I have just posted a circuit at abse and abpr fo those
who like a little brain teaser circuit application of the idea of error
correction in lieu
of applied NFB.

Consider the typically arranged LTP driver amp and 6550 PP UL output
stage
which is shown in the jpg.

The normal way to apply feedback if we *did not* want to include another
gain stage
so that we would reduce the thd by say 4 times would be to
apply the output voltage of 14.4v to to the g1/V3.
The input to V2 would then have to be 14.4v + 3.4 vg-g of the LTP, so
17.8 av all up.
This is a lot of drive voltage, and we'd have trouble keeping the
linearity of a preamp
to less than say 0.05% thd at full tilt.
However the FB would reduce thd in the same way
sensitivity is reduced from 3.4 v at the input to 17.8v which is a gain
reduction
of 17.8/3.4 = 5.23 times = 14.4 dB.

In the case of the NFB method, even with the FB connected, +D volts, the
distortion voltage
is able to be measured at the output.
But the output applies this to the feedback input port at g1/V3, so +Dv
gets amplified
by the open loop gain of the amp to make a signal of -4.24Dv at the
output.
But how can this be if we already know only Dv is present at the output?

But already a voltage of +5.24Dv will be trying to appear at the output
if we didn't have any NFB, and the -4.24v NFB correction voltage applied
subtracts from
the +5.24Dv to leave just Dv.
This further supports the idea Dn is reduced A+1 times with series
voltage NFB,
where Dn = open loop distortion, A = open loop gain, and there is little
phase shift
in frequencies of concern.

The price we had to pay for the NFB correction was a huge loss of
ampifier sensitivity.

If there was a way to apply only the Distortion voltage sample at g1/V3,
then the input voltage to the
g1/V2 would stay similar, or be able to be less than what is used when
no NFB is applied.

So we need to find some way of nulling out the signal content of the NFB
signal
and sifting out the Dv only.

Now let's look at my schematic again for the typical driver LTP and UL
output stage.

The input stage isn't part of the drive amp, but instead it is an
auxilliary to it.

The V1 is a paralleled 6SN7 which has a CCS DC anode supply, not
absolutely
necessary, but elegantly better for this scheme to work well.

The input voltage to the amp at g1/V2 is +1.7av, and this is also
applied to the
g1/V1 input stage.

There is a 22k resistor returned from the +14.4v to the
anode of V1 but separated by a DC blocking cap of say 0.47 uF or
greater.
The output of V1 is fed to g1V3, and all has been arranged so that
-1.7av
is generated at the output of V1 so V1 has a phase inverter function as
well as a "signal nuller"
and distortion re-director.

Normally, the anode of V1 would swing about -30 av with +1.7 av input.
But the 22k R from the output stops most but not quite all of this
swing.
The output resistance of the V1 at its anode is 47k, due to the use of
unbypassed Rk total
of about 2k, ( 1k plus about 1k, with provision for baising g1V1 via
220k).
In fact by adjustment of the lower Rk we could better balance the +/-
1.7 av applied to the
LTP.
The 22k supplies a signal current to the V1 anode which opposes the
anode normal flow as a result
of the g1 voltage effect, and in fact the tube sees an anode load of 1.7
/ 0.73 mA = 2.33 kOhms.
The cathode R of 2k make the total tube load = 4.33k, and the 2k is
used to make local current FB for V1, and since output voltage is so
low,
the thd will not be great; in fct V1 has the usual large amount of local
NFB as
one has in a concertina phase inverter, which it is, in this case..

Now as I said, the Ro of V1 at the anode is 47k; that's the dynamic
generator impedance,
which is calculated as Ra + [(U+1) x Rk], and the CCS doesn't count in
because its impedance in parallel
with Ro is in the megohm region.

So the 22k and 47k act as a R divider, and in this case the circuit
tries to
send the voltage in opposite polarities at each end of the divider..
Let us suppose that with set up as shown that we had Dv appear as thd at
the output.
Then (47 / 69) x Dv, or +0.68 Dv will appear at the divider junction,
which is the anode of V1.
This is applied to the g1/V3 input, as well as the 1.7 va signal, and
the
open loop gain of 4.24 amplifies the +0.68Dv to appear as -2.88Dv at the
output.
And this subtracts from whatever open loop Dv would have appeared
without this error
correction, so the original amount of D would have been +3.88Dv from
which the applied -2.88 subtracts,
leaving +Dv as we observe, with error connection applied.
So, we have removed the necessity for the additional usual gain stage in
the signal path,
and been able to reduce thd from 3.88 to 1, and this is about an 11 dB
thd reduction,
about equal to the use of the NFB.
And the amp is sensitive to only 1.7 av instead of 3.4v, open loop.

You mathematicians can draw up the formula for the amount of
error correction in Db with regards to the divider resistances, V1 Rk,
U, Ra, Gm, and amplifier open loop
gain between Vg-g of V2/3 to the output, but I ain't in any rush, since
whatever
one comes upo with will be more difficult to weild tham the normal
basic simple feedback equations.

The concept is also more complex, but as I show, its appliaction is dead
simple, and uses
no more tubes than an existing amp would.

It is possible to apply *more* correction, by using a higher U, higher
transconductance triode at V1,
such as a 12AT7, and get Ro to be a lot higher, so the
fraction of the distortion voltage fed back into the amp is increased,
whilst the linearity of V1 is increased. The linearity of V1 is
important, since we don't want
its thd to be fed into the amp.

Its also poosible to have 12AT7 for the LTP, and also possible to
arrange that the LTP
gets a signal input to only g2/V2, and have only the Dv applied to
g1/V3.

It works OK to reduce Ro at the output.
Suppose the load is reduced from 6 to 3 ohms, then the output voltage
would fall,
and the voltage at the divider nuller 22k and 47k would increase from
the
-1.7v, and the Vg-g applied to V2/3 will increase, thus tending to
boost the output v.

Similarly, if load was 20 ohms instead of 6ohms, the output voltage
would rise,
reducing the voltage level at V1 anode, and thus across Vg-g of V2/3,
so output voltage tends to get reduced from what it otherwise would be.

If you have already built a few things, and have time, try this circuit
on the
experimental amp you may have.

Regards to all,

Patrick Turner.

Pooh Bear wrote:

Gregg wrote:

Hi Pat,

That's *still* NFB, just a different kind ;-)

I didn't read the whole post - it was somewhat long-winded. Was overall
NFB replaced with local ? Some ppl think that local NFB ( e.g. cathode
resistors ) that linearises individual stages doesn't 'count'.

Graham

Someone is already saying I am still applying NFB.
But I ain't.

Globally applied series voltage includes a fraction of the output signal
voltage
as well as the distortion applied to a second amp input and the phase of
the
signal is the same as the signal input.
This is not being done in my design here; I deleiberately
have used an input auxilliary amp to NULL the signal current being sent
back
along the "feedback path", ie, along the 22 k R.

It is possible to adjust the set up so NO fedback or other signal voltage
appears at g1/V3 input, only the fraction of the distortion may appear.


To understand, don't be a clod; *read the whole post
and do try to follow what I am saying!!!!!!!!!!!!!!!!!!!!!!!*

If yer don't read, yer won't know :-)

There is local current FB in the way V1 is set up.

Never mind what folks say about whether local feedback counts or not.
Most of what such folks say about such should be taken with a block of
salt.

Patrick Turner.

Opps, somehow the image got sent to r.a.t.

OK, Mr God of NGs, I am so sorry!!!!!!!!.

Be so kind as to remove the image in replies to my post, lest it transfer
itself
uneccessarily, and an angry posse chase me down.

Patrick Turner.

"Patrick Turner" a écrit dans le message de news:
...

I have just posted a circuit at abse and abpr ......
[. . . ]

Does not appear to me, could you Email dirctly ?

Yves.

Patrick Turner said:

Opps, somehow the image got sent to r.a.t.

OK, Mr God of NGs, I am so sorry!!!!!!!!.

Soon, Rev. Phil Allison will be all over you!

--
Sander deWaal
"SOA of a KT88? Sufficient."

Patrick Turner wrote:
I have just posted a circuit at abse and abpr fo those
who like a little brain teaser circuit application of the idea of error
correction in lieu
of applied NFB.


Unless I've missed something back in electrical engineering
classes in college, NFB *is* error correction. Within the
dynamic range of the circuits in the amps. If you have
clipping though, correction cannot fix it.

Robert Casey wrote:

Patrick Turner wrote:
I have just posted a circuit at abse and abpr fo those
who like a little brain teaser circuit application of the idea of error
correction in lieu
of applied NFB.


Unless I've missed something back in electrical engineering
classes in college, NFB *is* error correction. Within the
dynamic range of the circuits in the amps. If you have
clipping though, correction cannot fix it.

All NFB schemes include "feeding back"
a part of the output signal to the input to be compared with the input
signal,
and the distortion in the signal fed back s amplified to cancel itself.

Error correction does the same thing with the distortion signal, but
without feeding back a fraction of the signal.

