NFB101 Part 3

NFB101 Part 3

I am working on the next part of NFB101. It is turning into a a huge
mass of maths which I think will put off a lot of people. So I decided
to split it into two parts. The first covers the basics and then when
the maths gets heavy it just quotes the result and refers instead to an
appendix where the heavy maths resides. That way the main section can
concentrate on the meaning of the results rather than their derivation
the the details are available to anyone who is interested in them.

Before I commit to this split the groups views would be welcome.

Cheers

Ian

Re: NFB101 Part 3



Ian Thompson-Bell wrote:
>
> I am working on the next part of NFB101. It is turning into a a huge
> mass of maths which I think will put off a lot of people. So I decided
> to split it into two parts. The first covers the basics and then when
> the maths gets heavy it just quotes the result and refers instead to an
> appendix where the heavy maths resides. That way the main section can
> concentrate on the meaning of the results rather than their derivation
> the the details are available to anyone who is interested in them.
>
> Before I commit to this split the groups views would be welcome.
>
> Cheers
>
> Ian

The maths doesn't need to get very complex unless you include all the
derivations and include for reactive elements.

So there isn't too much wrong with stating concepts
and then the equation that follows merely underlines the concept.

But after 30 editing attempts, what you end up with should be easily
read by anyone with some basic knowledge.

RDH4 is a hard act to follow on NFB for tubes I think.

But of course there are many ways NFB is applied in RDH4 which are never
done now.
And don't foget, some things are only explained with maths to be of any
use.

Patrick Turner.

Re: NFB101 Part 3

In article <47FB7988.2E607446@turneraudio . com .au>,
Patrick Turner <info@turneraudio . com .au> wrote:

> Ian Thompson-Bell wrote:
> >
> > I am working on the next part of NFB101. It is turning into a a huge
> > mass of maths which I think will put off a lot of people. So I decided
> > to split it into two parts. The first covers the basics and then when
> > the maths gets heavy it just quotes the result and refers instead to an
> > appendix where the heavy maths resides. That way the main section can
> > concentrate on the meaning of the results rather than their derivation
> > the the details are available to anyone who is interested in them.
> >
> > Before I commit to this split the groups views would be welcome.
> >
> > Cheers
> >
> > Ian
>
> The maths doesn't need to get very complex unless you include all the
> derivations and include for reactive elements.

Isn't including the reactive elements, at least a simplified
representation of them, the whole point of this exercise? If you ignore
the reactive elements, then there is no stability issue and one is led
to the erroneous conclusion that there is no problem applying large
amounts of feedback.


Regards,

John Byrns

--
Surf my web pages at, * fmamradios . com /

Re: NFB101 Part 3



"John Byrns" <byrnsj@sbcglobal . net > wrote in message
news:byrnsj-8D067C.09271808042008@newsclstr02.news.prodigy . com ...
> In article <47FB7988.2E607446@turneraudio . com .au>,
> Patrick Turner <info@turneraudio . com .au> wrote:
>
>> Ian Thompson-Bell wrote:
>> >
>> > I am working on the next part of NFB101. It is turning into a a huge
>> > mass of maths which I think will put off a lot of people. So I decided
>> > to split it into two parts. The first covers the basics and then when
>> > the maths gets heavy it just quotes the result and refers instead to an
>> > appendix where the heavy maths resides. That way the main section can
>> > concentrate on the meaning of the results rather than their derivation
>> > the the details are available to anyone who is interested in them.
>> >
>> > Before I commit to this split the groups views would be welcome.
>> >
>> > Cheers
>> >
>> > Ian
>>
>> The maths doesn't need to get very complex unless you include all the
>> derivations and include for reactive elements.
>
> Isn't including the reactive elements, at least a simplified
> representation of them, the whole point of this exercise? If you ignore
> the reactive elements, then there is no stability issue and one is led
> to the erroneous conclusion that there is no problem applying large
> amounts of feedback.
>

Agreed. It is the *limitation* in how much feedback can be applied
that is the problem for a great many of us. I have taken the first steps
in stabilising a couple of amps by just widening the stability margin.I
do this by empirical methods. I replace the Rfb with a pot (usually 10k)
and back off until the amp is on the edge of oscillation. Then I turn the
pot to increase the value (and increase the gain of the amp) by 10dB.

I for one am very grateful to Ian for the excellent work so far. I
have kept printed copies. Let's hope that this thread will lead to
a fruitful discussion on amp stability which for many homebrew
builders is a nightmare.

