[Engineer[_2_]]

This one's for Patrick in Oz.
Hi, Patrick,
I've looked at your "standard" VLF stabilization coupling network for
this amplifier class (0.05 parallel 1Mohm, with 220Kohm grid leak on
the o/p tubes.) It appears to provide a forward gain roll-off below
about 15 Hz with a shelf at about 1/6 of the normal gain (-15.5 dB)
below about 3.2 Hz. Assuming my calculations are correct, what is the
purpose of the 1 Mohm across the 0.05 uF? Surely just letting the roll-
off continue at 6 dB/octave below 15 Hz would scotch any VLF
oscillation. Is this a phase shift issue? The simple -6dB /octave
roll-off leave a 90 degree phase advance in place below, say, 10 Hz
whereas the shelf would appear to avoid that (I'm not certain as,
regrettably, no pspice to hand at present!) I stabilized my
Williamson clone with just 0.05 uF coupling caps to the 6L6's with an
"aggressive" 100K grid leak (- 3dB at 32 Hz, but I may increase the
100K a bit....) Do I really need a VLF shelf below that?
Thanks and cheers,
Roger

[John L Stewart]

Originally Posted by Engineer[_2_]

This one's for Patrick in Oz.
Hi, Patrick,
I've looked at your "standard" VLF stabilization coupling network for
this amplifier class (0.05 parallel 1Mohm, with 220Kohm grid leak on
the o/p tubes.) It appears to provide a forward gain roll-off below
about 15 Hz with a shelf at about 1/6 of the normal gain (-15.5 dB)
below about 3.2 Hz. Assuming my calculations are correct, what is the
purpose of the 1 Mohm across the 0.05 uF? Surely just letting the roll-
off continue at 6 dB/octave below 15 Hz would scotch any VLF
oscillation. Is this a phase shift issue? The simple -6dB /octave
roll-off leave a 90 degree phase advance in place below, say, 10 Hz
whereas the shelf would appear to avoid that (I'm not certain as,
regrettably, no pspice to hand at present!) I stabilized my
Williamson clone with just 0.05 uF coupling caps to the 6L6's with an
"aggressive" 100K grid leak (- 3dB at 32 Hz, but I may increase the
100K a bit....) Do I really need a VLF shelf below that?
Thanks and cheers,
Roger

One needs to be a bit careful with these cct values since they may be in order for Patrick T's OPTs. Your Hammond will probably benifit from a shelf at a higher f since Pat Ts transformers are sure to have a much larger primary inductance. The loudspeaker resonance will complicate things since it is in the same region you are trying to fix.

I've successfully used the parallel RC with differential amp driver to avoid the additional phase shift at LF. But with the split load phase inverter is a problem.

Patrick, your thoughts?

Cheers, John

[Patrick Turner]

On Jun 18, 2:39*am, Engineer wrote:
This one's for Patrick in Oz.
Hi, Patrick,
I've looked at your "standard" VLF stabilization coupling network for
this amplifier class (0.05 parallel 1Mohm, with 220Kohm grid leak on
the o/p tubes.) *It appears to provide a forward gain roll-off below
about 15 Hz with a shelf at about 1/6 of the normal gain (-15.5 dB)
below about 3.2 Hz. *Assuming my calculations are correct, what is the
purpose of the 1 Mohm across the 0.05 uF? Surely just letting the roll-
off continue at 6 dB/octave below 15 Hz would scotch any VLF
oscillation. *Is this a phase shift issue? *The simple -6dB /octave
roll-off leave a 90 degree phase advance in place below, say, 10 Hz
whereas the shelf would appear to avoid that (I'm not certain as,
regrettably, no pspice to hand at present!) *I stabilized my
Williamson clone with just 0.05 uF coupling caps to the 6L6's with an
"aggressive" 100K grid leak (- 3dB at 32 Hz, but I may increase the
100K a bit....) *Do I really need a VLF shelf below that?
Thanks and cheers,
Roger

The Wiiliamson has two RC coupled stages and OPT in the loop
surrounded by NFB so it is very prone to LF oscillations, especially
if you have a preamp powered off the pwr amp PSU and then you try to
boost the bass with a tone control network.

Williamson stipulated that the OPT have at least 100H at Vaa = 5Vrms
to get LF stability.

Nearly all OPTs made since 1949 have been by ppl who thought
Williamson was a ****** who made the amp cost of production too high.
These ppl didn't use enough P turns or big enough core size so lots of
Williamsons actually do oscillate at LF but it went un-noticed by the
dumb DIYers of yesteryear because they had no oscilloscopes to see the
small oscillations at below 10Hz. OPT core permeability rises with
applied voltage and so does Lp so the amplitude of LF oscillations may
be kept low if Lp rises enough to prevent any increase in amplitude.
Many people used CR coupilng values which encouraged phase shift
caused oscillations.
This is the background theory of how the Williamson and many other
tube amps become a phase shift oscillator at LF.

In a williamson with V1 direct coupled to V2 concertina and CR coupled
balanced amp with CR coupled OP tubes, the best place to put a
shelving network is between concertina and balanced amp, and there
must be TWO networks, one from concertina anode and one from cathode.

OK, the 0.47 + 1M strapped with 0.047 + 220k acts like this :-
At 1 kHz, all C have low Z so the phase shift is low and the driver,
ie concertina or whatever you have "sees" a load of 220k. As F is
reduced, there is a pole between 0.047 and 220 at 15.4Hz which is
usually above the F at which the amp may want to oscillate. Sometimes
I have used 0.022uF, so pole is at 32Hz. phase shift caused is less
than 90 degrees. As F is further reduced, the network response tends
to flatten to a shelf formed by 1M and 220k, ie, signal flattens at
roughly -15db, 0.18 times the 1kHz level. The 0.047 has gone to an
open circuit by 2Hz with little effect. But the 0.47 the rolls the
response off at the pole between 0.47uF and 1.22M ohms, ie at 0.27Hz,
below which there is an ultimate phase shift of 90d, but at such a low
F as to not cause bothers because its well below the poles of all
other sages.

The Williamson amp published in August 1949 has CR before balanced amp
with 0.05uF + 470k and before KT66 0.25uF + 100k. So poles are at
6.76Hz an 6.36Hz and so by 3Hz phase shift because of the two networks
would have been about 120d and with another 60d from OPT, maybe it
oscillated if the OPT didn't have enough Lp, and open loop gain 1
where 180d phase shift exists. Ppl tend to try using larger C values
only to find oscillation frequency Fo just goes lower, or smaller C
and Fo rises. But the GAIN at where Fo is likely to occur must be
lowered. The only good way is with a shelving network. This means the
open loop gain without GNFB, OLG, seems to roll off at say -3dB at
30Hz and ppl panic and say "OMG, I've lost the bass." But when they
connect the GNFB its nice and stable and -3db occurs at 7Hz and there
is no peak in the response below 30Hz.

What is happening with GNFB applied is that there is 20dB NFB applied
to all F down to about 35Hz below which the amount of NFB is reduced.
Maybe only 10dB at 10Hz, and hardly any at 5Hz, because OLG has been
reduced so much by the shelving network. But there is ENOUGH NFB being
applied at 10Hz to still get amp Rout quite low enough, and reduce THD
etc. Nobody has much of anything below 20Hz in music so signals below
20Hz are tiny so they don't create much THD/IMD so lots of NFB at 10Hz
is NOT required. This is especially valid where the OP stage is a
triode type which has Rout RL even with no GNFB. Williamson's
original KT66 triodes had Ra-a of 3k2, plus maybe Rw = 400 ohms, so DF
without NFB = 10k/3k6 = 2.77, and not bad, needing only 10dB NFB to
get DF to 10.