There are two things to consider. One is the output signal,
the other is the distortion voltage, Dv, within it.
If we have a resistance divider, and we create a pure oppositely phased
signal at the input of the amp,
we can create a null in the signal voltage, but because Dv only occurs at one
end
of the divider, the output end, then a fraction of it will occur in the
centre of the divider.

The Dv can be amplified, and re-applied to the amp in such a way to
cancel the distortion production, thus avoiding gain reduction, or it can be
applied to
the normally grounded end of the load, so the same distortion occurs at
both ends of the load, and thus no distortion currents flow through the load,

and Motzart sounds like MOtzart, not Led Zeppilin. And all this even though
no attempt has been made
to try to prevent the thd in the amplifiers concerned.

Sure there are limits, like with normal NFB.

Patrick Turner.

Patrick Turner wrote:

Robert Casey wrote:


Patrick Turner wrote:

I have just posted a circuit at abse and abpr fo those
who like a little brain teaser circuit application of the idea of error
correction in lieu
of applied NFB.


Unless I've missed something back in electrical engineering
classes in college, NFB *is* error correction. Within the
dynamic range of the circuits in the amps. If you have
clipping though, correction cannot fix it.


All NFB schemes include "feeding back"
a part of the output signal to the input to be compared with the input
signal,
and the distortion in the signal fed back s amplified to cancel itself.

Error correction does the same thing with the distortion signal, but
without feeding back a fraction of the signal.

There are two things to consider. One is the output signal,
the other is the distortion voltage, Dv, within it.
If we have a resistance divider, and we create a pure oppositely phased
signal at the input of the amp,
we can create a null in the signal voltage, but because Dv only occurs at one
end
of the divider, the output end, then a fraction of it will occur in the
centre of the divider.

The Dv can be amplified, and re-applied to the amp in such a way to
cancel the distortion production, thus avoiding gain reduction, or it can be
applied to
the normally grounded end of the load, so the same distortion occurs at
both ends of the load, and thus no distortion currents flow through the load

I think that this is called "feed forward". Sometimes used in CATV
system amps. Should work if you can use the main amp's output without
any of the error amp's correction getting into it, when you want to
get Dv. A very low output impedance would prevent most any of the
error amp's contribution to the load from "back-feeding" the main amp's
output, thus avoiding contamination of the main amp's output signal.

input----+------main amp--+----------load--+
| | |
| +-- -error-------+
+---Delta t--------- +amp

Presumably the errors will be small enough that a lower
powered error amp can null out the main amp's distortion.
However the time delay thru the main amp must be less than
a fraction of a wavelength of the highest desired audio
frequency's worth of time. Or else you can get reinforcement
of high frequency distortions and even high frequency clean signals.
So some sort of delta t delay line may be needed.

If the main amp clips, the error amp can provide the clipped
off portion of the waveform to the load. Regular NFB can't
do this.

In the practical side of engineering design, it may be
cheaper to just build a high quality main (high powered) amp
than to build both a high powered distorting amp plus a
high quality low powered error amp.

Patrick Turner wrote in message ...
Globally applied series voltage includes a fraction of the output
signal
voltage
as well as the distortion applied to a second amp input and the phase of
the
signal is the same as the signal input.
This is not being done in my design here; I deleiberately
have used an input auxilliary amp to NULL the signal current being sent
back
along the "feedback path", ie, along the 22 k R.

It is possible to adjust the set up so NO fedback or other signal voltage
appears at g1/V3 input, only the fraction of the distortion may appear.



Patrick Turner.

Hi Patrick,

You have re-invented Malcolm Hawksford's "Output Error Correction
Scheme" in a slightly different form. Hawksford's published (JAES)
design was added to the unity gain output stages (emitter follower or
source follower) of solid state amplifiers and is used in a number of
commercial amplifiers today. For an amplifier with significant gain,
the setting of the error correction feedback gain will be more
critical, so this is usually just added to a normal feedback design
with unity gain output stage. Also, the summation point of the error
feedback (V1 plate in your design) needs to have a pretty stable
impedance to maintain the correct level of error correction. Lots of
discussion of the Hawksford scheme on the "diyAudio" solid state
forum: ( a search on "Hawksford" will turn up quite a few discussion
threads and links to the original Hawksford articles)

http://www.diyaudio.com/forums/showt...113&highlight=

and one thread on the "diyAudio" tube forum too, where it is used
with a circlotron type output stage:

http://www.diyaudio.com/forums/showt...001&highlight=

Don

Robert Casey wrote:

Patrick Turner wrote:

Robert Casey wrote:


Patrick Turner wrote:

I have just posted a circuit at abse and abpr fo those
who like a little brain teaser circuit application of the idea of error
correction in lieu
of applied NFB.


Unless I've missed something back in electrical engineering
classes in college, NFB *is* error correction. Within the
dynamic range of the circuits in the amps. If you have
clipping though, correction cannot fix it.


All NFB schemes include "feeding back"
a part of the output signal to the input to be compared with the input
signal,
and the distortion in the signal fed back s amplified to cancel itself.

Error correction does the same thing with the distortion signal, but
without feeding back a fraction of the signal.

There are two things to consider. One is the output signal,
the other is the distortion voltage, Dv, within it.
If we have a resistance divider, and we create a pure oppositely phased
signal at the input of the amp,
we can create a null in the signal voltage, but because Dv only occurs at one
end
of the divider, the output end, then a fraction of it will occur in the
centre of the divider.

The Dv can be amplified, and re-applied to the amp in such a way to
cancel the distortion production, thus avoiding gain reduction, or it can be
applied to
the normally grounded end of the load, so the same distortion occurs at
both ends of the load, and thus no distortion currents flow through the load

I think that this is called "feed forward". Sometimes used in CATV
system amps. Should work if you can use the main amp's output without
any of the error amp's correction getting into it, when you want to
get Dv. A very low output impedance would prevent most any of the
error amp's contribution to the load from "back-feeding" the main amp's
output, thus avoiding contamination of the main amp's output signal.

In the sample of this type of error correction which I tendered,
the V1 tube acts as a low thd amp to provide the first R of the nulling divider.
There are other ways. A mu follower is OK, but you have to have a R1 arm to balance
the
R2 arm from the output, since a U follower is a low Ro amp.
If you you had a pentode for R1, it also works well to produce little if any gain,
and to oppose the current in R2.

In my case, I used V1 to allow a little phase inversion to get equal drive signals
to
the LTP of the power amp, but it all could have been set up so that for
a 6 ohm load, zero signal voltage appeared at g1/V3, and only a fraction of Dv
appeared.
But when the load is vried, the output stage gain varies, so the signal at g1V3 is
then going to get some signal voltage as part of its error message to keep the
output
voltage the same.

So if the amount of signal produced by the V1 tube is low, then its thd is also
reduced,
but where the high effective Ra is exploited as in my case, there is a reliance on
a linear
Gm transfer, and so if a pentode or higher gain triode is used, plenty of local
current FB is needed to linearize the operation of V1.
However, whatever thd is produced by V1, call it +Nv, is applied to g1/V3, and
amplified to become
-4.24Nv at the output, and this creates a current in the R1-R2 nulling network
which opposes the +Nv at the anode of V1, and so on.
So V1 has its own shunt FB network set up which helps it keep linear.



input----+------main amp--+----------load--+
| | |
| +-- -error-------+
+---Delta t--------- +amp

Presumably the errors will be small enough that a lower
powered error amp can null out the main amp's distortion.

That is what I suggest is true.


However the time delay thru the main amp must be less than
a fraction of a wavelength of the highest desired audio
frequency's worth of time. Or else you can get reinforcement
of high frequency distortions and even high frequency clean signals.
So some sort of delta t delay line may be needed.

In my experiments with a similar but slightly more complex arrangement some years
ago,
the was a limit for ho much Dv could be amplified and applied to the main amps
second port input.
It would go unstable with C loads due to the phase shift, so like all things,
there are limits.



If the main amp clips, the error amp can provide the clipped
off portion of the waveform to the load. Regular NFB can't
do this.

No, not quite right.
If ther main amp clips, then its because of rail clipping, or grid current,
and the ampo has run out of gain period, game over, and no amount
of NFB, or error correction can force the amp to stay linear beyond the clipping,
unless we did it all differently, and had a second amp which
acted to apply the main amps Dv to the earthy "dead" end of the load,
and applied some voltage to allow the voltage across the load to continue rising
linearly after clipping,
all done in such a way so that no distortion currents flowed in the load, although
massive distortions voltages could be seen at each end of the load.

This way the main amp's thd/imd and phase shift is all left uncorrected.
The second amp is grossly inefficient, and such a scheme is quite messy,
although perhaps not with SS.



In the practical side of engineering design, it may be
cheaper to just build a high quality main (high powered) amp
than to build both a high powered distorting amp plus a
high quality low powered error amp.

Well yes.

V1 can be a pentode, and so its Ra' can be megohms,
so R1 of the null network can be far higher than R2, which is 22k in my case.
Therefore nearly all the Dv at the output can be applied to g1/V3,
and as much thd reduction occurs as in the case I cited where all the output
voltage was applied
back as NFB to g1/V3.

