What tests should one employ for
unconditional stability when building
or testing a valve/tube PP amp with NFB?

Iain

On Thu, 10 Mar 2005 12:33:44 +0200, "Iain M Churches"
wrote:

What tests should one employ for
unconditional stability when building
or testing a valve/tube PP amp with NFB?

Iain


Output impedance must remain positive at all frequencies. An impedance
below zero at any frequency (not just in the audio band) means that
the amplifier will oscillate into some load impedances - usually
capacitive.

d

Pearce Consulting
http://www.pearce.uk.com

Iain M Churches wrote:

What tests should one employ for
unconditional stability when building
or testing a valve/tube PP amp with NFB?

The amplifier must not oscillate at LF or HF or both without any load
connected.
The amp will not oscillate with any value of inductance connected
The amp will not oscillate with any value of capacitance between 10 uF
and 0.01 uf
connected.

The amp will not oscillate when a low level square wave signal signal
is used in any of the above situtations.

Testing with F below 20 Hz with a sine wave at 1/2 full output voltage
will not
produces bursts of HF oscillations when OPT has begun to saturtate at
say 5 Hz.

There is much else written about stability and *** margins for
stability***
in RDH4 and other books.

The open loop response for most tube amps with a resistive load looks
like a
section drawn through a mountain plateau.

As NFB is applied in incrementally increasing amounts, the flat part of
the plateau
is lowered in levels, but at the slopes up to the plateau the response
becomes peaked,
since the reactive elements in the amplifier cause phase shift which
prevent the
feedback from being negative.

Part of testing requires that you plot the response with some accuracy
as the level of NFB is increased, at least at first with a pure
resistive load.
If you do not know how to plot a response in five minutes, then learn!
Suppose that you are able to apply 35 dB of global NFB with phase
tweaking networks
in place before it is impossible to prevent LF or HF oscillations with
any increase in applied NFB.
Then suppose you settle for 20 dB of NFB with the phase tweaking
networks left in place.
You then have a 15 dB margin of stability, but only with a resistive
load.
15 dB is a good margin of stability for a tube amp.

But the amp trimmed like this may be unstable with a capacitor load, so
the amp must have
phase tweaking networks set up to reduce the open loop gain and phase
shift
with a capacitor load in place, usually a 0.22 uF for most power amps,
and without any R load.

The amplifier schematic at
http://www.turneraudio.com.au/htmlwe...0ulabinteg.htm
has LF and HF phase tweaking networks to improve the bass stability
and the treble stablity.
Usually, there is no need to worry about stability between 50 Hz and 5
kHz
and its below and above these F that oscillations will always occur.

In the above schemo, there is a network using 0.47 uF, 0.047uF, 1M, and
220k
which attenuates the LF open loop gain by about 12 dB below 20 Hz,
and reduces the ultimate phase shift below 10 Hz, and since applied NFB
is dependant on open loop gain, and stability gets worse with more gain,

the less gain at LF means there is greater stability.
So in effect, although we may have 20 dB of applied NFB at 1 kHz, at
10 Hz there may be only 14 dB, and at 5 Hz about 10 dB, and our amp
wants to then be stable.
The bass response will still be very much improved overt he open loop
response, even with the
open loop LF response "artificially" reduced in the way shown in the
schematic.

In the schemo there is also a 330pF and 3.3k in series from the grid of
the
LTP input grid to ground.
This acts to provide an RL **at HF** of ultimately of 3.3k to the input
tube which would otherwise be 75k plus the 220k
in parallel as shown.
This is because at low RF, the 330 pF becomesa short circuit, so the RL

= 3.3k, ( in parallel with 220k and 75k, which have little effect ).

Therefore the HF pole caused by the miller effect of the LTP is
transfered to a higher F than would
be the case if the RC network were not present, and in addition the HF
open loop
gain is reduced in a frequency range to reduce the applied feedback as F
is increased above 20 kHz,
where there is little point in having the same amount of applied FB as
one wants at 1 kHz.


Other phase correction network tricks are employed to increase the
margins of stability
such as placing just the right size C across the global FB resistor, see
the
0.001 uF across the 1 k FB resistor.
Also, the FB network is a low impedance one, so the NFB is delivered via
the 1k and 100 ohm divider,
and the bias R of the input tube is bypoassed by a large value electro,
see the 2.2k and 1,000 uF cap
At really low F, the 1,000 uF becomes a high impedance, and the gain of
V1 ia slightly
reduced, but phase shift is reduced.