Regards to all
Iain

Re: NFB101 Part 3

Iain Churches wrote:
> "John Byrns" <byrnsj@sbcglobal . net > wrote in message
> news:byrnsj-8D067C.09271808042008@newsclstr02.news.prodigy . com ...
>> In article <47FB7988.2E607446@turneraudio . com .au>,
>> Patrick Turner <info@turneraudio . com .au> wrote:
>>
>>> Ian Thompson-Bell wrote:
>>>> I am working on the next part of NFB101. It is turning into a a huge
>>>> mass of maths which I think will put off a lot of people. So I decided
>>>> to split it into two parts. The first covers the basics and then when
>>>> the maths gets heavy it just quotes the result and refers instead to an
>>>> appendix where the heavy maths resides. That way the main section can
>>>> concentrate on the meaning of the results rather than their derivation
>>>> the the details are available to anyone who is interested in them.
>>>>
>>>> Before I commit to this split the groups views would be welcome.
>>>>
>>>> Cheers
>>>>
>>>> Ian
>>> The maths doesn't need to get very complex unless you include all the
>>> derivations and include for reactive elements.
>> Isn't including the reactive elements, at least a simplified
>> representation of them, the whole point of this exercise? If you ignore
>> the reactive elements, then there is no stability issue and one is led
>> to the erroneous conclusion that there is no problem applying large
>> amounts of feedback.
>>
>
> Agreed. It is the *limitation* in how much feedback can be applied
> that is the problem for a great many of us. I have taken the first steps
> in stabilising a couple of amps by just widening the stability margin.I
> do this by empirical methods. I replace the Rfb with a pot (usually 10k)
> and back off until the amp is on the edge of oscillation. Then I turn the
> pot to increase the value (and increase the gain of the amp) by 10dB.
>
> I for one am very grateful to Ian for the excellent work so far. I
> have kept printed copies. Let's hope that this thread will lead to
> a fruitful discussion on amp stability which for many homebrew
> builders is a nightmare.
>
> Regards to all
> Iain
>
>

Interesting comments from all. To put it into context I personally
dislike it when books do not give a full derivation of their results - I
somehow believe them less - so I wanted to include full derivations of
everything. However, doing this means even the basic stuff gets really
packed with formulae which I thought would put people off. SO to try to
cover both needs I thought of keeping just the *essential*maths in the
main part and the details in the appendix for anyone who wants to wade
through it.

Stability is of course the key, but there is a lot of basic stuff to
wade through before you have the tools/understanding to tackle it. I
wanted to make that initial process as easy as possible so that people
can get to the nitty gritty stability stuff at a pace that suits their
existing knowledge and interest.

So on balance I think I'll stick with the basics + appendix approach and
see how it goes down.

Thanks for all you inputs.

Cheers

ian

Re: NFB101 Part 3



John Byrns wrote:
>
> In article <47FB7988.2E607446@turneraudio . com .au>,
> Patrick Turner <info@turneraudio . com .au> wrote:
>
> > Ian Thompson-Bell wrote:
> > >
> > > I am working on the next part of NFB101. It is turning into a a huge
> > > mass of maths which I think will put off a lot of people. So I decided
> > > to split it into two parts. The first covers the basics and then when
> > > the maths gets heavy it just quotes the result and refers instead to an
> > > appendix where the heavy maths resides. That way the main section can
> > > concentrate on the meaning of the results rather than their derivation
> > > the the details are available to anyone who is interested in them.
> > >
> > > Before I commit to this split the groups views would be welcome.
> > >
> > > Cheers
> > >
> > > Ian
> >
> > The maths doesn't need to get very complex unless you include all the
> > derivations and include for reactive elements.
>
> Isn't including the reactive elements, at least a simplified
> representation of them, the whole point of this exercise? If you ignore
> the reactive elements, then there is no stability issue and one is led
> to the erroneous conclusion that there is no problem applying large
> amounts of feedback.

Well John, the world ain't purfect.

The trouble with tube amps is that there is a shirt & trouser load of
many reactive elements in the OLG character.

One could say gain, A, = 100. But a full equation would be
say A = 100 x ( do dah do dah blah blah tiddly wuptit )

The bracketed quantities would sum to 1.0 where the load was pure R and
at one F
where phase shift was zero. But they would sum to anything but one
if we enter in all the reactives, Lp, LL, Cshunt, and consecutive Miller
C and stray C and so on.

Trouble is Lp varies with ac voltage and frequency due to iron
permeability,
and at HF the effects of C and LL become exceedingly complex, and a full
equivalent
circuit for an OPT with 4P x 5S sections may have 30 sub quantity L and
C amonts to deal with

So finding a GUT, a Grand United Theory for tube amp stabilization
is a waste of time for the practical man who isn't paid like a
university
professor. The Practical Man and has 4 kids and a nagging wife to cop
with.

He only has 1/2 an unpaid sunday to stabilize his amp.

So, rather than measure very much, and rather than go utterly insane
trying to spend
20 years on developing maths nobody else can possibly understand, the PM
does it my way,
and uses a 0.22uF cap to see if oscillates at HF, then applies a Zobel
at V1 anode load
and he uses a radio tuning cap and pot to set the best values.
The PM applies Zobels across the two halves of the OPT to load the
OP tubes above 20kHz because as F rises the LL becomes a series high Z
element,
and resonates with whatever C it can resonate with.
Of course before during and after the amp stabilizing tweak,
the PM has played around for 20 x 1/2 sundays with a CRO and exercize
book and pencil to draw responses in
about 20 seconds and after reading books on basic LCR behaviour, has
learnt what to expect,
even though he hasn't a clue what "j" factoids or the square root of -1
has to do
with a bunch of L and C and R parts hooked up on the bench.