If you had 22k back to a true null point, the current in 22k is 14.4 / 22k = 0.65
mA,
and if a pentode V1 is used with Gm = 4 mA/v, like a 6AU6, then only
0.163 v between g1/V1 and k1/V1 is needed to oppose the I in the 22k,
thus nulling all its voltage at the null point.
Then the pentode will simply need whatever Rk to provide current FB
to allow the input voltage to g1/V1 to equal that applied to g1/V2.
If the input voltage was say 1.7v, then the CFB voltage is 1.7v - 0.163v, = 1.53v,
and in effect the local CFB at V1 with a pentode could be 20 dB, enough to
keep it linear enough, by comparison with a triode as V1.

Where there is no signal output voltage from V1 at all, even when a signal is
applied to its grid,
then its RL is zero ohms.
This is quite a queer idea, but from the tube's point of view, it sees a negative
load,
being the 22k in the case I posted. The tube has an even queerer idea of the world
around itself,
but is unaware of it.
Its g1 goes +ve, and normally Va swings a negative signal.
But no, the 22k has a +ve swing at the far away main amp output, and from the
tube's point of view,
its increased current flow matches that in the 22k, and so no voltage signal change
occurs
at the anode.
Its easier to understand if one draws a model of the V1 tube a low impedance
generator
producing U x Vg1 G1 at the gene output, with some resistance in series from the
output
to what is the anode in the real world. This series R is the dynamic anode
resistance, or Ra.
In the case of a pentode, U is perhaps 1,000, and Ra = 1 Mohm, but the model still
works.

From what I know about the method I have proposed, it seems important the BW
of V1 is high, and I have no reason to think it wouldn't be.
The miller effect at the LTP g1/V2 input would slightly effect the
stability perhaps, so some snubber network across V1 anode to 0V
may be prudent because we don't need any error correction above 50 kHz.
The 0.47 cap from V1 anode to 22k
may include a 0.1 and 100k, to give less error correction below 20 Hz,
where it also isn't needed.

I have not tried out all possibilities, there is a shortage of daily hours.

Patrick Turner.

smoking-amp wrote:

Patrick Turner wrote in message ...
Globally applied series voltage includes a fraction of the output
signal
voltage
as well as the distortion applied to a second amp input and the phase of
the
signal is the same as the signal input.
This is not being done in my design here; I deleiberately
have used an input auxilliary amp to NULL the signal current being sent
back
along the "feedback path", ie, along the 22 k R.

It is possible to adjust the set up so NO fedback or other signal voltage
appears at g1/V3 input, only the fraction of the distortion may appear.



Patrick Turner.

Hi Patrick,

You have re-invented Malcolm Hawksford's "Output Error Correction
Scheme" in a slightly different form.

No doubt someone has done similar before.
But I have never seen it done with tubes.



Hawksford's published (JAES)
design was added to the unity gain output stages (emitter follower or
source follower) of solid state amplifiers and is used in a number of
commercial amplifiers today. For an amplifier with significant gain,
the setting of the error correction feedback gain will be more
critical, so this is usually just added to a normal feedback design
with unity gain output stage. Also, the summation point of the error
feedback (V1 plate in your design) needs to have a pretty stable
impedance to maintain the correct level of error correction.

You can't wish for better than a triode with current FB.....

Lots of
discussion of the Hawksford scheme on the "diyAudio" solid state
forum: ( a search on "Hawksford" will turn up quite a few discussion
threads and links to the original Hawksford articles)

http://www.diyaudio.com/forums/showt...113&highlight=

and one thread on the "diyAudio" tube forum too, where it is used
with a circlotron type output stage:

http://www.diyaudio.com/forums/showt...001&highlight=

Don

I see someone has drawn up a tube version of error correction at

http://www.diyaudio.com/forums/attac...&postid=404499

where the input signal is applied to the output tube grid, but
signal off the OPT is nulled against the input V and delta Dv applied to a
tube where it is amplified and phase reversed and added also to the input grid of the output tube.

The stage shown isn't a 100% circlotron, but is like what I proposed days ago using a pair of floating
B+ supplies, and with a grounded UL tranny to achieve local cathode FB in the output stage.
It is a partial circlotron.
It could be mabe into the full circlotron by moving the cathode connections of the output tubes
out to the anode connections of the other tube.


The error correction is an add on.
R1 picks up a signal from the OPT at a point S which has an opposite signal phase to what is applied
via C1, R3, and R2.

The signal null point is at R2-R1.
Thd from the OPT, say let it be Dv, will appear at the g1/V10, which amplifies the Dv to make it into
-AxDv at V10 anode, where it is applied to the output tube g1/V1.
But unfortunately, the guy who so roughly scribbled up this schematic rather than waste 1,000 words
didn't see that -ADv applied to the output tube grid would make +ADv at the V1 anode, therby
increasing the THD, in other words, he is using positive feedback with regard to THD,
and any gains from the use of the cathode FB of the circuit may be lost.

The discussions continued amoungst the learned at the group for some time before the error
I noticed in seconds was picked up and a corrected schematic posted at

http://www.diyaudio.com/forums/attac... postid=405328

They don't appear to be doing much like what I proposed yesterday, although I see their
basic idea about allowing an anti Dv voltage to appear at the same input port
of the amp as the input signal of the amp.

In fact the use of two floating B+PS, and CFB from the OPT with 43% UL taps will
give an output stage needing about 120vrms drive to each output tube grid so the drive amp has to be a work of
art
to keep its thd low.
At least two stages would be needed to get the input voltage down to 1 or 2v for full PO.
Or something like what McIntosh or EAR use as driver amps.
And then you still need the error correction LTP as shown in the schemos above.

In my scheme, there is no need for extra LTP, floating PS, high drive voltages,
or multi stage drivers.
The same number of tubes as exist in any amp can be used.

And I remove one stage out of the signal path, because the open loop gain needed
is the same as the gain needed with error correction.

So I see that my scheme has the potential to operate better than plain old NFB.

Patrick Turner.

Patrick Turner wrote in message ...
smoking-amp wrote:

For an amplifier with significant gain,
the setting of the error correction feedback gain will be more
critical, so this is usually just added to a normal feedback design
with unity gain output stage. Also, the summation point of the error
feedback (V1 plate in your design) needs to have a pretty stable
impedance to maintain the correct level of error correction.

You can't wish for better than a triode with current FB.....


They don't appear to be doing much like what I proposed yesterday, although I see their
basic idea about allowing an anti Dv voltage to appear at the same input port
of the amp as the input signal of the amp.

In my scheme, there is no need for extra LTP, floating PS, high
drive voltages,
or multi stage drivers.
The same number of tubes as exist in any amp can be used.

And I remove one stage out of the signal path, because the open loop gain needed
is the same as the gain needed with error correction.

So I see that my scheme has the potential to operate better than plain old NFB.

Patrick Turner.


Hi Patrick,

I agree your design has a nice economy of parts and function by
including the entire amp in the error correction loop. I am not trying
to be critical of your design, but merely wish to point out the areas
where special attention to design will probably be required. Hawksford
in his original articles pointed out the difficulties of this
approach.

The error correction feedback gain has to be accurately set,
it is not self adjusting like normal NFB. With the amplifiers gain IN
the error loop, this setting becomes a lot more critical since any
error in setting gets multiplied by the gain. This also makes the
summation linearity all the more critical too. Since V1 plate in your
design is constrained to -1 gain, rather than -Mu, the internal V1
plate feedback will be mostly nulled out as far as linearity corr. is
concerned, so you are depending on the cathode degeneration resistor
for linearity. A triode by itself has a current dependant output
impedance, as I'm sure you are aware, so the cathode degeneration
resistor is very much needed to swamp this effect out too. A pentode
for V1 might have some advantage due to its already high output
impedance, which could then be shunted by just an ordinary plate load
resistor (instead of current source load) to set a constant impedance.
Significant cathode degeneration would still be required however, to
linearize transconductance, the usual 3/2 power gm just wouldn't do.
Just an idea. Otherwise, can always fall back to a full LTP in place
of V1 if problems persist.

The circlotron like design on diyAudio, plays it safe with the
usual Hawksford style design added on to a conventional global NFB
amplifier output stage only. The penalty is of course more tube
elements, but certain advantages do compensate. The output stage is
generally the worst offender for distortion, so using the corrector on
it still pays off well. Since most of the distortion in the amplifier
is thus nulled out, the global NFB loop is very effective at removing
the last residual dist. without generating higher harmonics or
requiring excessive open loop gain (which will help stability too).
In my opinion, the error correction idea is a godsend to the
usual ills of global NFB amplifiers, its about time tube designs
started using it. It's just the right fix for low or no global
feedback designs too, with no stability problems if implemented on the
primary side only of the output xfmr.
The low unity gain in the (circlotron like design) output
correction loop means that error corr. gain setting is very stable and
less sensitive to impedance variation at the summation point. Also,
the error correction loop does not in this case enclose the xfmr., so
phase/ bandwidth problems are avoided in its operation. The gobal NFB
loop is left to deal with the xfmr's secondary resistive losses, but
the low output impedance of the partial cathode follower enhances the
bandwidth and stability of this loop nicely.
Of course, the circlotron like output stage with its low gain
does impose a high drive requirement for its driver as you pointed
out. The UL tapped scheme drops the drive requirement to 28% of output
signal (plus grid to k signal) for a 43% UL tap xfmr. So is more
manageable than the 50% drive requirement of a full circlotron or
McIntosh design. A later schematic drops the dual floating B+ supply
requirement by using a center tapped inductor for cathode DC supply as
well.

