are not a loe enough value to critically damp the LC.

So there is a lot more at work to poop on the Motzart other than ther
Miller C.

Of course all these phase lag causing filter effects along the path of
any amp add up so that there often a very rapid rate of phase shift
with signal F increase so that when one tries GNFB the signal fed back
has more than 180 degrees of phase shift where the open loop response
as drooped to 1.0, so the amp oscillates and ppl are led to hate NFB
as a result.

BUT, if you apply a Zobel network to the anode load of V1, and perhaps
to the next gain stage then as F rises the load becomes a lower R
value much less than the load at 1kHz. The GAIN of such stages is
reduced above 20kHz where very little energy is to be amplified, and
so is the phase shift just where oscillation might otherwize take
place. In effect, Miller is defeated, or its effect shoved up a few
octaves where it cannot do any harm because gain is much less than 1
by that F.
Similarly, a Zobel can be placed across the OPT primary OR secondary
to load the stray "parasitic LL and Csh with resistance which then
helps tame the signal overshorts and peaked response above 20kHz.

When all is done well, one ends up with an amp which is happy to be
set at a maximum PO at 1kHz where THD is at say 0.5% and the bandwidth
into a resistance load is at least 20Hz to 65kHz, and without any R
load, no value of pure C load will make the amp oscillate.
And if one examines the workings of all stages, nothing is being
driven into saturation trying to cope with compensating for signal
droop at extremes of F in the open loop response curve.

So there has to be adequate idle current in each stage to cope with a
wanted signal swing into the the existing R loadings, the Miller C and
stray C and the lower loading due to the Zobel network which will
alone be the dominant loading above 30kHz.

Bean counters all around the world have worked so damned hard to
disobey all the golden rules to be followed to make a decent tube amp
because ones one arsole began making crap, they all had to make crap
or else they'd be rooned.

Of course just how you get more Ia to over come Miller etc etc is a
moot point.
But when you have a CCS load or a choke plus R for an SE gain stage
then the Ia can be much greater than were a pure R used to fetch Idc
to the triode so it can work a load plus whatever C is present.

The alternative to CCS and choke loading is to much increase the B+ so
much more Ia can flow while getting Ea high enough for an undistorted
wide swing. This can be difficult and I like the CCS for a V1 stage
and a choke + R for following gain stages. The R + L used as the load
is better than using just a pure L because then you have shunting
effects of the coil's inductance and capacitance and the ultimate 90
degree phase shifts such L and C cause. The series R much reduces the
effects of L and C and the gain does not droop so badly within the
audio band.
The partial removal of dc carrying resistive loading of gain stages
can dramatically reduce THD in the main audio band without using loop
NFB. However, with triodes, the increasing in ohm value of the load
from say a paltry 3Ra to say 15Ra the internal gain of the triode
increases and therefore the application of the available local NFB
within the triode. Hence THD reduces.

For more ideas about implementing these notions, try http://www.turneraudio..com.au

I apologise for top posting, but using Google to post at r.a.t makes
it more difficult to set posts out tidily. One has to post in a tiny
little window, and not all across the screen with full height.

Patrick Turner.


