[The Phantom]

Page 371 of RDH4 has a section on corrective networks of the "step" type.

Reference H9 discusses them in detail, and that paper, "Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email address here and get a
copy.

[Ian Thompson-Bell]

The Phantom wrote:
Page 371 of RDH4 has a section on corrective networks of the "step" type.

Reference H9 discusses them in detail, and that paper, "Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email address here and get a
copy.


Can't see it there right now. Can you email me a copy please. Send to
ianbellATukfsnDOTorg

Cheers

Ian

[Patrick Turner]

Ian Thompson-Bell wrote:

The Phantom wrote:
Page 371 of RDH4 has a section on corrective networks of the "step" type.

Reference H9 discusses them in detail, and that paper, "Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email address here and get a
copy.

Can't see it there right now. Can you email me a copy please. Send to
ianbellATukfsnDOTorg

Cheers

Ian

I cannot see the original post by The Phuntom
but he's brought the page 371 of RDH4 to my attention.

I hadn't realised RDH4 had devoted a page to such simple and worthwhile
advice
and with simple formulas to work out the two poles
of a step network at HF and at LF.

However, the comment at the bottom of page 371 re Fig 7.59C
shows a typical global FB arrangement in a tube amp
but there are two Rfb, R1 abd R2 and there is C across R2 so that at HF
the
feedback is is increased, ie, ß becomes larger and they say
stability is increased.

In practice, I have found this doesn't work very well, and the
last thing one wants to do is greatly increase ß at say 70kHz,
and therefore increase the dB of NFB applied and therefore reduce
bandwidth.

To my mind, its best to use only 1 Rfb, then apply a cap across the
whole R
which looks like one is trying to increase FB as F rises, and at some
very high
F the C is a short and ALL the output is fed back.

But the value of C is chosen so the point where the FB increases
by 3dB is at a very high F and where C is a short is at MHz,
where open loop gain is very low.

A typical full FB network will have 50 ohms between
V1 cathode biasing network and 0V, and say 1k from the top of 50 ohms to
the output,
so ß = 50 / 1050.

A typical value for C across the 1k might be 470pF, and this reacts with
1k
so that ZR&C = 700 ohms at 338kHz, and so the effect of the 470pF on
gain with FB at say
3.3kHz and 33kHz is quite negligible, except that it gives a slight
but useful phase advance to compensate for the lag in the open loop
gain.
If this C is made too large, the ultimate phase shift of +90 degrees
of output signals at HF might become 180 degrees out of phase with
open loop signals at some high F, maybe 500kHz, and the damn amp
oscillates.
Placing some R in series with the C usually lessens the wanted little
bit of
phase advance achieved with this C.

The RDH4 method of having the parallel R&C network connected with an
additional R in series
is worth trying though where you have an amp which has a peaked sine
wave response after FB has been applied
at say between 20kHz and 100kHz.
Usually a peaked response with NFB and with an R load shows that phase
shift is pretty poor
in the open loop, and when NFB is applied, the amp is barely stable and
the
phase difference between FB signal and input is over 90 degrees, but
less than 180 degrees.
Such amps usually always oscillate badly at HF when a 0.22 uF is
connected across the output,
or perhaps when no load is connected at all.

Rather than rely on FB network tailoring completely,
I prefer to apply the gain step networks to reduce the open loop gain
at below 20Hz and above say 10kHz, and provide zobel networks
across the OPT half primaries if they work well, and across the OPT
secondary,
if they work well. Not everything works as well as something else.
In triode output amps, a Zobel is usually needed across the secondary,
say 10 ohms plus 0.22 uF, to damp the parasitic LC of the OPT at HF when
no load is connected.
In beam/pentode/UL, CFB, the OPT promary may need Zobels.

Sometimes with Zobels connected, and a suitable C across Rfb,
there is no need for stepping the HF open loop gain.

One should only use the bare minimum of compensation and
step networks to ensure stability with any C load is fine.
There isn't any load an amp is ever going to work with that's worse than
a pure C at HF
except perhaps a series LC which will have very low Z at the series
resonant F.
But series L+C are never designed into a speakers
where the Fo is above 20kHz.
The pure C causes an ultimate additional phase shift of 90 degrees of
lag,
and as F rises on the way towards that 90d the open loop gain may have
90d caused by other accumulated
phase lags, and if gain is above 1, it will oscillate because the total
lag
has reached 180d where gain is still above 1.