Don

smoking-amp wrote:

Patrick Turner wrote in message ...
smoking-amp wrote:

For an amplifier with significant gain,
the setting of the error correction feedback gain will be more
critical, so this is usually just added to a normal feedback design
with unity gain output stage. Also, the summation point of the error
feedback (V1 plate in your design) needs to have a pretty stable
impedance to maintain the correct level of error correction.

You can't wish for better than a triode with current FB.....


They don't appear to be doing much like what I proposed yesterday, although I see their
basic idea about allowing an anti Dv voltage to appear at the same input port
of the amp as the input signal of the amp.

In my scheme, there is no need for extra LTP, floating PS, high
drive voltages,
or multi stage drivers.
The same number of tubes as exist in any amp can be used.

And I remove one stage out of the signal path, because the open loop gain needed
is the same as the gain needed with error correction.

So I see that my scheme has the potential to operate better than plain old NFB.

Patrick Turner.

Hi Patrick,

I agree your design has a nice economy of parts and function by
including the entire amp in the error correction loop.

Indeed.

Williamson said that one single loop of NFB enclosing as much of
the gain structure as possible was the most effective way of applying NFB.
McIntosh had the idea of lots of local NFB in the output stage,
then also the usual global NFB a la Williamson.
Professor Ed Cherry invented "nested feedback loops" to be able to supply
an enormpous total amount of NFB in SS circuits,
and many of us since Williamson have been playing around with NFB ever since.

It seems possible to have error correction at the driver /output tube interface,
then perhaps NFB right around the lot.



I am not trying
to be critical of your design, but merely wish to point out the areas
where special attention to design will probably be required. Hawksford
in his original articles pointed out the difficulties of this
approach.

The error correction feedback gain has to be accurately set,
it is not self adjusting like normal NFB.

This needs further verification. The amount of NFB applied
globally or in an output stage varies with amp gain which varies with load.
But NFB in early stages does not vary.
But yes, error correction is fixed, since the error correction stages are not
subject to gain variations caused by load variations.






With the amplifiers gain IN the error loop, this setting becomes a lot more critical since any
error in setting gets multiplied by the gain. This also makes the
summation linearity all the more critical too. Since V1 plate in your
design is constrained to -1 gain, rather than -Mu, the internal V1
plate feedback will be mostly nulled out as far as linearity corr. is
concerned, so you are depending on the cathode degeneration resistor
for linearity.

If you consider the V1 triode in my design as a voltage gene of U x Vg1
feeding through a plate resistance of approximately 5k, with the 2k
cathode R acting to increase the input resistance seen at the plate to 47k,
and a fairly linear effective Ra at that.

If the distortion produced by V1 is less than 0.1% of the 1.7v signal sent to
g1/v3, then the total thd input is 0.05% of the signal betweeen g1/V2 and gi/V3,
and the the output won't have any more than 0.05% due to V1's contribution.

And what if we were to have arranged V1 to not supply any of the drive voltage to the LTP driver stage?
Then only the extracted THD from the output would appear at g1/V3.
One could say some included amount of distortion from the non-linearity of the
V1's Gm characteristic would appear as well, but methinks its a trifling quantity
compared to the Dv produced from the output stage.




A triode by itself has a current dependant output
impedance, as I'm sure you are aware, so the cathode degeneration
resistor is very much needed to swamp this effect out too. A pentode
for V1 might have some advantage due to its already high output
impedance, which could then be shunted by just an ordinary plate load
resistor (instead of current source load) to set a constant impedance.

A pentode of the right choice would have a higher Gm than the triode,
and thus need a higher value of Rk to allow the same 1.7v input like I have in my schematic.
The use of a plate DC supply resistor immediately reduces the R1 value of the R1 - R2
relation for the nulling network, and a typical value would be 47k, which happens to be the same as
the triode's effective Ra, and so the fraction of the Dv will be the same, and
no increase of the amount of Dv applied to g1/V3 is obtained, so
there isn't a benefit using a pentode, since although it would have much more local current FB,
its non-linearity may be worse to begin with.
The CCS at the anode makes the R1 of the R divider so much greater than R2, in this case its 22k,
that the R1 / ( R1 + R2 ) becomes nearly 1.0, so nearly all the Dv of the output
is fed to g1/V3, and the error correction then becomes just as effective, but so much easier to implement
than if we had fed back all of the 14.4v output voltage at g1/V3.

If we had done this, the LTP has to cope with the extra input voltage
of 17.8 vrms at g1/V2, and 14.4vrms at g1/V2,
and the extra distortions produced in the process would exceed what
my V1 produces, surely, and the error correction is surely the lesser of two evils.


Significant cathode degeneration would still be required however, to
linearize transconductance, the usual 3/2 power gm just wouldn't do.
Just an idea. Otherwise, can always fall back to a full LTP in place
of V1 if problems persist.

One could use a Mu follower instead of V1 to give a low impedance voltage output,
and then the R1 of the R1-R2 network can be altered in value for the wanted null,
or in my case, adjusted to favour the -v signal produced by V1 output
to give equal +/- input signal voltages to the LTP.



The circlotron like design on diyAudio, plays it safe with the
usual Hawksford style design added on to a conventional global NFB
amplifier output stage only. The penalty is of course more tube
elements, but certain advantages do compensate.

That schematic I refered folks to yesterday has some idiosyncratic oddities which
may prevent it working as well as the theory suggests, because the
pick up points for error signals, and the cross-couplings from the
error amps all are very prone to phase reversals at HF, and I fear the whole darn thing
would make a nice HF oscillator.
The circuit is only a partial circlotron, but still the required drive voltage is enormous
even if the UL taps used in the sample are at 43%.
The huge drive voltage defeats the purpose, since error correction has to be applied
to the input voltages.
A way needs to be found to linearize a normal high gain plate loaded pentode output stage,
so that after error correction, the Vg-g is still only 20 vrms for full output,
but the local error correction has reduced the Dn down say 5 times, and Ra 10 times.

Its easy to procuce a low thd 20vrms a-a from a triode LTP.

I am giving it some thought.


The output stage is
generally the worst offender for distortion, so using the corrector on
it still pays off well. Since most of the distortion in the amplifier
is thus nulled out, the global NFB loop is very effective at removing
the last residual dist. without generating higher harmonics or
requiring excessive open loop gain (which will help stability too).
In my opinion, the error correction idea is a godsend to the
usual ills of global NFB amplifiers, its about time tube designs
started using it. It's just the right fix for low or no global
feedback designs too, with no stability problems if implemented on the
primary side only of the output xfmr.

Well they'll all be able to say what they ain't using.
The SET crowd hates NFB.

But will they take 20 years to understand error correction?....



The low unity gain in the (circlotron like design) output
correction loop means that error corr. gain setting is very stable and
less sensitive to impedance variation at the summation point.

The type of circuit does not use the gain of the output tubes for the correction.
It uses the gain of the tubes specifically used to ampify Dv which is inserted
in the input path, so correction is independant of load effects.

Also,
the error correction loop does not in this case enclose the xfmr., so
phase/ bandwidth problems are avoided in its operation.

No, I think this isn'r so.
The OPT with its leakage inductance / stray C effects at HF is very much involved here,
and the circuit as shown looks like a fine oscillator at HF.
The method of applying error correction is indirect, and the only best way
is to make the output voltage errors be applied back to the tube making them,
rather than by mutual cross coupling.

The gobal NFB
loop is left to deal with the xfmr's secondary resistive losses, but
the low output impedance of the partial cathode follower enhances the
bandwidth and stability of this loop nicely.

Well yes, the OPT secondary is a very low impedance drive
which normally allows a signal to be applied to an input tube's cathode
so the differential operation of the input tube is thus obtained.
It also is a fast circuit; there are fewer HF losses and phase shifts,
and I have found it the easiest FB path to stabilise compared to using an LTP
at the front end, with FB taken to a grid circuit.


Of course, the circlotron like output stage with its low gain
does impose a high drive requirement for its driver as you pointed
out. The UL tapped scheme drops the drive requirement to 28% of output
signal (plus grid to k signal) for a 43% UL tap xfmr. So is more
manageable than the 50% drive requirement of a full circlotron or
McIntosh design. A later schematic drops the dual floating B+ supply
requirement by using a center tapped inductor for cathode DC supply as
well.