If he really does know all the L, C and R values are in a network, and
there are 20 of them,
then figuring it all by hand on his own will take 20 x 20 x 1/2 sundays
long.
And that's be a tragic waste of 1/2 a year.
Better he get a simulation program for the PC, or go to a site with an
onsite
computator, just enter values, and out comes the response and phase
shift and input/output impedance result.
That'd be real cool, but a site that allows that does not exist afaik.

So the penniless PM who sure can't pay for a basic $40 LCR filter
program
is left to his own cunning devices. He exists in an state
of being helpless before a cruel and intolerant and uncaring world that
has gorn quite
crazy over transistors, and is hell bent on exterminating analog
behaviours in any electronics
wherever it raises its inefficient, ugly and expensive face.

But in 1955, before PCs came along, and if your parents could afford the
uni fees,
and if you had an IQ of 150, then after a bit of a tutor at uni the
LCR predictions were a doddle, and workable with a slide rule and a good
memory,
and if you didn't drink.

But the PM near you has IQ = 130 only, and had poor parents, so no uni,
and he'd like a drop sometimes.

90% of ppl designing prototypes of amps in the past were mere PMs.
They were starved of time and interferred with by accountants.

Their stabilization attempts were sometimes utterly bleedin awful,
showing that
they were UPMs, Unskilled Practical Manglers, a different beast to the
few PM
who could stabilize any fucking amp at all, even using one hand only,
and on a pay rate much lower than the uni educated fella.

My experience with the Jolida 502 amplifier I have worked on for a
couple of days
puts me in the class of PM who can now do it one handedly, and possibly
could do it asleep,
but for the risk of being woken abruptly by stray contact with 500
volts.
Yes there were a couple of very simple 3 prong linear equations to work
out,
but it only requires basic knowledge of the concept and school boy math.

I tried to make it as simple as possible somewhere on my website.

But most of the content leads to insanity if studied too deeply.

Certainly it leads to povety and divorce if you don't watch your step.

Patrick Turner.



> Regards,
>
> John Byrns
>
> --
> Surf my web pages at, * fmamradios . com /

Re: NFB101 Part 3



"Patrick Turner" <info@turneraudio . com .au> wrote in message
news:47FCB5BF.82703CCD@turneraudio . com .au...

> So, rather than measure very much, and rather than go utterly insane
> trying to spend
> 20 years on developing maths nobody else can possibly understand, the PM
> does it my way,
> and uses a 0.22uF cap to see if oscillates at HF, then applies a Zobel
> at V1 anode load
> and he uses a radio tuning cap and pot to set the best values.

But what does he do if he uses (as I do) a mu-follower
for the input stage. There is no anode load.

There are however two outputs. Can one shape the (unused)
HiZ output to modify the response of the LowZ output?


Regards to all
Iain

Re: NFB101 Part 3



Iain Churches wrote:
>
> "Patrick Turner" <info@turneraudio . com .au> wrote in message
> news:47FCB5BF.82703CCD@turneraudio . com .au...
>
> > So, rather than measure very much, and rather than go utterly insane
> > trying to spend
> > 20 years on developing maths nobody else can possibly understand, the PM
> > does it my way,
> > and uses a 0.22uF cap to see if oscillates at HF, then applies a Zobel
> > at V1 anode load
> > and he uses a radio tuning cap and pot to set the best values.
>
> But what does he do if he uses (as I do) a mu-follower
> for the input stage. There is no anode load.

The Jolida 502 amp has a SRPP input stage using a 12AX7 with 470 ohms Rk
for the
bottom triode and 1k for the top triode.

It has 16dB global NFB applied via a 27k R from the 8 ohm output
terminal.

The SRPP is actually like a two element stage, with 2 triodes in series.
The circuit resistance looking at the anode circuit point of the bottom
triode
is made up of the effective RL above the anode in parallel with the
effective Ra of the bottom triode
due to a virtually un-bypassed Rk of 1k.
Respectively, there is about 150k above the anode, and about 130k
effective Ra below the anode connection.
Total is about 70k, and I used a 3k9 plus 270pF which
created a first pole at 159,000 / ( 0.00027 x 70,000 ) which is approx
8.4kHz.
The 3k9 loads the bottom triode as the reactance of 270pF reduces so
that its gain as F rises
is much reduced.
By 200kHz, the Z 270pF = 294 ohms, so the Zobel has virtually no
reactive quality and it
has Z = 3k9, but it is resistive, and without phase shift. Gain at
200kHz
will therefore be about 3.3, but without the ultimate
90 degree phase lag caused by the C.

It is obvious to those who understand the most basic analysis of very
simple easy to figure out
tube circuits that the place to put a Zobel in SRPP is from the bottom
anode to 0V.
If it was placed at the top tube cathode, the top tube's local follower
NFB action and current regulation
would drive the top tube into saturation and distortion as F rises.