> group of American and British journalists the short tour. The
> Americans were interested in the two nubile virgins the loincloth-clad
> old sage slept with every evening to keep him warm and to test his
> willpower. The British, still hungry after war shortages, wanted to
> know what Gandhi ate.
>
> 'Only the simplest food,' the PR man said. 'A little lemon and honey
> with soda water.'
>
> 'Wow,' said the Brits and moved on. (This was before their tabloids
> became the scummiest in the world.)
>
> Afterwards the Time-Life reporter hung back. 'Aren't lemon and honey
> and soda water very great luxuries in India?' he asked cynically. 'How
> can you be so hypocritical as to describe that as simplicity?'
>
> The PR man, who had been an Inner Temple barrister with Gandhi in
> London a few decades before, knew when he was caught out. 'Quite,' he
> said smoothly, 'but I never said it didn't cost a fortune to keep
> Ghandiji in simplicity.'
>
> ***
>
> Ultrafi tube amps are like that. The simpler they appear, the greater
> the mental effort required to win the benefits of that simplicity.
> There hangs about the apparent simplicity of The KISS Amp 300B a whole
> miasma of invisible capacitances which can test one's willpower far
> more than two naked nubile virgins.
>
> Although we commonly say that 'a Class A power tube draws no current
> on its grid', a power tube requires current from the driver to
> overcome various capacitance which loiter with malicious intent in
> your amp. A useful shortcut to half a good answer is slew rate current
> (which tells us how fast a capacitance is charged up and discharged)
> and as usual experience provides the other half of the answer. The
> first half of the answer can be calculated with slew rate formulae and
> one empirical assumption generously suggested to me by Gordon Rankin
> (who isn't responsible for what I do with the information) on 15
> October 1996 while I was working on an 845 amp:
>
> ***
>
> The Slew Rate
> SR = 2*Pi*Bandwith*Vmax
> Where
> Bandwidth = 20kHz or whatever higher frequency your design will pass
> Vmax is the maximum voltage the stage will deliver
>
> The input capacitance
> Cin = (A + 1) Cgp + Cgc
> Where
> A = the gain of stage for which the input capacitance is being
> determined
> Cgp is the Capacitance of the Grid to Plate
> Cgc is the Capacitance of the Grid to Cathode
>
> The Slew Rate Current
> SRC = Cin*SR
>
> Erno Borbely, Jung and Ron Gunzler (according to Gordon Rankin)
> suggest that the stage current should be 5 times the slew rate current
> to overcome the input capacitance of the next stage.
>
> The Stage Current
> Scur = SRC * K
> where
> K = 5
>
> ***
>
> Bear with me while I put in some numbers:
>
> Desired Stage Current is therefore 5*Cin*SR.
>
> For a 300B,
> Cin =9 + (1 + 3.85)*15 = 81.85pf
>
> For a fullrange amp with 80V signal voltage,
> Slew Rate = 2*3.14*20000*80 = 10,048,000
> (note that to arrive at an irreducible minimum we enter only the audio
> range)
>
> therefore
>
> Desired Stage Current = 5*81.85*10,048,000
>
> and after moving the decimal we get
>
> 4.1mA
>
> ***
>
> Mmm. That is of course an absolute minimum requirement. Notice
> something above? No one has yet counted stray capacitances. In a
> simple 300B amp strays usually amount to between 15 and 25pf. They too
> have to dealt with, which is why the constant is a multiple.
>
> We know from experience that a 300B likes more current on the plate of
> the driver than 4mA. It was the excellent Steve Bench who first
> suggested to me that 8mA is the right minimum number for a driver for
> 300B. In fact, this is less of discrepancy between theory and practice
> than at first appears. We would normally design the amp out to at
> least 40kHz, not just 20kHz, so the calculated current requirement
> would match the 8mA from experience without altering the famous
> constant of 5.
>
> Do the calculations for an 845, which has a difficult signal
> requirement of around 150V in the more common designs, and you will
> discover that 20mA is just about a minimum on the driver, which
> accounts for why some of us laugh when we see designs for 12AX7
> driving 845. It also accounts for why I like to drive kilovolt
> transmitting tubes with a 300B booster amp, or at least a power tube
> for a driver. The KISS Amp 300B is just such a booster design,
> requiring only a switch to turn the primary of the output transformer
> into a choke load on the plate of the 300B and a polyprop cap in
> series on the output to the grid of the main kilovolt amp. In other
> words, the 300B is used as a preamp (control amp) tube and as a weapon
> to absolutely murder Miller.
>
> ***
>
> Now we come to the Miller Effect, because we must. Even the exemplary
> RDH treats this distasteful subject, somewhat akin to removing the
> nightsoil, less thoroughly than I do here, and believe me, I'm doing
> the minimum I can get away with and still claim a modest electronic
> respectability.
>
> Miller is a multiplicatory part of one of three shunt (read "unwanted
> but unavoidable") capacitances in an amplifier:
> Cpc, the output capacitance of the first stage (from plate to cathode)
>
> Cin, the input capacitance of the second stage, which is Cgc (from
> grid to cathode) augmented by the Miller Effect
>
> Cstray, stray wiring capacitance
> These shunt capacitances add up, with the numerals indicating the
> first and second tubes:
> Cs = Cpc1 + Cin2 + *Cstray
> where
> Cin = Ggc2 + Cgp2(A2 +1)
> A2 is the amplification factor of the second tube
> Cgp2 is the grid to plate capacitance of the second tube.
>
> The Miller Effect is the Cgp2(A2 +1) part of Cin. It is clearly
> important because the grid to plate capacitance is multiplied by the
> amplification factor of the tube.
>
> The high frequency rolloff of a stage due to shunt capacitance is
> f =1/(2*pi*R’eq*Cs)
> where the effective load on the plate is
> R’eq = Rg+((Rp*RL)/(Rp+RL))
> You can see the important part played by the plate resistance and it
> isn't rocket science to see that the you should wire tidily and keep
> leads short to avoid stray capacitances.
>
> ***
>
> You can use the capacitances loitering in your computer constructively
> if you think laterally. For instance, the bandwidth of an amp should
> be balanced. Any extension below a rather high level, say 50Hz, should
> be matched by an extension at the top end. Conversely, if the bottom
> end is being deliberately sloped off early to protect horn drivers
> (which rapidly become unloaded below Fs), the HF should not be
> designed out to infinity, whatever the iron may be capable of; it
> should instead be balanced with the LF you are plotting. One of the
> tools under your control is the amount of current you put on the
> driver, and the slew rate concept is your tool to calculate it. That
> ensures that your amp reaches the bandwidth you design to.
>
> Miller gives you the tools to calculate, with data off the spec sheets
> for the two tubes and off the schematic of the circuit you have
> designed, and perhaps to ensure that the amp will not roll off under
> the bandwidth you are trying to arrange by providing slew current. The
> calculations are exceedingly tiresome but you can't afford to skip
> them. Later we will see how to automate all this with a spreadsheet.
>
> ***
>
> The question arises, why do you want an amp that has HF extension so
> far beyond your hearing? What is this 'balance' good for? Does
> something lie beyond the achievement of 'balance'? The answer is that
> there is a subjective effect, which ultrafidelista sometimes refer to
> as 'speed' or the 'fast amp' syndrome, and which novices think is only
> about undersizing power supply caps. Unfortunately building a 'fast'
> amp is far, far more complicated than that and definitely requires
> attention at both ends of frequency scale.
>
> The interrelationships of these factors are not well understood but
> they are likely to be complicated by the usual difficulties of
> psychoacoustics. That is one reason I somewhat dislike the sound bite
> of 'a fast amp' and prefer the more sober phrase 'a responsive amp'.
>
> The implication of this caution is that net gossip that "mo' driva
> current is betta current" is not necessarily true. There is a correct
> bandwidth for every speaker/amp combination, and particular correct
> lower and related correct upper frequencies for each application. If
> you just throw current at the driver because you have it to spare and
> the iron can make the frequency, that can easily be as harmful to the
> sound of your finished amp as not enough current.
>
> Don't skimp. Don't overdo it. Calculate thrice, solder once. An
> elegant sufficiency will give you the right sound.
>
> THE VOLTAGES IN THIS AMP WILL KILL YOU.
> GET EXPERIENCED SUPERVISION IF IT IS YOUR FIRST TUBE AMP
>
> All text and illustration is Copyright © Andre Jute 1996, 2005
> and may not be reproduced except in the thread KISS xxx on
> rec.audio.tubes
>
> From JUTE ON AMPShttp://www.audio-talk.co.uk/fiultra/JUTE%20ON%20AMPS.htm