On page 346, RDH4, there is a Wlliamson with 807 in beam tetrode mode
which is VERY LIKELY to oscillate with the higher tetrode gain. CR
networks have 0.05 + 470k and 0.05 + 500k before 807. Looks like an
oscillator to me!

The Williamson rule was that there should be 100H for a load of 10k,
even at low Va-a levels. This meant RLa-a = Lp reactance at 15.9Hz. In
otherwords the amp will show less than 1 dB of response roll off due
to load reduction at 15.9Hz. Everyone mostly ROTFL at Willy, and they
made OPTs where ZLp = RL at 40Hz at low Va-a, and RLa-a was raised to
say 5k and tubes pushed into class AB with little class A, LF fidelity
dissappeared, power doubled though, bean conters were happy, and the
brandname sold yet another fraud to the public. One can build an amp
with only 20H of Lp, and the NFB will desperately try to compensate
the response towards flatness but don't expect the best bass
performance at loud levels and without core saturation effects. Unless
shelving networks are placed in such amps they will oscillate for
sure. Some makers avoided the issue by using only two amp stages, ie,
like Quad-II. This removed one of the 3 places where there could be an
ultimate 90d phase shift at F which is too high, ie, where OLG 1.0
and where total OLG phase shift 180d.

So Quad-II has a Tiny Toy OPT with 3,180 turns around a core Afe =
25mm x 25mm, while the Williamson original had 4,400t around 44 x 32.
I'd suggest the Quad-II has a lot less Lp than Williamson does, and I
know that the Quad-II saturates at 49Hz at 420Va-a while the
Williamson saturates at 15Hz at 420Va-a. The ratio of Lp to RL in the
original Williamson is much BETTER than in the Quad-II.

But Quad-II have 0.1uF + 680k between EF86 drivers and KT66, so as
years pass Eg1 goes positive without the 100k needed to stop the Eg1
rise. The Rout of EF86 is rather high, determined by the 180k and 680k
and EF86 Ra all in parallel, about 140k, so the pole in CR is at
1.9Hz, and low enough. Walker couldn't use less tha 680k for KT66 Rg
because that'd load the EF86 down an prevent the gain he needed. So
QUad-II has a really bad bodgie designed network in place. The triode
driver of the williamson allows the Rg = 100k without gain loss, and
the low Rg makes the KT66 last a long time. Leak also had bad design
problems. None of those old British brands got everything right.

Most of what I have said is incomprehensible. I bet you are entirely
baffled. You'll never learn until you build and measure everything
while asking questions and while sitting with a dual trace
oscilloscope to SEE the phase shifts and gain changes in networks. Its
real basic stuff and to build good amps you must know all about it.
After building 10 amps, maybe the penny will drop. Most Diyers build
one amp, and then forget the way they muddled through the process.
Read this page carefully, http://www.turneraudio.com.au/basic-tube-%283%29.htm

Its about time I renovated the page. I dunno how the "-%283%29" got
into the title line, years go by, I change ISP, and **** happens to
titles and text and formatting. I am re-editing pages now, lots to do.

Patrick Turner.

[Alex Pogossov]

"Patrick Turner" wrote in message
...

OK, the 0.47 + 1M strapped with 0.047 + 220k acts like this :-
At 1 kHz, all C have low Z so the phase shift is low and the driver,
ie concertina or whatever you have "sees" a load of 220k. As F is
reduced, there is a pole between 0.047 and 220 at 15.4Hz which is
usually above the F at which the amp may want to oscillate. Sometimes
I have used 0.022uF, so pole is at 32Hz. phase shift caused is less
than 90 degrees. As F is further reduced, the network response tends
to flatten to a shelf formed by 1M and 220k, ie, signal flattens at
roughly -15db, 0.18 times the 1kHz level. The 0.047 has gone to an
open circuit by 2Hz with little effect. But the 0.47 the rolls the
response off at the pole between 0.47uF and 1.22M ohms, ie at 0.27Hz,
below which there is an ultimate phase shift of 90d, but at such a low
F as to not cause bothers because its well below the poles of all
other sages.

Alex:
Shelving network does a good job, but it has a drawback. It is an attenuator
for low frequencies. If low frequency signal or DC step gets applied to the
input of the amp, the first stage will or might overload, while the shelving
circuit will protect the next stage from the overloading.

To prevent the front stage from overloading at VLF it is better to apply the
shelving as a local feedback, i.e. in the cathode circuit of the 1-st stage.
Imagine the cathode of the 1-st stage is not directly connected to a
feedback divider (say 1K/100R) but through a RC circuit of paralleled 10K
and 5uF (approx.) This cathode degeneration at LF will act similar to the
attenuator shelving, but will prevent overloading of the first stage by VLF.
Then you do not need 1M||0.05uF interstage attenuator.

Of course in this cathode shelving the grid DC bias voltage has to be
elevated, so the grid can not be directly connected to the volume control.
However it is not a big deal. there are three known ways of elevating grid
bias voltage:
- fixed divider from upply voltage;
- divider cathode-grid-ground;
- split cathode resistor and from the tap thus formed throw 1M to the grid.

Regards,
Alex

[Alex Pogossov]

"Engineer" wrote in message
...
This one's for Patrick in Oz.
Hi, Patrick,
I've looked at your "standard" VLF stabilization coupling network for
this amplifier class (0.05 parallel 1Mohm, with 220Kohm grid leak on
the o/p tubes.) It appears to provide a forward gain roll-off below
about 15 Hz with a shelf at about 1/6 of the normal gain (-15.5 dB)
below about 3.2 Hz. Assuming my calculations are correct, what is the
purpose of the 1 Mohm across the 0.05 uF? Surely just letting the roll-
off continue at 6 dB/octave below 15 Hz would scotch any VLF
oscillation. Is this a phase shift issue? The simple -6dB /octave
roll-off leave a 90 degree phase advance in place below, say, 10 Hz
whereas the shelf would appear to avoid that (I'm not certain as,
regrettably, no pspice to hand at present!) I stabilized my
Williamson clone with just 0.05 uF coupling caps to the 6L6's with an
"aggressive" 100K grid leak (- 3dB at 32 Hz, but I may increase the
100K a bit....) Do I really need a VLF shelf below that?
Thanks and cheers,
Roger

Your agressive LF frequency compensation might cause undesirable effects.

1. 100K bias is bit too low -- loading preceeding stage (driver, phase
splitter) unnecessarily. Perhaps you can achieve the same result by 220K and
0.022uF coupling.

2. By agressively cutting LF from 32Hz you reduce loop gain in the working
range (20Hz) thus increasing output impedance, reducing speaker damping,
increasing intermodulation and distortion (if it matters at 20Hz?) This is
aggravated by the fact that the driver stage has to labour 3dB harder at
20Hz because of the attenuation in your agressive circuit.. So I bet you
will get overall results twice worse than Mr Turner would have done with his
smart shelving.

3. Though your amp might appear stable, most likely it will be peaking close
to oscillation at 8Hz or so where the +90deg lead from your agressive
circuit will combine with +90deg lead from your OPT at the 0dB loop gain
crossing. Any LF rumble might drive your amp into overload.