The purpose of the zobel is to load the tube at HF with a resistance
load
while preventing the phase shift of any C, lest the added phase shift
make the amp even more unstable despite the use of sone pure C across
the
gain tube output.
The gain tube does not have to make a huge voltage output,
and even when a partially reactive load like an ESL speaker is used,
the V1 gain tube still should not overload at HF because of the
compensation generated by global NFB action.

With 1mA of anode current, and at 200kHz, an unclipped output
of 2.7vrms is possible into the 3k9 load.
>
> There are however two outputs. Can one shape the (unused)
> HiZ output to modify the response of the LowZ output?

Sure can, see above.

When I plotted the OLG response in my workbook with pencil while
watching voltage levels on the CRO, sure enough the
-3dB point is at about 9k.
This may seem to be awful, especially when deliberately created,
but its necessary to avoid other phase shifts further up the band.
It means that the 12dB is at about 35kHz, so because the OLG is lessened
by 12dB at 35 kHz,
there is only about 5 dB NFB applied, allowing for what phase shift does
happen anyway.
There is an F where OLG = 1.0, and even with a 0.22uF cap across the
output as the sole load,
phase shift is less than 180 degrees, so it does'nt oscillate.
The 20pF compensation cap across the RFB does give some phase advance of
the signal being fed back.

This helps matters. This cap shouldn't be too large lest the FB become
fully advanced by 90 degrees
and thus cause a difference with input signal of 180 degrees where gain
is above 1.0.

There is still some overshoot and "ring" with a 5 kHz square wave with
the Zobel networks in place.

The Zobel networks stop oscillation, and IF they were they arranged to
remove all ringing,
the response of the amp would be be lousy, with a far too early HF pole.

You see, a tube amp has many poles at HF and a few at LF, and they sum
to form
an equivalent of a band pass filter with multi L and C and R elements
so that the rate of gain attenuation beyond where OLG = 1.0 becomes
greater than 6dB/octave
and further out becomes very much greater, perhaps 18dB/octave.

Such rapid phase "turnover" at extremes of F cause oscillations in an
amp if we are not careful.

All such multi order LCR filters where the roll off exceeds 6dB/octave
will generate ringing
on square waves at the resonant F of the series/parallel LC components
involved.

Where such ringing is seen on the CRO, and where the ring exceeds twice
the amplitude
of the horizontal "settled" part of the squarewave, you will get a
peaked sine wave response.
Its normal to see this in tube amps when loaded with C loads.
It is tested at low level, tyically with 1Vrms output when the amp is in
pure class A and OLG is highest,
and NFB IS connected.
2uF will give a peaked response of 6dB at 15kHz in a poor amp, and 60kHz
in a very good amp.
Usually this means the poorer amp has 8 times the leakage inductance of
the better amp.
As C is reduced in value with a poorer amp, the F peak will move higher,
and be so high at some F that
oscillation begins.

A good amp will withstand 2uF being connected and the reponse at 20khz
will rise
by no more than 1dB and the sine wave peaking not more than 3dB at some
F above 20kHz.
All other values of C will not cause higher peaks than 3dB.

Music isn't a square wave, so as long as the above criteria are met, the
amp will be stable
and sound well.