Trying to achieve too much with corrective neworks is
catching the horse that has bolted.

In recent re-engineering attempts in a large 45W/chan VAC amp and
in two Audion 25W amps with 4 and 2 300B used respectively,
the GNFB becomes quite difficult to apply when you try going over 10dB
because the OPT quality in both amps is not perfect,
and most certainly not as good as the top line Partridge
trannies of yesteryear, and not as good as something i'd wind.
Its simply due to poorly conceived interleaving and winding insulation
thicknesses.
Very few makers get these issues anywhere near as good as they ought to.

Patrick Turner.

[The Phantom]

On Thu, 12 Jun 2008 13:15:58 GMT, Patrick Turner
wrote:



Ian Thompson-Bell wrote:

The Phantom wrote:
Page 371 of RDH4 has a section on corrective networks of the "step" type.

Reference H9 discusses them in detail, and that paper, "Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email address here and get a
copy.

Can't see it there right now. Can you email me a copy please. Send to
ianbellATukfsnDOTorg

Cheers

Ian

I cannot see the original post by The Phuntom
but he's brought the page 371 of RDH4 to my attention.


I'd be glad to send it to you if you post a suitable email address.

[Ian Iveson]

The Phantom wrote:

Page 371 of RDH4 has a section on corrective networks of
the "step" type.

Reference H9 discusses them in detail, and that paper,
"Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email
address here and get a
copy.

Good of you to post it, thanks, but it won't open for me.
Your other pdf on the cathodyne does open. I don't know much
about pdf files. Maybe my Acrobat reader doesn't, either.

Ian

[The Phantom]

On Fri, 13 Jun 2008 00:27:29 +0100, "Ian Iveson"
wrote:

The Phantom wrote:

Page 371 of RDH4 has a section on corrective networks of
the "step" type.

Reference H9 discusses them in detail, and that paper,
"Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email
address here and get a
copy.

Good of you to post it, thanks, but it won't open for me.
Your other pdf on the cathodyne does open. I don't know much
about pdf files. Maybe my Acrobat reader doesn't, either.

Ian

I created the Cathodyne paper with Acrobat 5, but I suspect the
"corrective network" paper (not created by me) was created with a much more
recent version of Acrobat. You should probably get the latest version of
Acrobat reader off the web.

[Patrick Turner]

The Phantom wrote:

On Thu, 12 Jun 2008 13:15:58 GMT, Patrick Turner
wrote:



Ian Thompson-Bell wrote:

The Phantom wrote:
Page 371 of RDH4 has a section on corrective networks of the "step" type.

Reference H9 discusses them in detail, and that paper, "Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email address here and get a
copy.

Can't see it there right now. Can you email me a copy please. Send to
ianbellATukfsnDOTorg

Cheers

Ian

I cannot see the original post by The Phuntom
but he's brought the page 371 of RDH4 to my attention.


I'd be glad to send it to you if you post a suitable email address.

RDH4 is online and can be downloaded.

Patrick Turner.

[The Phantom]

On Fri, 13 Jun 2008 15:36:38 GMT, Patrick Turner
wrote:



The Phantom wrote:

On Thu, 12 Jun 2008 13:15:58 GMT, Patrick Turner
wrote:



Ian Thompson-Bell wrote:

The Phantom wrote:
Page 371 of RDH4 has a section on corrective networks of the "step" type.

Reference H9 discusses them in detail, and that paper, "Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email address here and get a
copy.

Can't see it there right now. Can you email me a copy please. Send to
ianbellATukfsnDOTorg

Cheers

Ian

I cannot see the original post by The Phuntom
but he's brought the page 371 of RDH4 to my attention.


I'd be glad to send it to you if you post a suitable email address.

RDH4 is online and can be downloaded.

I was referring to the "original post by The Phuntom", not RDH4.


Patrick Turner.

[Ian Iveson]

flipper said:

Perhaps you're not doing OE's "combine and decode" and
it's a
multi-part post that won't be a valid file unless the
whole thing is
first put together.

Er, right. My fault. Never done this kind of thing before.

Generally I look at a document before I download it, so I
just selected part one and tried to open it.

Never noticed "combine and decode" before. Is there a
"separate and encode" somewhere?

Thanks. Missed the first document, but you helped me get the
second.