So I need to admit, shelving is wiser, because it does not take out precious
dBs from the loop gain in the whole audio range and gives a better phase
margin, and no peaking. Mr Turner makes a deep 12dB/octave nose dive below
15...20Hz by combining OPT effect with 220K/0.05uF effect, but then, closer
to 0dB crossing, he gently goes out of the deep nose dive, shelves the beast
out and happily crossing 0dB at 8dB/octave. Past that he can dive deep
again -- does not carem as no oscillation will occur below 0dB.

However, both shelving and "agressive" compensation have one drawback in
common. NFB gain is constant at all frequencies since NFB is simply a
resistive divider. Thus the NFB tries to have the amp's responce flat (at
low level of course) down to possibly 3Hz or so. This is crasy and
unnecessary.

For that reason, instead of shelving interstage coupling I shelf the NFB.
Instead of a resisive divider I would put a resistor (typically 100R) in
series with a capacitor (order of 47...100 uF). Interstage coupling is kept
a simple semi-agressive RC circuit (say 0.022uF and 330K). Overall loop gain
shelving effect is the same as in Mr Turner desihns, but because of the
deepening feedback at VLF the amp becomes a rumble filter itself. A drawback
of that method is that you need an electrolytic in the feedback, Some people
do not "trust" electrolytics as a frequency shaping components.

Regards,
Alex

[Patrick Turner]

On Jun 18, 7:01*pm, "Alex Pogossov" wrote:
"Engineer" wrote in message

...





This one's for Patrick in Oz.
Hi, Patrick,
I've looked at your "standard" VLF stabilization coupling network for
this amplifier class (0.05 parallel 1Mohm, with 220Kohm grid leak on
the o/p tubes.) *It appears to provide a forward gain roll-off below
about 15 Hz with a shelf at about 1/6 of the normal gain (-15.5 dB)
below about 3.2 Hz. *Assuming my calculations are correct, what is the
purpose of the 1 Mohm across the 0.05 uF? Surely just letting the roll-
off continue at 6 dB/octave below 15 Hz would scotch any VLF
oscillation. *Is this a phase shift issue? *The simple -6dB /octave
roll-off leave a 90 degree phase advance in place below, say, 10 Hz
whereas the shelf would appear to avoid that (I'm not certain as,
regrettably, no pspice to hand at present!) *I stabilized my
Williamson clone with just 0.05 uF coupling caps to the 6L6's with an
"aggressive" 100K grid leak (- 3dB at 32 Hz, but I may increase the
100K a bit....) *Do I really need a VLF shelf below that?
Thanks and cheers,
Roger

Your agressive LF frequency compensation might cause undesirable effects.

1. 100K bias is bit too low -- loading preceeding stage (driver, phase
splitter) unnecessarily. Perhaps you can achieve the same result by 220K and
0.022uF coupling.

But with 6SN7, and dc carrying RL = say 39k then total RL = 28k, and
this is about 3 x Ra of the tube which is fine. Bias R up to 150k in
OP stage is about right. I've seen too many tubes with several volts
dc across the high value bias resistance, so the tube is being turned
on by this positive bias and things only get worse if the tube heats
up more - there is a positive bias effect.
Just using 220k and 0.022uF from driver anode to output grid gives
pole at 32Hz, and 88d phase shift at say
8 Hz, maybe it oscillates.

2. By agressively cutting LF from 32Hz you reduce loop gain in the working
range (20Hz) thus increasing output impedance, reducing speaker damping,
increasing intermodulation and distortion (if it matters at 20Hz?) This is
aggravated by the fact that the driver stage has to labour 3dB harder at
20Hz because of the attenuation in your agressive circuit.. So I bet you
will get overall results twice worse than Mr Turner would have done with his
smart shelving.

The shelving leaves some gain available and reduces phase shift. So
usually there is enough FB operative at 20Hz, and with lower phase
shift at 20Hz its more effective than otherwise.

3. Though your amp might appear stable, most likely it will be peaking close
to oscillation at 8Hz or so where the +90deg lead from your agressive
circuit will combine with +90deg lead from your OPT at the 0dB loop gain
crossing. Any LF rumble might drive your amp into overload.

That's why I have referred him to some typical response graphs at my
website.

So I need to admit, shelving is wiser, because it does not take out precious
dBs from the loop gain in the whole audio range and gives a better phase
margin, and no peaking. Mr Turner makes a deep 12dB/octave nose dive below
15...20Hz by combining OPT effect with 220K/0.05uF effect, but then, closer
to 0dB crossing, he gently goes out of the deep nose dive, shelves the beast
out and happily crossing 0dB at 8dB/octave. Past that he can dive deep
again -- does not carem as no oscillation will occur below 0dB.

However, both shelving and "agressive" compensation have one drawback in
common. NFB gain is constant at all frequencies since NFB is simply a
resistive divider. Thus the NFB tries to have the amp's responce flat (at
low level of course) down to possibly 3Hz or so. This is crasy and
unnecessary.

Indeed. The trick is to get the shelving right, neither over done or
uderdone.
Same goes for HF shelving.

The other way to do LF shelving is to reduce the amount of NFB at LF
so you have a parallel network of C&R in series with the FB resistance
in the NFB divider network. I've never needed to use this method. Once
a shelving network is connected and a plotted F response shoes no
peaking outside the 20Hz to 20kHz band, it is wise to plot the
response at the anode output of V1 just before any shelving networks
in the input/driver line up between 1Hz and 1MHz. The signal from V1
output to OP tube grids is called the ERROR SIGNAL, because it
contains a pure version of the music signal PLUS a fraction of a phase
inverted version of the THD/IMD phase shift and any other artifact
generated by the amp. The Error Signal will never be a flat response
except in the middle of the AF band. The peaks in this signal at the
ends of the band and beyond the band should ideally not exceed 3dB
above the level in the centre of the band. Such peaking is inevitable
in most amps as the NFB causes more signal to be applied at band ends
to maintain the output level as flat.
But never should the peaking ever cause any input tube or driver tubes
to become overloaded, work into grid current, become cut off, etc.
Hence driver amps should always be able to make TWICE the voltage one
needs at the OP grids. When testing with a 5kHz square wave, then you
may see some huge peaks appear in the error signal. And especially at
the anodes of OP tubes. The amp is having troubles dealing with HF.
The shelving tends to prevent the amp from bothering to fix the HF
part if square waves above 20kHz, which is fine, lots of NFB above
20kHz, say at 60kHz, a typical HF oscillation frequency, is entirely
pointless.


For that reason, instead of shelving interstage coupling I shelf the NFB.

You have read my mind a bit.

Instead of a resisive divider I would put a resistor (typically 100R) in
series with a capacitor (order of 47...100 uF).

But what of the phase shift of that C? Isn't it better to have R&C in
parallel inserted from FB take off at OPT sec to the feedback R? Say
you have 1k0 and 100r as the normal FB divider so that 1/11 of the OPT
signal is applied to V1 cathode. Say one adds 3k3 so you then have
3k3, 1k0, then 100r at k to 0V at V1. Then ß becomes 0.022, much less
than 0.09, and at very low F there is no phase shift, so with less NFB
its probably going to be stable. But there isn't enough FB at higher F
so you shunt the 3k3 with say 6.8uF. So at 100Hz the 6.8uF = 233 ohms
reactance and 3k3 is well shunted.