Patrick Turner.
>
> Regards to all
> Iain

Re: NFB101 Part 3



"Patrick Turner" <info@turneraudio . com .au> wrote in message
news:47FDDE34.F381F389@turneraudio . com .au...
>
>
> Iain Churches wrote:
>>
>> "Patrick Turner" <info@turneraudio . com .au> wrote in message
>> news:47FCB5BF.82703CCD@turneraudio . com .au...
>>
>> > So, rather than measure very much, and rather than go utterly insane
>> > trying to spend
>> > 20 years on developing maths nobody else can possibly understand, the
>> > PM
>> > does it my way,
>> > and uses a 0.22uF cap to see if oscillates at HF, then applies a Zobel
>> > at V1 anode load
>> > and he uses a radio tuning cap and pot to set the best values.
>>
>> But what does he do if he uses (as I do) a mu-follower
>> for the input stage. There is no anode load.
>
> The Jolida 502 amp has a SRPP input stage using a 12AX7 with 470 ohms Rk
> for the
> bottom triode and 1k for the top triode.
>
> It has 16dB global NFB applied via a 27k R from the 8 ohm output
> terminal.
>
> The SRPP is actually like a two element stage, with 2 triodes in series.
> The circuit resistance looking at the anode circuit point of the bottom
> triode
> is made up of the effective RL above the anode in parallel with the
> effective Ra of the bottom triode
> due to a virtually un-bypassed Rk of 1k.
> Respectively, there is about 150k above the anode, and about 130k
> effective Ra below the anode connection.
> Total is about 70k, and I used a 3k9 plus 270pF which
> created a first pole at 159,000 / ( 0.00027 x 70,000 ) which is approx
> 8.4kHz.
> The 3k9 loads the bottom triode as the reactance of 270pF reduces so
> that its gain as F rises
> is much reduced.
> By 200kHz, the Z 270pF = 294 ohms, so the Zobel has virtually no
> reactive quality and it
> has Z = 3k9, but it is resistive, and without phase shift. Gain at
> 200kHz
> will therefore be about 3.3, but without the ultimate
> 90 degree phase lag caused by the C.
>
> It is obvious to those who understand the most basic analysis of very
> simple easy to figure out
> tube circuits that the place to put a Zobel in SRPP is from the bottom
> anode to 0V.
> If it was placed at the top tube cathode, the top tube's local follower
> NFB action and current regulation
> would drive the top tube into saturation and distortion as F rises.
>
> The purpose of the zobel is to load the tube at HF with a resistance
> load
> while preventing the phase shift of any C, lest the added phase shift
> make the amp even more unstable despite the use of sone pure C across
> the
> gain tube output.
> The gain tube does not have to make a huge voltage output,
> and even when a partially reactive load like an ESL speaker is used,
> the V1 gain tube still should not overload at HF because of the
> compensation generated by global NFB action.
>
> With 1mA of anode current, and at 200kHz, an unclipped output
> of 2.7vrms is possible into the 3k9 load.
>>
>> There are however two outputs. Can one shape the (unused)
>> HiZ output to modify the response of the LowZ output?
>
> Sure can, see above.
>
> When I plotted the OLG response in my workbook with pencil while
> watching voltage levels on the CRO, sure enough the
> -3dB point is at about 9k.
> This may seem to be awful, especially when deliberately created,
> but its necessary to avoid other phase shifts further up the band.
> It means that the 12dB is at about 35kHz, so because the OLG is lessened
> by 12dB at 35 kHz,
> there is only about 5 dB NFB applied, allowing for what phase shift does
> happen anyway.
> There is an F where OLG = 1.0, and even with a 0.22uF cap across the
> output as the sole load,
> phase shift is less than 180 degrees, so it does'nt oscillate.
> The 20pF compensation cap across the RFB does give some phase advance of
> the signal being fed back.
>
> This helps matters. This cap shouldn't be too large lest the FB become
> fully advanced by 90 degrees
> and thus cause a difference with input signal of 180 degrees where gain
> is above 1.0.
>
> There is still some overshoot and "ring" with a 5 kHz square wave with
> the Zobel networks in place.
>
> The Zobel networks stop oscillation, and IF they were they arranged to
> remove all ringing,
> the response of the amp would be be lousy, with a far too early HF pole.
>
> You see, a tube amp has many poles at HF and a few at LF, and they sum
> to form
> an equivalent of a band pass filter with multi L and C and R elements
> so that the rate of gain attenuation beyond where OLG = 1.0 becomes
> greater than 6dB/octave
> and further out becomes very much greater, perhaps 18dB/octave.
>
> Such rapid phase "turnover" at extremes of F cause oscillations in an
> amp if we are not careful.
>
> All such multi order LCR filters where the roll off exceeds 6dB/octave
> will generate ringing
> on square waves at the resonant F of the series/parallel LC components
> involved.
>
> Where such ringing is seen on the CRO, and where the ring exceeds twice
> the amplitude
> of the horizontal "settled" part of the squarewave, you will get a
> peaked sine wave response.
> Its normal to see this in tube amps when loaded with C loads.
> It is tested at low level, tyically with 1Vrms output when the amp is in
> pure class A and OLG is highest,
> and NFB IS connected.
> 2uF will give a peaked response of 6dB at 15kHz in a poor amp, and 60kHz
> in a very good amp.
> Usually this means the poorer amp has 8 times the leakage inductance of
> the better amp.
> As C is reduced in value with a poorer amp, the F peak will move higher,
> and be so high at some F that
> oscillation begins.
>
> A good amp will withstand 2uF being connected and the reponse at 20khz
> will rise
> by no more than 1dB and the sine wave peaking not more than 3dB at some
> F above 20kHz.
> All other values of C will not cause higher peaks than 3dB.
>
> Music isn't a square wave, so as long as the above criteria are met, the
> amp will be stable
> and sound well.
>
>
> Patrick Turner.
>>
>> Regards to all
>> Iain

Re: NFB101 Part 3


Iain,

I think you have relied to post from me but without your reply content
anywhere.

A slip of the click? or click of the slip perhaps?

Patrick Turner.