Let us suppose the Va-k of each tube is 250vrms when operating into a typical load.
With a pure circlotron, each tube has +/-125vrms at its anode and -/+125v at its cathode,
and if operating in pentode mode, open gain is say 20, so
12.5 vrms is needed between g1 and k, so input voltage to each g1
is 137.5vrms.
If you have partial circlotron with cathodes taken to 43% of the P windings,
then the tubes still have 250vrms across them, with +/-193vrms at the anodes,
and -/+57vrms at the cathodes.
If it runs in pentode mode, the 12.5 g1-k voltage is still required,
so 69.5 vrms is required to the g1 of each tube.

This would be with EL34/KT88 type of tubes.

The partial operation of the output tubes in circlotron
does give far better Ra reduction and thd reduction than any normal UL
way of operation, because all the FB is applied into the grid circuit, rather than to the screen
circuit.
If the screens are taken to a suitable value of B+ in all the forgoing cases,
then NFB is applied to the screen circuits and the tube gain drifts towards triode like gains,
and the Vg1-k voltage required will increase from the 12.5v neded in the pentode mode
to up to about 25v in the pure circlotron method, so 150vrms is needed at each g1,
and in the case of the 43% cathode tap option, g1 input would be around 80vrms.

Peter Walker and others, Hafler and Keroes thought about all this all this about 60 years ago,
after the shooting stopped in WW2, and hence the Quad II amp design,
which is an elegant blend of g1 and g2 applied NFB.

All other methods require floating PS, electro cap coupling from the OPT
back to cathodes plus a centre tapped choke, etc, all of which Mr Bean Kounter
severely disapproved of in 1946.
And the electros made at that time were bloomin awful. Copper and GO laminations were
expensive. Multiple floating PS were out of the question and uncompetitive
until silicon rectifiers became reliable enough to use with voltage doublers
in about 1960.

So Quad and McIntosh put their efforts into the OPT,
before wasting it elsewhere.

(Mr Kounter's sons went on to the dizzy heights of applied mediocrity
in so many areas that nothing seemed to escape them,
and finally we had chips which go phut real easy after making a lot of folks' ears sore. )

Nobody seemed to place the effort into error correction, and I wonder if Walker
and his pals ever gave it any thought.
They probably figured they'd overcome the problems with moderate NFB,
some 20 dB in all, and that was enough.
Everyone could understand and service the gear, and no special stablity
measures were needed, and the sun would shine on them for the next 50 years.

Patrick Turner.

Patrick Turner wrote in message ...


smoking-amp: The error correction feedback gain has to be
accurately set,
it is not self adjusting like normal NFB.

This needs further verification. The amount of NFB applied
globally or in an output stage varies with amp gain which varies with load.
But NFB in early stages does not vary.
But yes, error correction is fixed, since the error correction stages are not
subject to gain variations caused by load variations.


My point is that its easy to get rid of the bulk of the distortion
using error correction, say 99%, but the accuracy of the correction
(which depends on the linearity of the error subtraction and accuracy
of gain setting) loop gets very touchy when trying to get rid of
99.99%. Global NFB, on the other hand, causes HF harmonics if used to
get rid of high initial distortion, but does do well with reducing 1%
distortion down to .01% dist. Even if you manage to get the error
correction adjusted and working initially at .01% distortion, any tube
or component aging will whipe this out. Global NFB still keeps working
well after component aging, so is a better choice to do the last few
decimals. The two go together well.

The circlotron like design on diyAudio, plays it safe with the
usual Hawksford style design added on to a conventional global NFB
amplifier output stage only. The penalty is of course more tube
elements, but certain advantages do compensate.

That schematic I refered folks to yesterday has some idiosyncratic oddities which
may prevent it working as well as the theory suggests, because the
pick up points for error signals, and the cross-couplings from the
error amps all are very prone to phase reversals at HF, and I fear the whole darn thing
would make a nice HF oscillator.

Well, the circuit hasn't been built yet, but I don't agree with your
assessment on its stability. The two error correction feedback points
are both on the cathode driven section of the primary for whichever
tube is in conduction (class AB, class A fine too, there is no cross
coupling), so no coupling thru leakage inductances is required (this
WOULD be a problem for a non-circlotron like design however).
Distributed capacitance would cause some phase shift like in any
circuit, but is driven by a low impedance cathode follower here and
this should be easy to handle by appropriate rolloff of the correction
loop frequency response, like any feedback loop. On the other hand,
your circuit is taking its feedback from the secondary and is subject
to both these problems. Which just means that the error feedback must
be rolled off at a lower freq. Being able to extend the error corr.
freq. resp. further up by using the primary side connection enables it
to handle the class AB crossover distortion better (typically high
harmonics), which was very the reason Hawksford came up with the
design in the first place. But, building the circuit will be a better
judge of actual results.

The circuit is only a partial circlotron, but still the required drive voltage is enormous
even if the UL taps used in the sample are at 43%.
The huge drive voltage defeats the purpose, since error correction has to be applied
to the input voltages.
Let us suppose the Va-k of each tube is 250vrms when operating into a typical load.
With a pure circlotron, each tube has +/-125vrms at its anode and -/+125v at its cathode,
and if operating in pentode mode, open gain is say 20, so
12.5 vrms is needed between g1 and k, so input voltage to each g1
is 137.5vrms.
If you have partial circlotron with cathodes taken to 43% of the P windings,
then the tubes still have 250vrms across them, with +/-193vrms at the anodes,
and -/+57vrms at the cathodes.
If it runs in pentode mode, the 12.5 g1-k voltage is still required,
so 69.5 vrms is required to the g1 of each tube.

This would be with EL34/KT88 type of tubes.

Getting 70 V drive into a grid only (no actual load) shouldn't be
that much of a problem, just have to use the same B+ for the driver as
the output tube uses. EL84 or 5687 driver maybe. Not nearly as bad as
a full circlotron or McIntosh design. The 43% UL taps were only chosen
since cheap off-the-shelf xfmrs are readily available (Hammond 1650T),
a 10-20% tap would be nicer of course.

All other methods require floating PS, electro cap coupling from the OPT
back to cathodes plus a centre tapped choke, etc, all of which Mr Bean Kounter
severely disapproved of in 1946.

My low cost, low distributed capacitance center tapped choke:
http://www.mpja.com/productview.asp?product=4567+TR
I use two in series connection. The multi-bobbin windings make for
very low distributed capacitance. (Note, the shield straps are on
wrong, probably why they are surplus) SMPS Electrolytics are cheap and
plentiful surplus these days too.

Oh, on the linearized pentode need ..., this was solved way back
in the 1930s according to S. Okamura in the book: "History of Electron
Tubes" and then forgotten with the invention of feedback theory
shortly thereafter. We call them current mirrors with gain today (not
the same as a typical current source), but one never sees them in
vacuum tube form these days. (The IC guys use them all the time in
chips. And usually they are thought of as invented in solid state form
in the 60s, but tubes actually did it first) Its a mystery to me why
people make solid state ones to use in their tube amps as auxilliary
circuits, but never consider a tube version. I use them frequently,
they work wonderfully in tube form, linear as can be, and can be
adapted to linearize triodes with real loads too. Unfortunately, to
make good use of them nowadays, one would really need some custom
designed VT diodes with the correct internal geometry/dimensions to
get useful current gain ratios.

Don

smoking-amp wrote:

Patrick Turner wrote in message ...

smoking-amp: The error correction feedback gain has to be
accurately set,
it is not self adjusting like normal NFB.

This needs further verification. The amount of NFB applied
globally or in an output stage varies with amp gain which varies with load.
But NFB in early stages does not vary.
But yes, error correction is fixed, since the error correction stages are not
subject to gain variations caused by load variations.


My point is that its easy to get rid of the bulk of the distortion
using error correction, say 99%, but the accuracy of the correction
(which depends on the linearity of the error subtraction and accuracy
of gain setting) loop gets very touchy when trying to get rid of
99.99%. Global NFB, on the other hand, causes HF harmonics if used to
get rid of high initial distortion, but does do well with reducing 1%
distortion down to .01% dist. Even if you manage to get the error
correction adjusted and working initially at .01% distortion, any tube
or component aging will whipe this out. Global NFB still keeps working
well after component aging, so is a better choice to do the last few
decimals. The two go together well.

I have never tried to get the distortions of my tube amps down to 0.01%
at 1 dB below clipping, because I'd need to apply
40 dB of NFB to an amp with 1% open loop Dn.
Usually, the limit is about 30 dB before intractable instability at LF and HF appear
even with pure R loads, so 16 dB is all I ever use, with thd at about 0.2%
at 1 dB below clip.
which means the amp has a margin of 12 dB.
At 1/10 of full power, Vo is 1/3 of full Vo, so
thd will be around 0.05%, and all is well, as thd falls as Po is reduced,
especially when considering all tube amps have class A operation for the first
few watts.

So trying to use error correction to reduce thd to absurdly low levels isn't
a priority for me, and is fraught with instability problems as the amount
of error correction increases, just like with NFB application..



The SS amps I have made with no OPT have a shirtload of global NFB,
and 0.005% at 299 watts is a doddle to achieve, with about a total of 80 dB of
FB, both in the source follower connection and global loop.