Peaking still has to be checked in output and in error signal.



Interstage coupling is kept
a simple semi-agressive RC circuit (say 0.022uF and 330K). Overall loop gain
shelving effect is the same as in Mr Turner desihns, but because of the
deepening feedback at VLF the amp becomes a rumble filter itself. A drawback
of that method is that you need an electrolytic in the feedback, Some people
do not "trust" electrolytics as a frequency shaping components.

You can use 63V rated plastic caps OK. I doubt you need 47uF caps.

Switched networks in FB loops were often used for tone controls, not
my scene in power amps though.

Patrick Turner.



Regards,
Alex- Hide quoted text -

- Show quoted text -

[Alex Pogossov]

"Patrick Turner" wrote in message
...

Patrick:
But what of the phase shift of that C? Isn't it better to have R&C in
parallel inserted from FB take off at OPT sec to the feedback R? Say
you have 1k0 and 100r as the normal FB divider so that 1/11 of the OPT
signal is applied to V1 cathode. Say one adds 3k3 so you then have
3k3, 1k0, then 100r at k to 0V at V1. Then ß becomes 0.022, much less
than 0.09, and at very low F there is no phase shift, so with less NFB
its probably going to be stable. But there isn't enough FB at higher F
so you shunt the 3k3 with say 6.8uF. So at 100Hz the 6.8uF = 233 ohms
reactance and 3k3 is well shunted.

Peaking still has to be checked in output and in error signal.

Alex:
I suggested something quite opposite. The NFB divider looks like:
- 1K from the speaker terminal to cathode of the driver stage;
- 100R from the cathode to 100uF capacitor;
- the other end of the 100uF capacitor is tied to GND.

Thus the feedback "beta" increases (!) at low frequencies (below 10Hz in
this case), giving -90deg phase lag in the loop below 10Hz. This is in
effect turning the amp into a anti-rumble filter.

From the first glance it might sound crazy to increase the loop gain at VLF
where we want an overal reduction of yje loop gain, but consider this:
- OPT typically gives +90deg lead below 15...20Hz;
- interstage coupling (simple RC with no shelving, 220K and 0.033uF) is
calculated to give -3dB corner at say 15...20Hz and also gives +90 deg lead
below;
- but this "funny" NFB with a 100R and 100uF gives -90deg LAG below 10Hz!
And this lag maintains down until 1Hz!

At LF one lag subtracts from two leads and in combination we have only
+90deg below 10Hz. Therefore, 0dB line at about 3Hz (typically) will be
crossed safely at only +90deg!

Killing many birds with one stone:
- perfect transient with no peaking;
- natural low cut off in the whole amp (sort of built-in anti-rumble
filtering);
- reduced error signal at VLF as more feedback is applied;
- no need to use two identical shelving circuits in push-pull amps -- just
one extra electrolytic.

A drawback - a despised electrolytic as a shaping component.

[Patrick Turner]

On Jun 18, 9:56*pm, "Alex Pogossov" wrote:
"Patrick Turner" wrote in message

...

Patrick:
But what of the phase shift of that C? Isn't it better to have R&C in
parallel inserted from FB take off at OPT sec to the feedback R? Say
you have 1k0 and 100r as the normal FB divider so that 1/11 of the OPT
signal is applied to V1 cathode. Say one adds 3k3 so you then have
3k3, 1k0, then 100r at k to 0V at V1. Then ß becomes 0.022, much less
than 0.09, and at very low F there is no phase shift, so with less NFB
its probably going to be stable. But there isn't enough FB at higher F
so you shunt the 3k3 with say 6.8uF. So at 100Hz the 6.8uF = 233 ohms
reactance and 3k3 is well shunted.

Peaking still has to be checked in output and in error signal.

Alex:
I suggested something quite opposite. The NFB divider looks like:
- 1K from the speaker terminal to cathode of the driver stage;
- 100R from the cathode to 100uF capacitor;
- the other end of the 100uF capacitor is tied to GND.

My apologies, I didn't understand you. And the reason I didn't
understand was because the technique you are suggesting is completely
unknown in the range of traditional means of applying NFB in tube
amps.
Your method here is normal procedure in solid state amps which are all
dc coupled and without any large phase shifts at very low frequencies
down to DC.

Thus the feedback "beta" increases (!) at low frequencies (below 10Hz in
this case), giving -90deg phase lag in the loop below 10Hz. This is in
effect turning the amp into a anti-rumble filter.

Yes, I see you are against rumble. But in any tube amp with 2 CR
couplings and an OPT the gain below 10Hz quickly falls to zero at DC
and there is no need to roll off LF any more than it naturally is
rolled off. So if a signal at 1Hz enters the amp, the FB fed back is
extremely small because the output signal is so small.
So I cannot see your method would improve LF behaviour. I've read most
of the texts about NFB applications in tube amps and I don't recall a
single instance where your idea has been applied by any commercial
manufacturer nor have I read of anyone supporting it.


From the first glance it might sound crazy to increase the loop gain at VLF
where we want an overal reduction of yje loop gain, but consider this:
- OPT typically gives +90deg lead below 15...20Hz;
- interstage coupling (simple RC with no shelving, 220K and 0.033uF) is
calculated to give -3dB corner at say 15...20Hz and also gives +90 deg lead
below;
- but this "funny" NFB with a 100R and 100uF gives -90deg LAG below 10Hz!
And this lag maintains down until 1Hz!

Well, your NFB method does not increase OLG. OLG remains what it is
regardless of the NFB network. But th closed loop gain, CLG, could be
boosted with positive FB. 3 lots of reactive phase advances in the amp
will add to a rapid phase turnover per octave below 10Hz. So maximum
phase shift can be more than 180d then because the Lp inductance falls
to near zero near dc, you have only dc winding resistance in the OP
stage so phase shift lessens a bit towards DC. But exactly what anyone
measures depends on voltage applied and fiddling around with trying to
increase NFB as one gets close to DC makes no sense to me at all.

When you have measured and demonstrated your technique and analysised
it all with maybe 10 detailed pages on a website with photos of the
amps, then ppl might say you have something to offer. People here are
very difficult people. We cannot agree with anything anyone says
unless they offer the truth, the whole truth, and nothing but the
truth, so help them in the eyes of the God Of Triodes :-)


At LF one lag subtracts from two leads and in combination we have only
+90deg below 10Hz. Therefore, 0dB line at about 3Hz (typically) will be
crossed safely at only +90deg!

I'll believe works when I see it.

Killing many birds with one stone:
- perfect transient with no peaking;
- natural low cut off in the whole amp (sort of built-in anti-rumble
filtering);
- reduced error signal at VLF as more feedback is applied;
- no need to use two identical shelving circuits in push-pull amps -- just
one extra electrolytic.

A drawback - a despised electrolytic as a shaping component.

Perhaps there are other drawbacks you have not thought about.

I suggest you embark on a course of soldering in your laboratory to
prove your idea works. We all look forward to results.

Patrick Turner.

[Engineer[_2_]]

On Jun 18, 7:22*pm, Patrick Turner wrote:
On Jun 18, 9:56*pm, "Alex Pogossov" wrote:









"Patrick Turner" wrote in message

....