Iain Churches wrote:
>
> "Patrick Turner" <info@turneraudio . com .au> wrote in message
> news:47FDDE34.F381F389@turneraudio . com .au...
> >
> >
> > Iain Churches wrote:
> >>
> >> "Patrick Turner" <info@turneraudio . com .au> wrote in message
> >> news:47FCB5BF.82703CCD@turneraudio . com .au...
> >>
> >> > So, rather than measure very much, and rather than go utterly insane
> >> > trying to spend
> >> > 20 years on developing maths nobody else can possibly understand, the
> >> > PM
> >> > does it my way,
> >> > and uses a 0.22uF cap to see if oscillates at HF, then applies a Zobel
> >> > at V1 anode load
> >> > and he uses a radio tuning cap and pot to set the best values.
> >>
> >> But what does he do if he uses (as I do) a mu-follower
> >> for the input stage. There is no anode load.
> >
> > The Jolida 502 amp has a SRPP input stage using a 12AX7 with 470 ohms Rk
> > for the
> > bottom triode and 1k for the top triode.
> >
> > It has 16dB global NFB applied via a 27k R from the 8 ohm output
> > terminal.
> >
> > The SRPP is actually like a two element stage, with 2 triodes in series.
> > The circuit resistance looking at the anode circuit point of the bottom
> > triode
> > is made up of the effective RL above the anode in parallel with the
> > effective Ra of the bottom triode
> > due to a virtually un-bypassed Rk of 1k.
> > Respectively, there is about 150k above the anode, and about 130k
> > effective Ra below the anode connection.
> > Total is about 70k, and I used a 3k9 plus 270pF which
> > created a first pole at 159,000 / ( 0.00027 x 70,000 ) which is approx
> > 8.4kHz.
> > The 3k9 loads the bottom triode as the reactance of 270pF reduces so
> > that its gain as F rises
> > is much reduced.
> > By 200kHz, the Z 270pF = 294 ohms, so the Zobel has virtually no
> > reactive quality and it
> > has Z = 3k9, but it is resistive, and without phase shift. Gain at
> > 200kHz
> > will therefore be about 3.3, but without the ultimate
> > 90 degree phase lag caused by the C.
> >
> > It is obvious to those who understand the most basic analysis of very
> > simple easy to figure out
> > tube circuits that the place to put a Zobel in SRPP is from the bottom
> > anode to 0V.
> > If it was placed at the top tube cathode, the top tube's local follower
> > NFB action and current regulation
> > would drive the top tube into saturation and distortion as F rises.
> >
> > The purpose of the zobel is to load the tube at HF with a resistance
> > load
> > while preventing the phase shift of any C, lest the added phase shift
> > make the amp even more unstable despite the use of sone pure C across
> > the
> > gain tube output.
> > The gain tube does not have to make a huge voltage output,
> > and even when a partially reactive load like an ESL speaker is used,
> > the V1 gain tube still should not overload at HF because of the
> > compensation generated by global NFB action.
> >
> > With 1mA of anode current, and at 200kHz, an unclipped output
> > of 2.7vrms is possible into the 3k9 load.
> >>
> >> There are however two outputs. Can one shape the (unused)
> >> HiZ output to modify the response of the LowZ output?
> >
> > Sure can, see above.
> >
> > When I plotted the OLG response in my workbook with pencil while
> > watching voltage levels on the CRO, sure enough the
> > -3dB point is at about 9k.
> > This may seem to be awful, especially when deliberately created,
> > but its necessary to avoid other phase shifts further up the band.
> > It means that the 12dB is at about 35kHz, so because the OLG is lessened
> > by 12dB at 35 kHz,
> > there is only about 5 dB NFB applied, allowing for what phase shift does
> > happen anyway.
> > There is an F where OLG = 1.0, and even with a 0.22uF cap across the
> > output as the sole load,
> > phase shift is less than 180 degrees, so it does'nt oscillate.
> > The 20pF compensation cap across the RFB does give some phase advance of
> > the signal being fed back.
> >
> > This helps matters. This cap shouldn't be too large lest the FB become
> > fully advanced by 90 degrees
> > and thus cause a difference with input signal of 180 degrees where gain
> > is above 1.0.
> >
> > There is still some overshoot and "ring" with a 5 kHz square wave with
> > the Zobel networks in place.
> >
> > The Zobel networks stop oscillation, and IF they were they arranged to
> > remove all ringing,
> > the response of the amp would be be lousy, with a far too early HF pole.
> >
> > You see, a tube amp has many poles at HF and a few at LF, and they sum
> > to form
> > an equivalent of a band pass filter with multi L and C and R elements
> > so that the rate of gain attenuation beyond where OLG = 1.0 becomes
> > greater than 6dB/octave
> > and further out becomes very much greater, perhaps 18dB/octave.
> >
> > Such rapid phase "turnover" at extremes of F cause oscillations in an
> > amp if we are not careful.
> >
> > All such multi order LCR filters where the roll off exceeds 6dB/octave
> > will generate ringing
> > on square waves at the resonant F of the series/parallel LC components
> > involved.
> >
> > Where such ringing is seen on the CRO, and where the ring exceeds twice
> > the amplitude
> > of the horizontal "settled" part of the squarewave, you will get a
> > peaked sine wave response.
> > Its normal to see this in tube amps when loaded with C loads.
> > It is tested at low level, tyically with 1Vrms output when the amp is in
> > pure class A and OLG is highest,
> > and NFB IS connected.
> > 2uF will give a peaked response of 6dB at 15kHz in a poor amp, and 60kHz
> > in a very good amp.
> > Usually this means the poorer amp has 8 times the leakage inductance of
> > the better amp.
> > As C is reduced in value with a poorer amp, the F peak will move higher,
> > and be so high at some F that
> > oscillation begins.
> >
> > A good amp will withstand 2uF being connected and the reponse at 20khz
> > will rise
> > by no more than 1dB and the sine wave peaking not more than 3dB at some
> > F above 20kHz.
> > All other values of C will not cause higher peaks than 3dB.
> >
> > Music isn't a square wave, so as long as the above criteria are met, the
> > amp will be stable
> > and sound well.
> >
> >
> > Patrick Turner.
> >>
> >> Regards to all
> >> Iain