In the cases of the error correction I posted day before yesterday,
and about 5 minutyes ago, the error correction
would tend to counter parts drift and ageing.






The circlotron like design on diyAudio, plays it safe with the
usual Hawksford style design added on to a conventional global NFB
amplifier output stage only. The penalty is of course more tube
elements, but certain advantages do compensate.

That schematic I refered folks to yesterday has some idiosyncratic oddities which
may prevent it working as well as the theory suggests, because the
pick up points for error signals, and the cross-couplings from the
error amps all are very prone to phase reversals at HF, and I fear the whole darn thing
would make a nice HF oscillator.

Well, the circuit hasn't been built yet, but I don't agree with your
assessment on its stability.

I have done a few attempts at such queer circuits as error correction types are,
and I was rewarded with instabilities.
Build the circuits, and you will find out.

The two error correction feedback points
are both on the cathode driven section of the primary for whichever
tube is in conduction (class AB, class A fine too, there is no cross
coupling), so no coupling thru leakage inductances is required (this
WOULD be a problem for a non-circlotron like design however).
Distributed capacitance would cause some phase shift like in any
circuit, but is driven by a low impedance cathode follower here and
this should be easy to handle by appropriate rolloff of the correction
loop frequency response, like any feedback loop. On the other hand,
your circuit is taking its feedback from the secondary and is subject
to both these problems. Which just means that the error feedback must
be rolled off at a lower freq.

Some trimming and R&C networks mat be needed, just like in any NFB circuit.
But in the basic circuits I propose, the amount of error correction is no greater than a mild
usage of NFB, and so global NFB abd global error correction will work easily
if the OPT is a relatively low leakage inductance low shunt capacitance type.

To bake a cake, the quality of eggs should be right.

Being able to extend the error corr.
freq. resp. further up by using the primary side connection enables it
to handle the class AB crossover distortion better (typically high
harmonics), which was very the reason Hawksford came up with the
design in the first place. But, building the circuit will be a better
judge of actual results.

If you have a 10 kHz square wave applied to an average feedback tube amp
even when its still in class A mode at low level, the output tube
anode wave forms are often full of over shoot and ring
even thought the output waves are relatively a good looking
wave form, with the desired amount of curvature to the uprights of the
square wave.

I have tried local plate to cathode FB to try to tame the output plate behaviour of tube amps,
but what works best is RC damping networks across each 1/2 primary.
Or wherever they are found to work best.

Class AB operation is always going to cause a kink in the gain transfer character
between class A and AB with bjts, mosfets, pentodes, and to a lesser extent
with UL or triodes, which are the least likely to produce high amounts of odd order
crossover products.
Regardless of the OPT, the kink will cause the harmonics, simply
because of the gain change with the sharp cut off behaviour
with many devices.


The circuit is only a partial circlotron, but still the required drive voltage is enormous
even if the UL taps used in the sample are at 43%.
The huge drive voltage defeats the purpose, since error correction has to be applied
to the input voltages.
Let us suppose the Va-k of each tube is 250vrms when operating into a typical load.
With a pure circlotron, each tube has +/-125vrms at its anode and -/+125v at its cathode,
and if operating in pentode mode, open gain is say 20, so
12.5 vrms is needed between g1 and k, so input voltage to each g1
is 137.5vrms.
If you have partial circlotron with cathodes taken to 43% of the P windings,
then the tubes still have 250vrms across them, with +/-193vrms at the anodes,
and -/+57vrms at the cathodes.
If it runs in pentode mode, the 12.5 g1-k voltage is still required,
so 69.5 vrms is required to the g1 of each tube.

This would be with EL34/KT88 type of tubes.

Getting 70 V drive into a grid only (no actual load) shouldn't be
that much of a problem, just have to use the same B+ for the driver as
the output tube uses. EL84 or 5687 driver maybe. Not nearly as bad as
a full circlotron or McIntosh design.

I agree, and I use a centre tapped choke plus RLs for the LTP load,
to increase the load impedance on the triodes, and the undistorted swing.

The 43% UL taps were only chosen
since cheap off-the-shelf xfmrs are readily available (Hammond 1650T),
a 10-20% tap would be nicer of course.

All other methods require floating PS, electro cap coupling from the OPT
back to cathodes plus a centre tapped choke, etc, all of which Mr Bean Kounter
severely disapproved of in 1946.

My low cost, low distributed capacitance center tapped choke:
http://www.mpja.com/productview.asp?product=4567+TR
I use two in series connection. The multi-bobbin windings make for
very low distributed capacitance. (Note, the shield straps are on
wrong, probably why they are surplus) SMPS Electrolytics are cheap and
plentiful surplus these days too.

Oh, on the linearized pentode need ..., this was solved way back
in the 1930s according to S. Okamura in the book: "History of Electron
Tubes" and then forgotten with the invention of feedback theory
shortly thereafter. We call them current mirrors with gain today (not
the same as a typical current source), but one never sees them in
vacuum tube form these days. (The IC guys use them all the time in
chips. And usually they are thought of as invented in solid state form
in the 60s, but tubes actually did it first) Its a mystery to me why
people make solid state ones to use in their tube amps as auxilliary
circuits, but never consider a tube version.

I'd be interested in a schematic showing what you mean for a tube version
of a current mirror, and that of what you mean by a transistor current mirror.


I use CCS in many of my signal stages of my amps.
A single MJE340 can be so easily arranged as a CCS with a finite value
of 30 Mohms, but getting Z that high with a tube is a pita,
and totally uneccessary, imho.

Their use reduces the power produced in the R used to get DC to a tube,
and with less unecessary current change, a triode is always going to be
more linear.
I also like the mu follower stage for the same reason, but then Ro is low, and Rin is high.

I'd just as soon use a CT choke + RLs for the load of an LTP,
as usually I have the grib bias R of the output stage to power and a current
mirror of any type won't remove the effect of a resistive load.
And the R load is needed to balance the circuit with a CC tail.
See the circuit at
http://www.turneraudio.com.au/websch...ma550w335h.gif


I use them frequently,
they work wonderfully in tube form, linear as can be, and can be
adapted to linearize triodes with real loads too. Unfortunately, to
make good use of them nowadays, one would really need some custom
designed VT diodes with the correct internal geometry/dimensions to
get useful current gain ratios.

Really. Why diodes?

Post a basic schematic.

Patrick Turner.



Don

I posted the Vacuum Tube Current Mirror schematics on ABSE. I will
post the same on the diyAudio tube thread in case anyone has trouble
accessing ABSE.
I use some assorted fixtures with a bunch of diode tubes connected in
series with selectable series tap points. 6JU8's and 9006's are
suitable diode tubes.

Just looked at your latest ABSE post of an Error Correcting class A
amplifier. Interesting, similar to what I had envisioned would be
required for a conventional (non-circlotron) P-P output transformer,
using separate LTPs for correcting each output tube. If you leave out
the cross coupled plates, might it work for class AB? Probably will
need to provide a trimpot on each LTP for fine adjusting the error
loop gains since the two output tubes will have slightly different
gains, same required for all the P-P error corr. designs actually.

I am not trying to reach .01% dist. particularly either, just was
throwing some numbers around to illustrate.

Oh, on the cross-coupling leading to oscillation point, I maybe was
mis-reading your meaning earlier. There is hopefully no or little
cross-coupling in the circlotron like xfmr itself due to the cathode
drives being fully across the sampling points, but the circuit itself
does have cross coupling in the LTP corrector by design. This does
make it look like a conventional oscillator. The gain around the loop
is just shy of unity, so should be stable when adjusted correctly
(gain trimpots for each output tube). If adjusted incorrectly it can
indeed become an oscillator by providing positive feedback around the
loop. This problem is generic to the error corr. scheme in any form I
think. If the subtractor isn't exactly spot on it either leaves some +
or - phase signal in its output, which then acts as negative or
positive feedback in a loop with near unity gain. It is quite similar
to the bootstrapped load resistor problem. At exactly unity feedback
you get a current source, below unity you get a high resistance, and
above unity you get positive feedback. In the error corr. case, to
theoretically null distortion to .0001% say, one would adjust the
error loop gain to .999999, perilously close to oscillation. So any
drift of loop gain could likely lead to instability. Thus my point
about using error corr. to get the bulk of dist. cancelled, but not
the last decimals. Class AB output makes oscillation less likely since
only one tube is on most of the time, and the tube gains are lower
during crossover. My goal in looking at the error correction idea was
to fix class AB crossover distortion just like the original Hawksford
scheme was used as a crossover dist. fix in SS amplifiers.

Don

Here is a reference. Not sure if anyone has already posted this one.

http://www.quadesl.org/home/walalb.html

Cheers, John Stewart

Patrick Turner wrote:

I have just posted a circuit at abse and abpr fo those
who like a little brain teaser circuit application of the idea of error
correction in lieu
of applied NFB.

Consider the typically arranged LTP driver amp and 6550 PP UL output
stage
which is shown in the jpg.