In another schemo at
http://www.turneraudio.com.au/htmlwe...00ulabmono.htm
there is some RC networks in series across each 1/2 primary of the OPT
to give the output tubes a resistive load at HF as the leakage
inductance gradually
decouples the tubes from the load at the output, but while it does,
there is a serious phas shift and tendency to HF instability, and the
series RC "zobel" networks add to the margin of stability.

When an amp is stabilised to the list above,
there will still be some overshoot on square waves with capacitor loads.

The sine wave response is best measured with a cap load when output
voltage = 1/10
of maximum for clipping with an R load at 1 kHz.
Therefore the amp should be able to produce a true non saturated
response for the cap load, since the cap load becomes a low impedance at
HF,
and no amp is able to put an infinite amount of current into a cap at HF

in order to keep the output voltage equal to what it is at 1 kHz.

If there is up to 6 dB of peaking in the sine wave response at say 40
kHz with a 0.22 uF,
and other values of cap won't produce a higher peak, and you have 20 dB
of applied NFB,
then you have a nicely stable tube amp.
Usually the addition of an R across the C load, or a small R in series
with the C should improve the
response.

A test for ability to drive Quad ESL57 will have a
16 ohm R as a load with 2 uF and 1.6 ohms in series, also as a load.
many amps will show a peaked response at between 16 and 40 kHz, maybe 6
dB,
and 3 dB at 20 kHz, and this is a poor outcome, usually because tyhe OPT
is poor,
or the builder hasn't done his stabilising networks properly.

Despite phase tweaking networks, it should be still possible to get a
resistive load response with no peaking outside tha 20 Hz to 20 kHz
band,
and with -3 dB points at least from 5 Hz to 70 kHz at 1/10 full output
voltage,
and on the really good amps, between 16 Hz and 65 kHz at the full power
clipping level at 1 kHz, even for where RL = 1/2 the rated RL for the
amp.

The idea of trimming the response is called critical damping, critical
because the choice of
RC correction components is a narrow one because we aim to have the
maximum
bandwidth with minimum phase shift with FB applied and still have
stability into any kind of reactive load.

The amplifier should be considered as a bandpass filter, and one around
which we apply
NFB, a loop which includes the phase delays in the transfer of signals.

There is no theoretical way we can select damping components or exactly
calculate the values, since the stray OPT
C and leakage L are quantities not able to be quantified accurately
enough
to allow optimal damping, and its only after we have done a few amps
that we get good at the cut and try
approach to critical damping.

I use the empirical notion that for zobels across the halves of the OPT,

each R = 1/2 RLa-a, and C has Z = 1/2 RLa-a at about 100 kHz.

For the RC across the output of V1, I use a variable radio tuning gang
and a pot to find the least ring
with a 5 kHz square wave wehn using 0.22 uF across the output.
The C across the FB resistance is adjusted until HF oscillations at RF
start, then the value halved.
Its somewhat guesswork.

But at the end of the procedure, no tube of the the amp should saturate,
ie, be forced into grid current
or cut off with an R load at full 1 kHz power between 20 Hz and 65 kHz.

A pure cap load of 2 uF should produce a peak in the sine wave response
at an F above
25 kHz, not below, and the peak no more than +3 dB, so that at 20 kHz,
such a C
should cause a lift in response of no more than 1.5 dB.

Since such C loads are not encountered in real world speakers, and there
is usually
some series and parallel R also present, and so
the response into ESL should be substantially flat if we have got it all
right,
and in fact the tube amp will often provide a better response than many
SS amps.

The secret to a good response is an OPT with a wide bandwidth.

This means high primary L and low saturation F, low stray capacitance
and low leakage inductance.
( and then such transfromers usually contribute a very small amount of
HD to the
THD of the amp ).

This means all out phase correction tricks only affect frequencies well
outside the
audio band, where there is little energy in the recordings.

Many amplifier makers pay only lipservice to the above concerns because
its cheaper
to make a crummy amp, and its easy to get away with it.

There are other statements I have made on NFB and stability in the RAT
archives.


Patrick Turner.











Iain

"Patrick Turner" wrote in message
...


Iain M Churches wrote:

What tests should one employ for
unconditional stability when building
or testing a valve/tube PP amp with NFB?

The amplifier must not oscillate at LF or HF or both without any load
connected.
The amp will not oscillate with any value of inductance connected
The amp will not oscillate with any value of capacitance between 10 uF
and 0.01 uf
connected.