Patrick:
But what of the phase shift of that C? Isn't it better to have R&C in
parallel inserted from FB take off at OPT sec to the feedback R? Say
you have 1k0 and 100r as the normal FB divider so that 1/11 of the OPT
signal is applied to V1 cathode. Say one adds 3k3 so you then have
3k3, 1k0, then 100r at k to 0V at V1. Then ß becomes 0.022, much less
than 0.09, and at very low F there is no phase shift, so with less NFB
its probably going to be stable. But there isn't enough FB at higher F
so you shunt the 3k3 with say 6.8uF. So at 100Hz the 6.8uF = 233 ohms
reactance and 3k3 is well shunted.

Peaking still has to be checked in output and in error signal.

Alex:
I suggested something quite opposite. The NFB divider looks like:
- 1K from the speaker terminal to cathode of the driver stage;
- 100R from the cathode to 100uF capacitor;
- the other end of the 100uF capacitor is tied to GND.

My apologies, I didn't understand you. And the reason I didn't
understand was because the technique you are suggesting is completely
unknown in the range of traditional means of applying NFB in tube
amps.
Your method here is normal procedure in solid state amps which are all
dc coupled and without any large phase shifts at very low frequencies
down to DC.



Thus the feedback "beta" increases (!) at low frequencies (below 10Hz in
this case), giving -90deg phase lag in the loop below 10Hz. This is in
effect turning the amp into a anti-rumble filter.

Yes, I see you are against rumble. But in any tube amp with 2 CR
couplings and an OPT the gain below 10Hz quickly falls to zero at DC
and there is no need to roll off LF any more than it naturally is
rolled off. So if a signal at 1Hz enters the amp, the FB fed back is
extremely small because the output signal is so small.
So I cannot see your method would improve LF behaviour. I've read most
of the texts about NFB applications in tube amps and I don't recall a
single instance where your idea has been applied by any commercial
manufacturer nor have I read of anyone supporting it.



From the first glance it might sound crazy to increase the loop gain at VLF
where we want an overal reduction of yje loop gain, but consider this:
- OPT typically gives +90deg lead below 15...20Hz;
- interstage coupling (simple RC with no shelving, 220K and 0.033uF) is
calculated to give -3dB corner at say 15...20Hz and also gives +90 deg lead
below;
- but this "funny" NFB with a 100R and 100uF gives -90deg LAG below 10Hz!
And this lag maintains down until 1Hz!

Well, your NFB method does not increase OLG. *OLG remains what it is
regardless of the NFB network. But th closed loop gain, CLG, could be
boosted with positive FB. 3 lots of reactive phase advances in the amp
will add to a rapid phase turnover per octave below 10Hz. So maximum
phase shift can be more than 180d then because the Lp inductance falls
to near zero near dc, you have only dc winding resistance in the OP
stage so phase shift lessens a bit towards DC. But exactly what anyone
measures depends on voltage applied and fiddling around with trying to
increase NFB as one gets close to DC makes no sense to me at all.

When you have measured and demonstrated your technique and analysised
it all with maybe 10 detailed pages on a website with photos of the
amps, then ppl might say you have something to offer. People here are
very difficult people. We cannot agree with anything anyone says
unless they offer the truth, the whole truth, and nothing but the
truth, so help them in the eyes of the God Of Triodes :-)

At LF one lag subtracts from two leads and in combination we have only
+90deg below 10Hz. Therefore, 0dB line at about 3Hz (typically) will be
crossed safely at only +90deg!

I'll believe works when I see it.



Killing many birds with one stone:
- perfect transient with no peaking;
- natural low cut off in the whole amp (sort of built-in anti-rumble
filtering);
- reduced error signal at VLF as more feedback is applied;
- no need to use two identical shelving circuits in push-pull amps -- just
one extra electrolytic.

A drawback - a despised electrolytic as a shaping component.

Perhaps there are other drawbacks you have not thought about.

I suggest you embark on a course of soldering in your laboratory to
prove your idea works. We all look forward to results.

Patrick Turner.

Thanks for all the valuable comment. Here's what I've gleaned from
this so far...
A VLF shelf has merit in that it leaves some Global NFB at VLF and
mitigates the -90 deg phase shift at critical instability frequencies
that adds to the -90 deg in the OPT and the -90 in the other RC
coupler... easy to get to 180 ! This suggest use large C's (and
normal Rg's) inside the NFB loop except for ONE shelf network.
My Fisher OPT is presumably "medium quality" (is that more or less
than the Hammond 1620, 1650F, 1645, etc. series?) in that it came from
a claimed 30 watt RMS Fisher amp (model KX-200PP, not mine... OPT
bought on eBay.)
My 100K grid resistor on the 6L6's is too low... I'll up it to 220
Kohms and recalculate the C for the shelf I need.
Rumble will be cut off in the pre-amp so none hits the PA.
BTW, I do know about phase shift... also gain and phase margins,
Nyquist diagrams, PID control, feedback controller tuning and
stability, open and closed loop Bode plots... and the whole ball of
wax of stochastic sampled data control (off topic!), as I was a
practicing control systems engineer for some 30 years. But the last
18 years were spent in engineering management (pays better!), so you
get a bit rusty, but I still knew when any staffers tried to BS me...
fortunately, not often, as I never forgot the basics! Now retired.
However, tube audio is pretty straightforward in principle but, of
course, needs close design attention or it will bite you.
Again, my thanks to all, particularly Patrick. We shall overcome!
Cheers,
Roger

[Alex Pogossov]

"Patrick Turner" wrote in message
news:2169d98c-5935-4e7c-

When you have measured and demonstrated your technique and analysised
it all with maybe 10 detailed pages on a website with photos of the
amps, then ppl might say you have something to offer. People here are
very difficult people. We cannot agree with anything anyone says
unless they offer the truth, the whole truth, and nothing but the
truth, so help them in the eyes of the God Of Triodes :-)


At LF one lag subtracts from two leads and in combination we have only
+90deg below 10Hz. Therefore, 0dB line at about 3Hz (typically) will be
crossed safely at only +90deg!

I'll believe works when I see it.

Killing many birds with one stone:
- perfect transient with no peaking;
- natural low cut off in the whole amp (sort of built-in anti-rumble
filtering);
- reduced error signal at VLF as more feedback is applied;
- no need to use two identical shelving circuits in push-pull amps -- just
one extra electrolytic.

A drawback - a despised electrolytic as a shaping component.

Perhaps there are other drawbacks you have not thought about.

I suggest you embark on a course of soldering in your laboratory to
prove your idea works. We all look forward to results.

Alex:
I am not into building audio tube amps, because it is crazy to do while SS
works (or can potentially work) much better in all respects, apart from
creating a warm fuzzy feeling. But I restore and improve radios, and use
this feedback increase at LF technique. Of course there is no rumble in an
AM receiver, but fadings, beating of two stations on close but not
synchronised carriers and simply skimming the band makes lots of VLF.

As you know to minimise AF load on the AM detector, it is not uncommon to
have grid leak in the first audio stage of 10M with 0.05uF of coupling cap.
This huge time constant makes the first tube (6SJ7, e.g.) virtually open to
DC. VLF undulations result in 30V swings on the plate of the 1-st stage
bringing it close to saturation. Part of this reaches the grid of the power
stage (say, 6V6) unnecessarily swinging its bias and plate current. The
later in turn cause undulations on the lightly filtered +B rail, and even
might fall into resonance with the supply CLC filter.

I am not very impressed with the shelving approach, because it is only an
attenuator. While it would cut off 6V6 grid excursions, it will not prevent
from 6SJ7 front overloading.