Re: NFB101 Part 3



"Patrick Turner" <info@turneraudio . com .au> wrote in message
news:47FF29D5.8DC1E1F4@turneraudio . com .au...
>
> Iain,
>
> I think you have relied to post from me but without your reply content
> anywhere.
>
> A slip of the click? or click of the slip perhaps?
>
> Patrick Turner.



Yes. Sorry :-(( My routine is to click "reply" and then store the
post without the reply added, in the drafts folder. Sometimes
one escapes.

I added the proper reply a few mins later.

regards
Iain



>
>
>
> Iain Churches wrote:
>>
>> "Patrick Turner" <info@turneraudio . com .au> wrote in message
>> news:47FDDE34.F381F389@turneraudio . com .au...
>> >
>> >
>> > Iain Churches wrote:
>> >>
>> >> "Patrick Turner" <info@turneraudio . com .au> wrote in message
>> >> news:47FCB5BF.82703CCD@turneraudio . com .au...
>> >>
>> >> > So, rather than measure very much, and rather than go utterly insane
>> >> > trying to spend
>> >> > 20 years on developing maths nobody else can possibly understand,
>> >> > the
>> >> > PM
>> >> > does it my way,
>> >> > and uses a 0.22uF cap to see if oscillates at HF, then applies a
>> >> > Zobel
>> >> > at V1 anode load
>> >> > and he uses a radio tuning cap and pot to set the best values.
>> >>
>> >> But what does he do if he uses (as I do) a mu-follower
>> >> for the input stage. There is no anode load.
>> >
>> > The Jolida 502 amp has a SRPP input stage using a 12AX7 with 470 ohms
>> > Rk
>> > for the
>> > bottom triode and 1k for the top triode.
>> >
>> > It has 16dB global NFB applied via a 27k R from the 8 ohm output
>> > terminal.
>> >
>> > The SRPP is actually like a two element stage, with 2 triodes in
>> > series.
>> > The circuit resistance looking at the anode circuit point of the bottom
>> > triode
>> > is made up of the effective RL above the anode in parallel with the
>> > effective Ra of the bottom triode
>> > due to a virtually un-bypassed Rk of 1k.
>> > Respectively, there is about 150k above the anode, and about 130k
>> > effective Ra below the anode connection.
>> > Total is about 70k, and I used a 3k9 plus 270pF which
>> > created a first pole at 159,000 / ( 0.00027 x 70,000 ) which is approx
>> > 8.4kHz.
>> > The 3k9 loads the bottom triode as the reactance of 270pF reduces so
>> > that its gain as F rises
>> > is much reduced.
>> > By 200kHz, the Z 270pF = 294 ohms, so the Zobel has virtually no
>> > reactive quality and it
>> > has Z = 3k9, but it is resistive, and without phase shift. Gain at
>> > 200kHz
>> > will therefore be about 3.3, but without the ultimate
>> > 90 degree phase lag caused by the C.
>> >
>> > It is obvious to those who understand the most basic analysis of very
>> > simple easy to figure out
>> > tube circuits that the place to put a Zobel in SRPP is from the bottom
>> > anode to 0V.
>> > If it was placed at the top tube cathode, the top tube's local follower
>> > NFB action and current regulation
>> > would drive the top tube into saturation and distortion as F rises.
>> >
>> > The purpose of the zobel is to load the tube at HF with a resistance
>> > load
>> > while preventing the phase shift of any C, lest the added phase shift
>> > make the amp even more unstable despite the use of sone pure C across
>> > the
>> > gain tube output.
>> > The gain tube does not have to make a huge voltage output,
>> > and even when a partially reactive load like an ESL speaker is used,
>> > the V1 gain tube still should not overload at HF because of the
>> > compensation generated by global NFB action.
>> >
>> > With 1mA of anode current, and at 200kHz, an unclipped output
>> > of 2.7vrms is possible into the 3k9 load.
>> >>
>> >> There are however two outputs. Can one shape the (unused)
>> >> HiZ output to modify the response of the LowZ output?
>> >
>> > Sure can, see above.
>> >
>> > When I plotted the OLG response in my workbook with pencil while
>> > watching voltage levels on the CRO, sure enough the
>> > -3dB point is at about 9k.
>> > This may seem to be awful, especially when deliberately created,
>> > but its necessary to avoid other phase shifts further up the band.
>> > It means that the 12dB is at about 35kHz, so because the OLG is
>> > lessened
>> > by 12dB at 35 kHz,
>> > there is only about 5 dB NFB applied, allowing for what phase shift
>> > does
>> > happen anyway.
>> > There is an F where OLG = 1.0, and even with a 0.22uF cap across the
>> > output as the sole load,
>> > phase shift is less than 180 degrees, so it does'nt oscillate.
>> > The 20pF compensation cap across the RFB does give some phase advance
>> > of
>> > the signal being fed back.
>> >
>> > This helps matters. This cap shouldn't be too large lest the FB become
>> > fully advanced by 90 degrees
>> > and thus cause a difference with input signal of 180 degrees where gain
>> > is above 1.0.
>> >
>> > There is still some overshoot and "ring" with a 5 kHz square wave with
>> > the Zobel networks in place.
>> >
>> > The Zobel networks stop oscillation, and IF they were they arranged to
>> > remove all ringing,
>> > the response of the amp would be be lousy, with a far too early HF
>> > pole.
>> >
>> > You see, a tube amp has many poles at HF and a few at LF, and they sum
>> > to form
>> > an equivalent of a band pass filter with multi L and C and R elements
>> > so that the rate of gain attenuation beyond where OLG = 1.0 becomes
>> > greater than 6dB/octave
>> > and further out becomes very much greater, perhaps 18dB/octave.
>> >
>> > Such rapid phase "turnover" at extremes of F cause oscillations in an
>> > amp if we are not careful.
>> >
>> > All such multi order LCR filters where the roll off exceeds 6dB/octave
>> > will generate ringing
>> > on square waves at the resonant F of the series/parallel LC components
>> > involved.
>> >
>> > Where such ringing is seen on the CRO, and where the ring exceeds twice
>> > the amplitude
>> > of the horizontal "settled" part of the squarewave, you will get a
>> > peaked sine wave response.
>> > Its normal to see this in tube amps when loaded with C loads.
>> > It is tested at low level, tyically with 1Vrms output when the amp is
>> > in
>> > pure class A and OLG is highest,
>> > and NFB IS connected.
>> > 2uF will give a peaked response of 6dB at 15kHz in a poor amp, and
>> > 60kHz
>> > in a very good amp.
>> > Usually this means the poorer amp has 8 times the leakage inductance of
>> > the better amp.
>> > As C is reduced in value with a poorer amp, the F peak will move
>> > higher,
>> > and be so high at some F that
>> > oscillation begins.
>> >
>> > A good amp will withstand 2uF being connected and the reponse at 20khz
>> > will rise
>> > by no more than 1dB and the sine wave peaking not more than 3dB at some
>> > F above 20kHz.
>> > All other values of C will not cause higher peaks than 3dB.
>> >
>> > Music isn't a square wave, so as long as the above criteria are met,
>> > the
>> > amp will be stable
>> > and sound well.
>> >
>> >
>> > Patrick Turner.
>> >>
>> >> Regards to all
>> >> Iain