The amp will not oscillate when a low level square wave signal signal
is used in any of the above situtations.

Testing with F below 20 Hz with a sine wave at 1/2 full output voltage
will not
produces bursts of HF oscillations when OPT has begun to saturtate at
say 5 Hz.

There is much else written about stability and *** margins for
stability***
in RDH4 and other books.

The open loop response for most tube amps with a resistive load looks
like a
section drawn through a mountain plateau.

As NFB is applied in incrementally increasing amounts, the flat part of
the plateau
is lowered in levels, but at the slopes up to the plateau the response
becomes peaked,
since the reactive elements in the amplifier cause phase shift which
prevent the
feedback from being negative.

Part of testing requires that you plot the response with some accuracy
as the level of NFB is increased, at least at first with a pure
resistive load.
If you do not know how to plot a response in five minutes, then learn!
Suppose that you are able to apply 35 dB of global NFB with phase
tweaking networks
in place before it is impossible to prevent LF or HF oscillations with
any increase in applied NFB.
Then suppose you settle for 20 dB of NFB with the phase tweaking
networks left in place.
You then have a 15 dB margin of stability, but only with a resistive
load.
15 dB is a good margin of stability for a tube amp.

But the amp trimmed like this may be unstable with a capacitor load, so
the amp must have
phase tweaking networks set up to reduce the open loop gain and phase
shift
with a capacitor load in place, usually a 0.22 uF for most power amps,
and without any R load.

The amplifier schematic at
http://www.turneraudio.com.au/htmlwe...0ulabinteg.htm
has LF and HF phase tweaking networks to improve the bass stability
and the treble stablity.
Usually, there is no need to worry about stability between 50 Hz and 5
kHz
and its below and above these F that oscillations will always occur.

In the above schemo, there is a network using 0.47 uF, 0.047uF, 1M, and
220k
which attenuates the LF open loop gain by about 12 dB below 20 Hz,
and reduces the ultimate phase shift below 10 Hz, and since applied NFB
is dependant on open loop gain, and stability gets worse with more gain,

the less gain at LF means there is greater stability.
So in effect, although we may have 20 dB of applied NFB at 1 kHz, at
10 Hz there may be only 14 dB, and at 5 Hz about 10 dB, and our amp
wants to then be stable.
The bass response will still be very much improved overt he open loop
response, even with the
open loop LF response "artificially" reduced in the way shown in the
schematic.

In the schemo there is also a 330pF and 3.3k in series from the grid of
the
LTP input grid to ground.
This acts to provide an RL **at HF** of ultimately of 3.3k to the input
tube which would otherwise be 75k plus the 220k
in parallel as shown.
This is because at low RF, the 330 pF becomesa short circuit, so the RL

= 3.3k, ( in parallel with 220k and 75k, which have little effect ).

Therefore the HF pole caused by the miller effect of the LTP is
transfered to a higher F than would
be the case if the RC network were not present, and in addition the HF
open loop
gain is reduced in a frequency range to reduce the applied feedback as F
is increased above 20 kHz,
where there is little point in having the same amount of applied FB as
one wants at 1 kHz.


Other phase correction network tricks are employed to increase the
margins of stability
such as placing just the right size C across the global FB resistor, see
the
0.001 uF across the 1 k FB resistor.
Also, the FB network is a low impedance one, so the NFB is delivered via
the 1k and 100 ohm divider,
and the bias R of the input tube is bypoassed by a large value electro,
see the 2.2k and 1,000 uF cap
At really low F, the 1,000 uF becomes a high impedance, and the gain of
V1 ia slightly
reduced, but phase shift is reduced.

In another schemo at
http://www.turneraudio.com.au/htmlwe...00ulabmono.htm
there is some RC networks in series across each 1/2 primary of the OPT
to give the output tubes a resistive load at HF as the leakage
inductance gradually
decouples the tubes from the load at the output, but while it does,
there is a serious phas shift and tendency to HF instability, and the
series RC "zobel" networks add to the margin of stability.

When an amp is stabilised to the list above,
there will still be some overshoot on square waves with capacitor loads.

The sine wave response is best measured with a cap load when output
voltage = 1/10
of maximum for clipping with an R load at 1 kHz.
Therefore the amp should be able to produce a true non saturated
response for the cap load, since the cap load becomes a low impedance at
HF,
and no amp is able to put an infinite amount of current into a cap at HF

in order to keep the output voltage equal to what it is at 1 kHz.