Instead I use a circuit which helps reduce VLF voltage applied between grid
and cathode of the first stage 6SJ7.

You would say: Use a low cut-off filter. But it is a complication. Much
better is to turn the amp into an active high pass filter by means of the
NFB.

[John Byrns]

In article ,
Patrick Turner wrote:

On Jun 18, 9:56*pm, "Alex Pogossov" wrote:
"Patrick Turner" wrote in message

...

Patrick:
But what of the phase shift of that C? Isn't it better to have R&C in
parallel inserted from FB take off at OPT sec to the feedback R? Say
you have 1k0 and 100r as the normal FB divider so that 1/11 of the OPT
signal is applied to V1 cathode. Say one adds 3k3 so you then have
3k3, 1k0, then 100r at k to 0V at V1. Then ß becomes 0.022, much less
than 0.09, and at very low F there is no phase shift, so with less NFB
its probably going to be stable. But there isn't enough FB at higher F
so you shunt the 3k3 with say 6.8uF. So at 100Hz the 6.8uF = 233 ohms
reactance and 3k3 is well shunted.

Peaking still has to be checked in output and in error signal.

Alex:
I suggested something quite opposite. The NFB divider looks like:
- 1K from the speaker terminal to cathode of the driver stage;
- 100R from the cathode to 100uF capacitor;
- the other end of the 100uF capacitor is tied to GND.

My apologies, I didn't understand you. And the reason I didn't
understand was because the technique you are suggesting is completely
unknown in the range of traditional means of applying NFB in tube
amps.
Your method here is normal procedure in solid state amps which are all
dc coupled and without any large phase shifts at very low frequencies
down to DC.

Thus the feedback "beta" increases (!) at low frequencies (below 10Hz in
this case), giving -90deg phase lag in the loop below 10Hz. This is in
effect turning the amp into a anti-rumble filter.

Yes, I see you are against rumble. But in any tube amp with 2 CR
couplings and an OPT the gain below 10Hz quickly falls to zero at DC
and there is no need to roll off LF any more than it naturally is
rolled off. So if a signal at 1Hz enters the amp, the FB fed back is
extremely small because the output signal is so small.
So I cannot see your method would improve LF behaviour. I've read most
of the texts about NFB applications in tube amps and I don't recall a
single instance where your idea has been applied by any commercial
manufacturer nor have I read of anyone supporting it.


From the first glance it might sound crazy to increase the loop gain at VLF
where we want an overal reduction of yje loop gain, but consider this:
- OPT typically gives +90deg lead below 15...20Hz;
- interstage coupling (simple RC with no shelving, 220K and 0.033uF) is
calculated to give -3dB corner at say 15...20Hz and also gives +90 deg lead
below;
- but this "funny" NFB with a 100R and 100uF gives -90deg LAG below 10Hz!
And this lag maintains down until 1Hz!

Well, your NFB method does not increase OLG. OLG remains what it is
regardless of the NFB network. But th closed loop gain, CLG, could be
boosted with positive FB. 3 lots of reactive phase advances in the amp
will add to a rapid phase turnover per octave below 10Hz. So maximum
phase shift can be more than 180d then because the Lp inductance falls
to near zero near dc, you have only dc winding resistance in the OP
stage so phase shift lessens a bit towards DC. But exactly what anyone
measures depends on voltage applied and fiddling around with trying to
increase NFB as one gets close to DC makes no sense to me at all.

When you have measured and demonstrated your technique and analysised
it all with maybe 10 detailed pages on a website with photos of the
amps, then ppl might say you have something to offer. People here are
very difficult people. We cannot agree with anything anyone says
unless they offer the truth, the whole truth, and nothing but the
truth, so help them in the eyes of the God Of Triodes :-)


At LF one lag subtracts from two leads and in combination we have only
+90deg below 10Hz. Therefore, 0dB line at about 3Hz (typically) will be
crossed safely at only +90deg!

I'll believe works when I see it.

Killing many birds with one stone:
- perfect transient with no peaking;
- natural low cut off in the whole amp (sort of built-in anti-rumble
filtering);
- reduced error signal at VLF as more feedback is applied;
- no need to use two identical shelving circuits in push-pull amps -- just
one extra electrolytic.

A drawback - a despised electrolytic as a shaping component.

Perhaps there are other drawbacks you have not thought about.

I suggest you embark on a course of soldering in your laboratory to
prove your idea works. We all look forward to results.

Hi Patrick,

I like Alex's idea. As you point out, this feedback scheme was widely used in
solid state amps, although as you allude to, it didn't have anything to do with
feedback stability, my take was that the reason for its use was to increase the
amount of feedback at DC in order to minimize the DC offset at the output of the
amp. The thing I always liked about these solid state amps is the built in
rumble filter, which Alex also seems to like.

So this discussion gets me thinking, how could I apply Alex's idea to a tube amp
without using the despised electrolytic capacitor.

My solution, multiply the impedance of the feedback network by 220 X, making
Alex's 1K resistor 220k, his 100 Ohm resistor 22k, and his 100uF capacitor
0.47uF. Of course this network will no longer drive the cathode of V1 because
its impedance is too high, so I propose to add an additional triode operating as
a cathode follower driven by the high impedance feedback network and direct
coupled to the cathode of V1.

--
Regards,

John Byrns

Surf my web pages at, http://fmamradios.com/

[Patrick Turner]

At LF one lag subtracts from two leads and in combination we have only
+90deg below 10Hz. Therefore, 0dB line at about 3Hz (typically) will be
crossed safely at only +90deg!

Interesting idea but I'm having trouble wrapping my head around the
lead/lag analysis because, if I understand your circuit correctly,
that cap is at the summing junction so shouldn't it introduce poles
into both the NFB and the input signal?

I'm not the only one questioning Alex's FB network. You are having
trouble accepting Alex's claims.

Where's the evidence? its no ****ing good making claims without backup
evidence.

Patrick Turner.

[Patrick Turner]

On Jun 24, 5:20*am, flipper wrote:
On Wed, 22 Jun 2011 16:57:27 -0700 (PDT), Patrick Turner

wrote:
At LF one lag subtracts from two leads and in combination we have only
+90deg below 10Hz. Therefore, 0dB line at about 3Hz (typically) will be
crossed safely at only +90deg!

Interesting idea but I'm having trouble wrapping my head around the
lead/lag analysis because, if I understand your circuit correctly,
that cap is at the summing junction so shouldn't it introduce poles
into both the NFB and the input signal?

I'm not the only one questioning Alex's FB network.

You aren't 'questioning'. You've got your fingers stuck in your ears.

You are having
trouble accepting Alex's claims.

I asked only one specific question about the phase analysis. That it
prevents peaking, the POINT of the suggestion, is self evident.

Where's the evidence? its no ****ing good making claims without backup
evidence.

The evidence is in the analysis that you keep snipping and ignoring.

What sort of fish gave you the idea of calling yourself FLIPPER?
Doncha feel embarassed most days to be seen by others as a half
brained fish? Where is the evidence you can think about anything?

NOBODY HAS POSTED EVIDENCE THAT THE NFB IDEA AS DESCRIBED BY ALEX WILL
WORK TO TO ANYTHING WORTHWHILE.

I do not have my hands across my eyes, hands across my ears, or hands
across my mouth like the three unwise monkeys who would not see the
truth, hear the truth, or speak the truth.