Hopefully I can now undo all the fuss and palaver of
installing reader 8 and go back to 4, which occupies a tenth
of the space.

Ian

[Ian Iveson]

"The Phantom" wrote in message
...
Page 371 of RDH4 has a section on corrective networks of
the "step" type.

Reference H9 discusses them in detail, and that paper,
"Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email
address here and get a
copy.

Missed it now

Please email me a copy, thanks.



Ian

[The Phantom]

On Sat, 14 Jun 2008 12:07:21 +0100, "Ian Iveson"
wrote:

flipper said:

Perhaps you're not doing OE's "combine and decode" and
it's a
multi-part post that won't be a valid file unless the
whole thing is
first put together.

Er, right. My fault. Never done this kind of thing before.

Generally I look at a document before I download it, so I
just selected part one and tried to open it.

Never noticed "combine and decode" before. Is there a
"separate and encode" somewhere?

Thanks. Missed the first document, but you helped me get the
second.

Hopefully I can now undo all the fuss and palaver of
installing reader 8 and go back to 4, which occupies a tenth
of the space.

Get the free .pdf reader at http://www.foxitsoftware.com/

It's small and it will read the latest .pdf format.


Ian

[Ian Thompson-Bell]

Patrick Turner wrote:

Ian Thompson-Bell wrote:
The Phantom wrote:
Page 371 of RDH4 has a section on corrective networks of the "step" type.

Reference H9 discusses them in detail, and that paper, "Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email address here and get a
copy.
Can't see it there right now. Can you email me a copy please. Send to
ianbellATukfsnDOTorg

Cheers

Ian

I cannot see the original post by The Phuntom
but he's brought the page 371 of RDH4 to my attention.

I hadn't realised RDH4 had devoted a page to such simple and worthwhile
advice
and with simple formulas to work out the two poles
of a step network at HF and at LF.


Just a word of warning about these simple formulas, the RDH4 example is
for a pentode and it looks as though they therefore ignore ra as being
much higher than the anode load. Obviously triodes are different. For a
triode if we say Rt is the parallel combination of ra and the anode
load, a R and C are the compensation network then:

There is a pole at 1/(Rt+R)C which is where the response will start to
drop at 6db/octave.

There is a zero at 1/RC which RDH4 calls the flattening out frequency.

Cheers

Ian

[Patrick Turner]

Ian Thompson-Bell wrote:

Patrick Turner wrote:

Ian Thompson-Bell wrote:
The Phantom wrote:
Page 371 of RDH4 has a section on corrective networks of the "step" type.

Reference H9 discusses them in detail, and that paper, "Corrective Networks
for Feedback Circuits", is posted over on ABSE.

Anybody who can't get it there can provide an email address here and get a
copy.
Can't see it there right now. Can you email me a copy please. Send to
ianbellATukfsnDOTorg

Cheers

Ian

I cannot see the original post by The Phuntom
but he's brought the page 371 of RDH4 to my attention.

I hadn't realised RDH4 had devoted a page to such simple and worthwhile
advice
and with simple formulas to work out the two poles
of a step network at HF and at LF.


Just a word of warning about these simple formulas, the RDH4 example is
for a pentode and it looks as though they therefore ignore ra as being
much higher than the anode load. Obviously triodes are different. For a
triode if we say Rt is the parallel combination of ra and the anode
load, a R and C are the compensation network then:

There is a pole at 1/(Rt+R)C which is where the response will start to
drop at 6db/octave.

There is a zero at 1/RC which RDH4 calls the flattening out frequency.

Cheers

Ian

Circuit designers will realise that if you have an EF86 with 200k RLdc,
and following cap coupled grid bias R of say 470k, and the Ra is
1M or greater at say 0.6mA of Ia, the Ra can be ignored
because of the tiny difference it makes to include it.
These stepper networks are not sharply critical
unless a lot of NFB is applied, and the OPT is a ****ty bean counter
designed POS.

Triode Ra is much lower than the total RL connected, so it must be
included in calcs
but with the adjustment for an unbypassed Rk.

Turbe amp design is all about considering the 10 different issues
surrounding the selection of
every single damn part in the box.

For those who solder many amps together routinely,
such as myself, we don't need to calculate because we KNOW
what the stepper network does in principle very well and we
just use practical measures to apply R&C by well considered experiments
to avoid making silly mistakes on paper.

Patrick Turner.



podf SSED uT