[Patrick Turner]

On Feb 25, 4:29*pm, "Alex" > wrote:
> "Patrick Turner" > wrote in message
>
> ...
>
> I sympathize entirely with Mr J's ideas about Miller C and how it does
> audio not one bit of good.
>
> Alex:
>
> I found good use of the Miller effect. I can elaborate:
>
> To achieve stability one needs to roll of the loop gain before the OPT
> leakage inductance and winding capacitance kicks in. To do that some people
> use C or RC loads on the plate of a driver stage.
>
> This works, but still if the amp load varies in a wide range, unity gain
> bandwidthmight be too high when output is open circuit, and a bit small if
> someone connects too low-ohm speakers.

Triode output stages tend to have low gain variations whether loaded
or not.

So Williamson used the 4k7 + 470pF across the 47k load R for V1 1/2
6SN7.

But pentode output stage gain varies with load connected.

The use of a Zobel R&C across the OPT primary loads the OP stage at HF
where the stability is otherwise challenged by loading variations and
gain changes.

>
> To stabilise unity gain bandwidth one cam use the same approach as was used
> for op-amp compensation, namely a capacitor across an inverting gain stage.
> In a tube world this would be a capacitor between plate and grid of an
> output pentode. Typically it will be a quality 1...3kV rated 10...33pF
> ceramic cap. Multiplied by the Miller effect, it would be equivalent to
> 300...1000pF applied to the driver plate.