SO WHERE IS THE TRUTH?

Truth needs to be proven to be true.

So Flipper, perhaps you can spend a few days to prepare a lecture on
the issue and explore all possibilities but I won't accept
simulations. You shall build a circuit, you shall apply a healthy self
critical attitude, and shall carefully measure all voltages and
currents and waveforms and distortions and phase shifts. You shall not
dither about on news groups throwing mud and accusing ppl of crapping
when they don't.

When that's all done we all might learn and beneft.

Patrick Turner

[John Byrns]

In article ,
flipper wrote:

On Sat, 18 Jun 2011 21:56:00 +1000, "Alex Pogossov"
wrote:


"Patrick Turner" wrote in message
...

Patrick:
But what of the phase shift of that C? Isn't it better to have R&C in
parallel inserted from FB take off at OPT sec to the feedback R? Say
you have 1k0 and 100r as the normal FB divider so that 1/11 of the OPT
signal is applied to V1 cathode. Say one adds 3k3 so you then have
3k3, 1k0, then 100r at k to 0V at V1. Then ß becomes 0.022, much less
than 0.09, and at very low F there is no phase shift, so with less NFB
its probably going to be stable. But there isn't enough FB at higher F
so you shunt the 3k3 with say 6.8uF. So at 100Hz the 6.8uF = 233 ohms
reactance and 3k3 is well shunted.

Peaking still has to be checked in output and in error signal.

Alex:
I suggested something quite opposite. The NFB divider looks like:
- 1K from the speaker terminal to cathode of the driver stage;
- 100R from the cathode to 100uF capacitor;
- the other end of the 100uF capacitor is tied to GND.

Thus the feedback "beta" increases (!) at low frequencies (below 10Hz in
this case), giving -90deg phase lag in the loop below 10Hz. This is in
effect turning the amp into a anti-rumble filter.

From the first glance it might sound crazy to increase the loop gain at VLF
where we want an overal reduction of yje loop gain, but consider this:
- OPT typically gives +90deg lead below 15...20Hz;
- interstage coupling (simple RC with no shelving, 220K and 0.033uF) is
calculated to give -3dB corner at say 15...20Hz and also gives +90 deg lead
below;
- but this "funny" NFB with a 100R and 100uF gives -90deg LAG below 10Hz!
And this lag maintains down until 1Hz!

At LF one lag subtracts from two leads and in combination we have only
+90deg below 10Hz. Therefore, 0dB line at about 3Hz (typically) will be
crossed safely at only +90deg!

Interesting idea but I'm having trouble wrapping my head around the
lead/lag analysis because, if I understand your circuit correctly,
that cap is at the summing junction so shouldn't it introduce poles
into both the NFB and the input signal?

I'm having some trouble wrapping my mind around this summing junction issue.
What is the problem if it does introduce a pole into the input signal? That
would not affect the feedback signal, and at worst would only enhance the
"Rumble Filter" effect, if it even does anything additional to the input signal
at all. I will have to write out the transfer function and see if there is any
additional affect on the input signal. If it does have some additional effect
on the input signal, which is considered undesirable, my cathode follower
modification described in an earlier post should eliminate the unwanted effect.

--
Regards,

John Byrns

Surf my web pages at, http://fmamradios.com/

[John Byrns]

In article ,
flipper wrote:

On Wed, 22 Jun 2011 19:24:10 -0500, John Byrns
wrote:

In article ,
flipper wrote:

On Sat, 18 Jun 2011 21:56:00 +1000, "Alex Pogossov"
wrote:

I suggested something quite opposite. The NFB divider looks like:
- 1K from the speaker terminal to cathode of the driver stage;
- 100R from the cathode to 100uF capacitor;
- the other end of the 100uF capacitor is tied to GND.

Thus the feedback "beta" increases (!) at low frequencies (below 10Hz in
this case), giving -90deg phase lag in the loop below 10Hz. This is in
effect turning the amp into a anti-rumble filter.

From the first glance it might sound crazy to increase the loop gain at
VLF
where we want an overal reduction of yje loop gain, but consider this:
- OPT typically gives +90deg lead below 15...20Hz;
- interstage coupling (simple RC with no shelving, 220K and 0.033uF) is
calculated to give -3dB corner at say 15...20Hz and also gives +90 deg
lead
below;
- but this "funny" NFB with a 100R and 100uF gives -90deg LAG below 10Hz!
And this lag maintains down until 1Hz!

At LF one lag subtracts from two leads and in combination we have only
+90deg below 10Hz. Therefore, 0dB line at about 3Hz (typically) will be
crossed safely at only +90deg!

Interesting idea but I'm having trouble wrapping my head around the
lead/lag analysis because, if I understand your circuit correctly,
that cap is at the summing junction so shouldn't it introduce poles
into both the NFB and the input signal?

I'm having some trouble wrapping my mind around this summing junction issue.

I don't know why. I wrote it at the exact spot in the message where it
matters, the lead/lag analysis, and said "lead/lag analysis."

If you don't know why, then please explain exactly where the summing junction is
located in this circuit? In inverting amplifiers the summing junction is
usually explicit and easy to see in the schematic diagram. Non inverting
amplifiers like this one are a different matter, and the summing junction is
buried somewhere inside the active devices, be they tubes, transistors, or ICs.
What I was trying to say is that I was trying to figure out where the "summing
junction" is actually located in this circuit.

What is the problem if it does introduce a pole into the input signal?

The whole point of that paragraph is adding the lead/lags and if
there's *another* lead/lag it changes the phase analysis.

Does, it? If there is a pole added to the input circuit, it doesn't affect the
"phase analysis" around the feedback loop, or the feedback stability issue,
although it would affect the closed loop gain of the entire amplifier circuit,
and hence what we see on the CRO after we solder the circuit together and apply
a test signal to the input.

That
would not affect the feedback signal, and at worst would only enhance the
"Rumble Filter" effect,

First Patrick was obsessed with the "Rumble Filter" and now you are. I
*specifically* said my question was about "the lead/lag analysis."

And now it seems to be you that is obsessed with the "Rumble Filter"! In any
case it's not a "Rumble Filter", it's a high pass filter with an infrasonic
cutoff, it wouldn't do a thing for rumble!

--
Regards,

John Byrns

Surf my web pages at, http://fmamradios.com/

[Patrick Turner]

snip

A drawback - a despised electrolytic as a shaping component.

Btw, there's another means of achieving a modest 'LF shelf', using a
concertina, and that's to boostrap the gain stage load off the
concertina cathode. Gain drops at LF as the boostrap rolls off.

This has been done in some Dynaco schematics which have the concertina
cathode driving the top of a RLdc via an electro cap, therefore
supplying positive FB to the input pentode by means of increasing the
RL the pentode anode "sees" so that pentode gain goes much higher
therefore making a given amount of GNFB much more effective. Basically
while peter robs paul and you light a fire under both, they tend to
become very active indeed.

If the input tube is a lowly triode who's gain is determined by its
lower µ the PFB effect is minimal, and whether there is a bootstrap or
not won't change VLF gain very much, so the shelving networks seen
throughout my website are perhaps the better way with triodes.