The normal way to apply feedback if we *did not* want to include another
gain stage
so that we would reduce the thd by say 4 times would be to
apply the output voltage of 14.4v to to the g1/V3.
The input to V2 would then have to be 14.4v + 3.4 vg-g of the LTP, so
17.8 av all up.
This is a lot of drive voltage, and we'd have trouble keeping the
linearity of a preamp
to less than say 0.05% thd at full tilt.
However the FB would reduce thd in the same way
sensitivity is reduced from 3.4 v at the input to 17.8v which is a gain
reduction
of 17.8/3.4 = 5.23 times = 14.4 dB.

In the case of the NFB method, even with the FB connected, +D volts, the
distortion voltage
is able to be measured at the output.
But the output applies this to the feedback input port at g1/V3, so +Dv
gets amplified
by the open loop gain of the amp to make a signal of -4.24Dv at the
output.
But how can this be if we already know only Dv is present at the output?

But already a voltage of +5.24Dv will be trying to appear at the output
if we didn't have any NFB, and the -4.24v NFB correction voltage applied
subtracts from
the +5.24Dv to leave just Dv.
This further supports the idea Dn is reduced A+1 times with series
voltage NFB,
where Dn = open loop distortion, A = open loop gain, and there is little
phase shift
in frequencies of concern.

The price we had to pay for the NFB correction was a huge loss of
ampifier sensitivity.

If there was a way to apply only the Distortion voltage sample at g1/V3,
then the input voltage to the
g1/V2 would stay similar, or be able to be less than what is used when
no NFB is applied.

So we need to find some way of nulling out the signal content of the NFB
signal
and sifting out the Dv only.

Now let's look at my schematic again for the typical driver LTP and UL
output stage.

The input stage isn't part of the drive amp, but instead it is an
auxilliary to it.

The V1 is a paralleled 6SN7 which has a CCS DC anode supply, not
absolutely
necessary, but elegantly better for this scheme to work well.

The input voltage to the amp at g1/V2 is +1.7av, and this is also
applied to the
g1/V1 input stage.

There is a 22k resistor returned from the +14.4v to the
anode of V1 but separated by a DC blocking cap of say 0.47 uF or
greater.
The output of V1 is fed to g1V3, and all has been arranged so that
-1.7av
is generated at the output of V1 so V1 has a phase inverter function as
well as a "signal nuller"
and distortion re-director.

Normally, the anode of V1 would swing about -30 av with +1.7 av input.
But the 22k R from the output stops most but not quite all of this
swing.
The output resistance of the V1 at its anode is 47k, due to the use of
unbypassed Rk total
of about 2k, ( 1k plus about 1k, with provision for baising g1V1 via
220k).
In fact by adjustment of the lower Rk we could better balance the +/-
1.7 av applied to the
LTP.
The 22k supplies a signal current to the V1 anode which opposes the
anode normal flow as a result
of the g1 voltage effect, and in fact the tube sees an anode load of 1.7
/ 0.73 mA = 2.33 kOhms.
The cathode R of 2k make the total tube load = 4.33k, and the 2k is
used to make local current FB for V1, and since output voltage is so
low,
the thd will not be great; in fct V1 has the usual large amount of local
NFB as
one has in a concertina phase inverter, which it is, in this case..

Now as I said, the Ro of V1 at the anode is 47k; that's the dynamic
generator impedance,
which is calculated as Ra + [(U+1) x Rk], and the CCS doesn't count in
because its impedance in parallel
with Ro is in the megohm region.

So the 22k and 47k act as a R divider, and in this case the circuit
tries to
send the voltage in opposite polarities at each end of the divider..
Let us suppose that with set up as shown that we had Dv appear as thd at
the output.
Then (47 / 69) x Dv, or +0.68 Dv will appear at the divider junction,
which is the anode of V1.
This is applied to the g1/V3 input, as well as the 1.7 va signal, and
the
open loop gain of 4.24 amplifies the +0.68Dv to appear as -2.88Dv at the
output.
And this subtracts from whatever open loop Dv would have appeared
without this error
correction, so the original amount of D would have been +3.88Dv from
which the applied -2.88 subtracts,
leaving +Dv as we observe, with error connection applied.
So, we have removed the necessity for the additional usual gain stage in
the signal path,
and been able to reduce thd from 3.88 to 1, and this is about an 11 dB
thd reduction,
about equal to the use of the NFB.
And the amp is sensitive to only 1.7 av instead of 3.4v, open loop.

You mathematicians can draw up the formula for the amount of
error correction in Db with regards to the divider resistances, V1 Rk,
U, Ra, Gm, and amplifier open loop
gain between Vg-g of V2/3 to the output, but I ain't in any rush, since
whatever
one comes upo with will be more difficult to weild tham the normal
basic simple feedback equations.

The concept is also more complex, but as I show, its appliaction is dead
simple, and uses
no more tubes than an existing amp would.

It is possible to apply *more* correction, by using a higher U, higher
transconductance triode at V1,
such as a 12AT7, and get Ro to be a lot higher, so the
fraction of the distortion voltage fed back into the amp is increased,
whilst the linearity of V1 is increased. The linearity of V1 is
important, since we don't want
its thd to be fed into the amp.

Its also poosible to have 12AT7 for the LTP, and also possible to
arrange that the LTP
gets a signal input to only g2/V2, and have only the Dv applied to
g1/V3.

It works OK to reduce Ro at the output.
Suppose the load is reduced from 6 to 3 ohms, then the output voltage
would fall,
and the voltage at the divider nuller 22k and 47k would increase from
the
-1.7v, and the Vg-g applied to V2/3 will increase, thus tending to
boost the output v.

Similarly, if load was 20 ohms instead of 6ohms, the output voltage
would rise,
reducing the voltage level at V1 anode, and thus across Vg-g of V2/3,
so output voltage tends to get reduced from what it otherwise would be.

If you have already built a few things, and have time, try this circuit
on the
experimental amp you may have.

Regards to all,

Patrick Turner.

smoking-amp wrote:

I posted the Vacuum Tube Current Mirror schematics on ABSE. I will
post the same on the diyAudio tube thread in case anyone has trouble
accessing ABSE.
I use some assorted fixtures with a bunch of diode tubes connected in
series with selectable series tap points. 6JU8's and 9006's are
suitable diode tubes.

Just looked at your latest ABSE post of an Error Correcting class A
amplifier. Interesting, similar to what I had envisioned would be
required for a conventional (non-circlotron) P-P output transformer,
using separate LTPs for correcting each output tube. If you leave out
the cross coupled plates, might it work for class AB?

I am sure it will work for class AB, since the errors are voltage errors
at either end of the OPT.
The crosscoupled LTP plates make the whole LTP dual into a pair of
paralleled
LTPs, and thus with symetrical signals and better balance, imho.

Probably will
need to provide a trimpot on each LTP for fine adjusting the error
loop gains since the two output tubes will have slightly different
gains, same required for all the P-P error corr. designs actually.

Well yes, but the OPT keeps the errors very similar at each side of the PP
circuit,
except they are opposite phase.

Most of the correctiion signal is signal voltage compensation to keep
the effective Ro at a low value. The distortion signal is quite low.



I am not trying to reach .01% dist. particularly either, just was
throwing some numbers around to illustrate.

To get 0.01% thd, the error correction circuit becomes more critical
to set up and stabilise; it has similar constraints as NFB.



Oh, on the cross-coupling leading to oscillation point, I maybe was
mis-reading your meaning earlier. There is hopefully no or little
cross-coupling in the circlotron like xfmr itself due to the cathode
drives being fully across the sampling points, but the circuit itself
does have cross coupling in the LTP corrector by design. This does
make it look like a conventional oscillator. The gain around the loop
is just shy of unity, so should be stable when adjusted correctly
(gain trimpots for each output tube).

Sure, and the circlotron which has the cathodes of obe tube taken to the
anodes of the other
may be more inherently stalbe than if you have a partial circlotron
using a UL OPT.

If adjusted incorrectly it can
indeed become an oscillator by providing positive feedback around the
loop. This problem is generic to the error corr. scheme in any form I
think.

Same with NFB.
If you have a poor OPT, and apply some internal FB some place,
and then also try to apply global on top of all that, it usually
will try to oscillate no matter what is done to stabilise the amp,
because the margins of stability already have been used up in the first
lot of NFB app.

If the subtractor isn't exactly spot on it either leaves some +
or - phase signal in its output, which then acts as negative or
positive feedback in a loop with near unity gain. It is quite similar
to the bootstrapped load resistor problem. At exactly unity feedback
you get a current source, below unity you get a high resistance, and
above unity you get positive feedback. In the error corr. case, to
theoretically null distortion to .0001% say, one would adjust the
error loop gain to .999999, perilously close to oscillation. So any
drift of loop gain could likely lead to instability. Thus my point
about using error corr. to get the bulk of dist. cancelled, but not
the last decimals. Class AB output makes oscillation less likely since
only one tube is on most of the time, and the tube gains are lower
during crossover. My goal in looking at the error correction idea was
to fix class AB crossover distortion just like the original Hawksford
scheme was used as a crossover dist. fix in SS amplifiers.