Yes, indeed, but usually such added shunt C solutions ruin the ability
of the amp to deal with pure C loads.

So I NEVER use a pure added C shunt of any place along the signal path
in the amp.

Often in SS amps where there is a huge open loop gain at 100Hz, there
is a C across the driver base to collector and open loop gain is
reduced from its maximum at 100Hz down at a rate of 20dB per decade so
that by 10kHz the applied GNFB is effectively much less than at 100Hz.

The dominant open loop phase shift is 90 degrees from well below 1kHz
but it is all removed by the action of GNFB so that maybe there is
only 10 degrees by 20 kHz.
But you can do this OK when there is no OPT in the signal path to be
included in the ZNFB loop

> Now unity gain bandwidth is
> defined by transconductance of the driver stage and this capacitor. Output
> pentode gain almost does not matter.
>
> Indeed if the amp is unloaded, and the gain is very high, the Miller effect
> will be stronger, and vica versa, thus unity gain bandwidth is stable. This
> approach almost makes redundant an RC Zobel at the OPT primary, with some
> reservation: you shall not pulling tubes out of the sockets on the working
> amp. If the output pentode current is interrupted, and the amp is unloaded,
> it may cause a *huge* inductive voltage spike causing breakdown of the OPT
> primary insulation or a spark of some sort.

Well, I don't follow all that. But i like Zobels across OPT primaries
or secondaries.


>
> Such Miller frequency compensation is especially useful in the amps where
> the volume control is connected not between the signal source and the
> ground, but between the signal source and the output speaker terminal. (Such
> ingenious solutions were common in old radios.)

Yes indeed. I have often found the volume control is a variable R in a
shunt FB network.
It works remarkably well if done right.
Crown also used such a scheme around opamps in preamps.



>At low volume such amp would
> work with nearly 100% feedback, which could reach 30...40dB. Almost no hum
> and noise! At volume control at full, the feedback will be around 10...20dB.

And ppl rarely used full volume. And radios were pretty awful
concerning noise.


>
> Without the "Miller-effect" frequency compemsation technique described above
> it is very difficult to achieve stability in such wide range of feedback and
> load from 3ohm to infinity.

Indeed, and especially so with tube amps with more parasitic L and C
elements to cause phase shift close to the audio band.

Patrick Turner.
>
> Regards,
> Alex

[Alex]

"Patrick Turner" > wrote in message
...

I sympathize entirely with Mr J's ideas about Miller C and how it does
audio not one bit of good.

Alex:

I found good use of the Miller effect. I can elaborate:

To achieve stability one needs to roll of the loop gain before the OPT
leakage inductance and winding capacitance kicks in. To do that some people
use C or RC loads on the plate of a driver stage.

This works, but still if the amp load varies in a wide range, unity gain
bandwidthmight be too high when output is open circuit, and a bit small if
someone connects too low-ohm speakers.

To stabilise unity gain bandwidth one cam use the same approach as was used
for op-amp compensation, namely a capacitor across an inverting gain stage.
In a tube world this would be a capacitor between plate and grid of an
output pentode. Typically it will be a quality 1...3kV rated 10...33pF
ceramic cap. Multiplied by the Miller effect, it would be equivalent to
300...1000pF applied to the driver plate. Now unity gain bandwidth is
defined by transconductance of the driver stage and this capacitor. Output
pentode gain almost does not matter.

Indeed if the amp is unloaded, and the gain is very high, the Miller effect
will be stronger, and vica versa, thus unity gain bandwidth is stable. This
approach almost makes redundant an RC Zobel at the OPT primary, with some
reservation: you shall not pulling tubes out of the sockets on the working
amp. If the output pentode current is interrupted, and the amp is unloaded,
it may cause a *huge* inductive voltage spike causing breakdown of the OPT
primary insulation or a spark of some sort.

Such Miller frequency compensation is especially useful in the amps where
the volume control is connected not between the signal source and the
ground, but between the signal source and the output speaker terminal. (Such
ingenious solutions were common in old radios.) At low volume such amp would
work with nearly 100% feedback, which could reach 30...40dB. Almost no hum
and noise! At volume control at full, the feedback will be around 10...20dB.

Without the "Miller-effect" frequency compemsation technique described above
it is very difficult to achieve stability in such wide range of feedback and
load from 3ohm to infinity.

Regards,
Alex