If there is up to 6 dB of peaking in the sine wave response at say 40
kHz with a 0.22 uF,
and other values of cap won't produce a higher peak, and you have 20 dB
of applied NFB,
then you have a nicely stable tube amp.
Usually the addition of an R across the C load, or a small R in series
with the C should improve the
response.

A test for ability to drive Quad ESL57 will have a
16 ohm R as a load with 2 uF and 1.6 ohms in series, also as a load.
many amps will show a peaked response at between 16 and 40 kHz, maybe 6
dB,
and 3 dB at 20 kHz, and this is a poor outcome, usually because tyhe OPT
is poor,
or the builder hasn't done his stabilising networks properly.

Despite phase tweaking networks, it should be still possible to get a
resistive load response with no peaking outside tha 20 Hz to 20 kHz
band,
and with -3 dB points at least from 5 Hz to 70 kHz at 1/10 full output
voltage,
and on the really good amps, between 16 Hz and 65 kHz at the full power
clipping level at 1 kHz, even for where RL = 1/2 the rated RL for the
amp.

The idea of trimming the response is called critical damping, critical
because the choice of
RC correction components is a narrow one because we aim to have the
maximum
bandwidth with minimum phase shift with FB applied and still have
stability into any kind of reactive load.

The amplifier should be considered as a bandpass filter, and one around
which we apply
NFB, a loop which includes the phase delays in the transfer of signals.

There is no theoretical way we can select damping components or exactly
calculate the values, since the stray OPT
C and leakage L are quantities not able to be quantified accurately
enough
to allow optimal damping, and its only after we have done a few amps
that we get good at the cut and try
approach to critical damping.

I use the empirical notion that for zobels across the halves of the OPT,

each R = 1/2 RLa-a, and C has Z = 1/2 RLa-a at about 100 kHz.

For the RC across the output of V1, I use a variable radio tuning gang
and a pot to find the least ring
with a 5 kHz square wave wehn using 0.22 uF across the output.
The C across the FB resistance is adjusted until HF oscillations at RF
start, then the value halved.
Its somewhat guesswork.

But at the end of the procedure, no tube of the the amp should saturate,
ie, be forced into grid current
or cut off with an R load at full 1 kHz power between 20 Hz and 65 kHz.

A pure cap load of 2 uF should produce a peak in the sine wave response
at an F above
25 kHz, not below, and the peak no more than +3 dB, so that at 20 kHz,
such a C
should cause a lift in response of no more than 1.5 dB.

Since such C loads are not encountered in real world speakers, and there
is usually
some series and parallel R also present, and so
the response into ESL should be substantially flat if we have got it all
right,
and in fact the tube amp will often provide a better response than many
SS amps.

The secret to a good response is an OPT with a wide bandwidth.

This means high primary L and low saturation F, low stray capacitance
and low leakage inductance.
( and then such transfromers usually contribute a very small amount of
HD to the
THD of the amp ).

This means all out phase correction tricks only affect frequencies well
outside the
audio band, where there is little energy in the recordings.

Many amplifier makers pay only lipservice to the above concerns because
its cheaper
to make a crummy amp, and its easy to get away with it.

There are other statements I have made on NFB and stability in the RAT
archives.


Patrick Turner.

Thanks Patrick for taking the trouble to reply to my question so fully.
I shall print your post for careful study.

Iain

The amplifier must not oscillate at LF or HF or both without any load
connected.
The amp will not oscillate with any value of inductance connected
The amp will not oscillate with any value of capacitance between 10 uF
and 0.01 uf connected.
The amp will not oscillate when a low level square wave signal signal
is used in any of the above situtations.


Not in a box.
Not with a fox.
Not in a house.
Not with a mouse.
It will not oscillate here or there.
It will not oscillate anywhere.
I would not eat green eggs and ham.
I do not like them, Sam-I-am.


Sorry, couldn't resist ;-)

Gilbert Bates wrote:


The amplifier must not oscillate at LF or HF or both without any load
connected.
The amp will not oscillate with any value of inductance connected
The amp will not oscillate with any value of capacitance between 10 uF
and 0.01 uf connected.
The amp will not oscillate when a low level square wave signal signal
is used in any of the above situtations.


Not in a box.
Not with a fox.
Not in a house.
Not with a mouse.
It will not oscillate here or there.
It will not oscillate anywhere.
I would not eat green eggs and ham.
I do not like them, Sam-I-am.

Sorry, couldn't resist ;-)

Its good to see that in some situtations, some ppl have a sense of
humerator that works :-)

Patrick Turner.