It has the added advantage of boosting stage gain and theoretically
lowering distortion from the (idealized) 'infinite impedance' of the
bootstrap

The bootstrap from concertina cathode could be taken from an
additional triode cathode follower which is driven directly off the
cathode so concertina R values remain unchanged, ie, anode RL =
cathode RL. The CF can then be used to drive a two resistor R divider
to V1 anode. One has to be careful that cut off distortion with the AC
coupling does not happen.

The concertina resistors need adjusting to re-balance the load, and
that causes an offset toward B+, but that can be an advantage as well
since direct coupled concertinas on modest B+ rails can end up with
rather low anode voltage on the gain stage and the offset bumps that
up.

Yes, indeed. I would say I prefer the LTP driver with cathode CCS with
say 6SN7/6CG7/12BH7/12AU7 or EL84. Then the LTP acts as a balanced
pair with very low THD. Whatever the input tube is, pentode or
triode, it has only to produce a low signal in SE mode so THD remains
lower than if you have an input tube needing to make slightly more Va
than is applied to each output tube grid.

There's plenny of headroom for the function of the shelving network.
The LTP does not need to be directly
driven from input gain tube anode, so the LTP can have much more
headroom than found in most samples of amps like Leak and Manley Labs
et all.

Patrick Turner.

[Patrick Turner]

On Jul 16, 3:07*pm, flipper wrote:
On Fri, 15 Jul 2011 02:03:07 -0700 (PDT), Patrick Turner





wrote:
snip

A drawback - a despised electrolytic as a shaping component.

Btw, there's another means of achieving a modest 'LF shelf', using a
concertina, and that's to boostrap the gain stage load off the
concertina cathode. Gain drops at LF as the boostrap rolls off.

This has been done in some Dynaco schematics which have the concertina
cathode driving the top of a RLdc via an electro cap, therefore
supplying positive FB to the input pentode by means of increasing the
RL the pentode anode "sees" so that pentode gain goes much higher
therefore making a given amount of GNFB much more effective. Basically
while peter robs paul and you light a fire under both, they tend to
become very active indeed.

Of course it's been done before.

If the input tube is a lowly triode who's gain is determined by its
lower µ the PFB effect is minimal, and whether there is a bootstrap or
not won't change VLF gain very much, so the shelving networks seen
throughout my website are perhaps the better way with triodes.

Depends, but 6 dB isn't a bad rough cut rule of thumb.

Sometimes 'just a little more' is all it takes and, at least with the
ones I've done, the capacitor values are usually, or can be, small
enough to use film and avoid electrolytics.

My "Stealth AX" amp does it sort of 'in reverse'. The gain triode Rl
is under the concertina splitter, for the 'infinite impedance'
bootstrap, and the signal is cap coupled to the concertina grid.

It has the added advantage of boosting stage gain and theoretically
lowering distortion from the (idealized) 'infinite impedance' of the
bootstrap

The bootstrap from concertina cathode could be taken from an
additional triode cathode follower which is driven directly off the
cathode so concertina R values remain unchanged, ie, anode RL =
cathode RL. The CF can then be used to drive a two resistor R divider
to V1 anode. One has to be careful that cut off distortion with the AC
coupling does not happen.

It "could be" but it seems an unnecessary waste of a tube.

The concertina resistors need adjusting to re-balance the load, and
that causes an offset toward B+, but that can be an advantage as well
since direct coupled concertinas on modest B+ rails can end up with
rather low anode voltage on the gain stage and the offset bumps that
up.

Yes, indeed. I would say I prefer the LTP driver with cathode CCS with
say 6SN7/6CG7/12BH7/12AU7 or EL84. Then the LTP acts as a balanced
pair with very low THD.

It isn't as low as a 100% NFB cathode follower, like the concertina,
and it's more tubes.

It depends. Have a look at
http://www.turneraudio.com.au/300w-1...tput-jan06.htm

Here I have an LTP which produces less than 0.5% THD at two phases of
85Vrms from each driver triode.

To do a similar thing with concertina, you'd have a triode making
170Vrms Va-k, and one might do that using an EL84 in triode, and the
drive voltage to the Concer grid would be 95Vrms. The previous stage
has to make a shirt&trouser load sized signal, maybe with 5% THD. Now
the concer stage OLG THD might be 10%, and with the CLG gain reduction
to about 2, the 10% is reduced to 1.2%, but basically you end up with
more 2H than I get with LTP which has natural cancelling of 2H. Now in
a typical Williamson, the balanced amp needs to make only 2 phases of
about say 32Vrms to power a couple of 6550 in triode. So the concer
makes only two phases of 2V and input tube makes only 2.2V at
clipping. Now with the Willy, the input concer stages usually have
such LOW signals at normal listening levels, say less than 0.2Vrms,
the THD is down at 0.05% and is reduced by the GNFB to utterly
negligible levels. The same goes for the use of the LTP in my amps.

So the discussion of "what is best" becomes of academic interest only.
In small amps with UL output stages with EL84, the SE input triode
with SE concer stage is all one needs and sounds/measured fine. If you
use an extra triode and make an LTP, you might find the amp can be
made more sensitive, and the THD will be almost indentical to the
concer stage but with less 2H present. Both ideas work fine where
signals are low, but in my high powered amps the concer drive stage is
not so hot. McIntosh amps have balanced driver stages.
Most makers use LTP or balanced with larger tubes.

*Whatever the input tube is, pentode or
triode, it has only to produce a low signal in SE mode so THD remains
lower than if you have an input tube needing to make slightly more Va
than is applied to each output tube grid.

Distortion is also increased by the extra tubes.

I happen to like the LTP too but that doesn't mean everything else is
crap and, like all design tradeoffs, "it depends" (on everything
else).

There's more than one way to do things.

A concertina is enough to drive a pair of 6BQ5s all by itself without
interposing another set of tubes and while that may not be your cup of
tea it makes for a perfectly fine little amp.

I agree entirely.

An LTP driven single ended has only half the gain of Williamson's
concertina double driven short tail and while I know you think
converting that to an LTP 'reduces distortion' it's at a place in the
amp likely to be of little consequence since the vast majority of
distortion is in the output stage. So far that's a "why not?" but the
LPT does cause overload interaction on positive grid drive, as does a
concertina *if* it's the thing doing the driving (but it's buffered in
the Williamson). It might not seem like such a big deal but I've seen
that cause HF instability on both the concertina and LPT and is the
reason some add a series grid resistor between the concertina and
output tubes when driving them directly.

Usually HF instability is due to the OLG phase shift characteristic
and once you have the correct gain shelving and zobels the HF becomes
entirely stable. And of course the grid stoppers are always a good
idea.

There's plenny of headroom for the function of the shelving network.
The LTP does not need to be directly
driven from input gain tube anode, so the LTP can have much more
headroom than found in most samples of amps like Leak and Manley Labs
et all.

What 'headroom'? The shelving network is zero loss in band.

Or do you mean for your shelf's 'peaking', which a bootstrap roll off
doesn't cause?

All I said is there was "another means," and I don't know that I'd do
it for 'just that', but it's an interesting aspect to consider if one
also wants the added gain.

Observation of the error signal at V1 anode output when feeding the
amp a level input signal between 1Hz to 100Hz will show that the LF
shelving network is OK. The reference 1kHz signal should produce half
maximum l Vo. Then if you drop the input F and plot the response at
all electrodes, you'll understand. And what will help matters is that
you have passive input CR input filter with pole at 8 Hz. You just
don't need to try to have the amp vainly try to make a big effort with
a 3Hz signal.

Patrick Turner.