Re: NFB101 Part 3

On Tue, 08 Apr 2008 08:21:00 +0100, Ian Thompson-Bell <nospam@free.co.uk> wrote:

>I am working on the next part of NFB101. It is turning into a a huge
>mass of maths which I think will put off a lot of people. So I decided
>to split it into two parts. The first covers the basics and then when
>the maths gets heavy it just quotes the result and refers instead to an
>appendix where the heavy maths resides. That way the main section can
>concentrate on the meaning of the results rather than their derivation
>the the details are available to anyone who is interested in them.
>
>Before I commit to this split the groups views would be welcome.
>
>Cheers
>
>Ian

In addition to RDH4, another good reference is:

* w w w .amazon . com /Feedback-Amplifier-Principles-Sol-Rosenstark/dp/067222545X/ref=sr_1_1?ie=UTF8&s=books&qid=1207888217&sr=1-1

Re: NFB101 Part 3

The Phantom wrote:
> On Tue, 08 Apr 2008 08:21:00 +0100, Ian Thompson-Bell <nospam@free.co.uk> wrote:
>
>> I am working on the next part of NFB101. It is turning into a a huge
>> mass of maths which I think will put off a lot of people. So I decided
>> to split it into two parts. The first covers the basics and then when
>> the maths gets heavy it just quotes the result and refers instead to an
>> appendix where the heavy maths resides. That way the main section can
>> concentrate on the meaning of the results rather than their derivation
>> the the details are available to anyone who is interested in them.
>>
>> Before I commit to this split the groups views would be welcome.
>>
>> Cheers
>>
>> Ian
>
> In addition to RDH4, another good reference is:
>
> * w w w .amazon . com /Feedback-Amplifier-Principles-Sol-Rosenstark/dp/067222545X/ref=sr 1 1?ie=UTF8&s=books&qid=1207888217&sr=1-1

Difficult to tell - it seems fairly rare and rather expensive. I'll get
my local library to find me a copy. I don't suppose it is available on
line anywhere?

Cheers

Ian