I don't like trying to catch the horse after it has bolted,
so I like to have a lot of class A in the amps I make before they begin
working class AB.
Then there is less reliance on correction or NFB, and the amp measures
well and sounds dreamy.

Some designs, like the McIntosh, and EAR509 use a "unity coupled output
stage"
with special tranny constructions, and they then go on to apply
more loops of NFB to total about 40 dB.

In the last pair of EAR amps I serviced, and seriously modified to be more
stable
and to keep their bias better, the total NFB was 44 dB,
and yet the Ro was 0.5 ohms, since the OPT sec wasn't in the loop, and it
couldn't
be, because of stability issues, and the thd at 100 watts was around 0.3%.

To me the amp was harsh sounding.

If I want 100 watts, I use a six pack of EL34, or quad of KT88,
not a lousy pair of EL509, which are not very linear when flogged to death

like in an EAR, working mainly in class B,
thus producing far more open loop thd than anything I make does,
and therefore needing a shirtload more NFB and gain stages to
correct all the thd.

Its dead easy to make a 100 watt UL amp with a quad of KT88 which uses
10 times less NFB, ie, 16 dB instead of 36 dB, and which then measures
the same 0.3% at a dB below clip, and 0.02% at 3 watts, where we listen,
and Ro 0.5ohms.

The error correction I was considering must be simple, easy to set up,
effective as NFB, use a minimum of parts, and be non critical in its daily
use.

The first EC amp I posted with an SET error nuller amp I posted days ago
conforms to the simple elegance I like to see in audio amps.

The balanced EC amp is obviously far more complex.

I have another absolute bottler you all will like when I draw it up later
and post it.

Patrick Turner.




Don

My reply from Sept 07 seems to have disappeared into cyber space. I
will try to remember what I said:

I posted the vacuum tube current mirror circuit on ABSE. Also is a
thread on this at the diyAudio tube forum too, in case someone can't
get ABSE. I use an assortment of fixtures with many diode tubes
connected in series, with selectable series tap points for
experimenting and designing with this idea. 6JU8 and 9006 diode tubes
are decent choices. Smaller diodes and higher gm pentodes in general
give higher current gain.

I don't try to reach extremely low levels of dist. in my designs
either, my .01% figures were just for illustration.

Saw your new class A error correction circuit on ABSE, similar to what
I had imagined would be required for a non-circlotron P-P with an
error corr. An LTP for each tube since the output xfmr. cannot be so
depended on to accurately cross couple signal between primary halves
in a normal P-P. The cross connected plates is interesting. If this
cross connecting of plates is left out, would it not work for class
AB? Most likely will want to use a trimpot on each error corr.
feedback network to set subtractor loop gain precisely, this is
something I think is needed for all P-P err. corr. designs in general,
since the output tubes will never be exactly matched.

I was thinking over your earlier comment about the cross coupling and
oscillator similarity in the circlotron like design, and maybe missed
your point somewhat in my earlier reply. My comments were on the
cathode outputs fully driving the primary between the sampling points,
so that no cross coupling thru the transformer was necessary or
likely, which is the major worry in a normal P-P err. corr. design.
But there is obvious cross coupling in the LTP error corr. circuitry
itself by design, which does make it look like a typical osc. circuit.
This similarity to an oscillator is actually inherent in all error
corr. designs, some just more obvious than others. The subtractor
circuit must be spot on in accuracy or some residual signal is left in
its output with either + or - polarity. This residual amounts to
additional positive or negative feedback added to a loop that is right
at unity gain, so oscillation is a constant threat. This is a generic
problem with all error corr. type circuits. This is somewhat like the
problem with bootstrapped load resistors, too much gain and its
positive feedback, too little gain and its not quite a current source
load but a high value resistor instead, and just the right gain makes
it look like a current source load as wanted. An err. corr. design
operating at .9999 loop gain for say .01% distortion residual (in
theory anyway) would be susceptible to any .01% variation in the
feedback loop gain, potentially causing oscillation. Hence my comment
about using error corr. circuitry to just lop off the bulk, say 90% of
distortion, but using conventional NFB for pushing the decimal point
beyond that.

Don

Could someone please explain how this so called "error correction" system
is any different from ordinary negative feedback? It still seems to
depend on an error in the output signal, just as with negative feedback to
generate the "correction" signal, unlike for example a feed forward signal
which can theoretically cancel the entire error signal in the output.
Looking at the schematics that have been posted before all I see is
slightly complicated implementations of ordinary negative feedback, what
am I missing?


Regards,

John Byrns


In article ,
(smoking-amp) wrote:

My reply from Sept 07 seems to have disappeared into cyber space. I
will try to remember what I said:

I posted the vacuum tube current mirror circuit on ABSE. Also is a
thread on this at the diyAudio tube forum too, in case someone can't
get ABSE. I use an assortment of fixtures with many diode tubes
connected in series, with selectable series tap points for
experimenting and designing with this idea. 6JU8 and 9006 diode tubes
are decent choices. Smaller diodes and higher gm pentodes in general
give higher current gain.

I don't try to reach extremely low levels of dist. in my designs
either, my .01% figures were just for illustration.

Saw your new class A error correction circuit on ABSE, similar to what
I had imagined would be required for a non-circlotron P-P with an
error corr. An LTP for each tube since the output xfmr. cannot be so
depended on to accurately cross couple signal between primary halves
in a normal P-P. The cross connected plates is interesting. If this
cross connecting of plates is left out, would it not work for class
AB? Most likely will want to use a trimpot on each error corr.
feedback network to set subtractor loop gain precisely, this is
something I think is needed for all P-P err. corr. designs in general,
since the output tubes will never be exactly matched.

I was thinking over your earlier comment about the cross coupling and
oscillator similarity in the circlotron like design, and maybe missed
your point somewhat in my earlier reply. My comments were on the
cathode outputs fully driving the primary between the sampling points,
so that no cross coupling thru the transformer was necessary or
likely, which is the major worry in a normal P-P err. corr. design.
But there is obvious cross coupling in the LTP error corr. circuitry
itself by design, which does make it look like a typical osc. circuit.
This similarity to an oscillator is actually inherent in all error
corr. designs, some just more obvious than others. The subtractor
circuit must be spot on in accuracy or some residual signal is left in
its output with either + or - polarity. This residual amounts to
additional positive or negative feedback added to a loop that is right
at unity gain, so oscillation is a constant threat. This is a generic
problem with all error corr. type circuits. This is somewhat like the
problem with bootstrapped load resistors, too much gain and its
positive feedback, too little gain and its not quite a current source
load but a high value resistor instead, and just the right gain makes
it look like a current source load as wanted. An err. corr. design
operating at .9999 loop gain for say .01% distortion residual (in
theory anyway) would be susceptible to any .01% variation in the
feedback loop gain, potentially causing oscillation. Hence my comment
about using error corr. circuitry to just lop off the bulk, say 90% of
distortion, but using conventional NFB for pushing the decimal point
beyond that.

Don


Surf my web pages at, http://users.rcn.com/jbyrns/

On Wed, 08 Sep 2004 11:51:51 -0500, (John Byrns) wrote:


Could someone please explain how this so called "error correction" system
is any different from ordinary negative feedback? It still seems to
depend on an error in the output signal, just as with negative feedback to
generate the "correction" signal, unlike for example a feed forward signal
which can theoretically cancel the entire error signal in the output.
Looking at the schematics that have been posted before all I see is
slightly complicated implementations of ordinary negative feedback, what
am I missing?


Regards,

John Byrns


[snip]

You are missing nothing.

The toooob types, however, are missing quite a few important brain
cells ;-)

...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona Voice480)460-2350 | |
| E-mail Address at Website Fax480)460-2142 | Brass Rat |
| http://www.analog-innovations.com | 1962 |

I love to cook with wine. Sometimes I even put it in the food.

I read in alt.binaries.schematics.electronic that John Byrns
wrote (in
act.com) about 'Who needs NFB when there is error correction?', on Wed,
8 Sep 2004:
Could someone please explain how this so called "error correction"
system is any different from ordinary negative feedback?

Negative feedback feeds back the *whole* output signal, which is then
subtracted from the input signal. Error feedback has a first subtraction
that eliminates the component of the feedback signal which is an exact
(we hope) scale copy of the input signal. The *residual* is then
subtracted from the input signal. So the signal in the forward path
appears to be 'pre-distorted' in such a way as to cancel the distortion
introduced in the later stages of the amplifier. But there is many a
slip 'twixt error subtraction and extreme output linearity.
--
Regards, John Woodgate, OOO - Own Opinions Only.
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

On Wednesday 08 September 2004 10:58 am, John Woodgate
did deign to grace us with the following:

I read in alt.binaries.schematics.electronic that John Byrns
wrote (in
act.com) about 'Who needs NFB when there is error correction?', on Wed,
8 Sep 2004:
Could someone please explain how this so called "error correction"
system is any different from ordinary negative feedback?

Negative feedback feeds back the *whole* output signal, which is then
subtracted from the input signal. Error feedback has a first subtraction
that eliminates the component of the feedback signal which is an exact
(we hope) scale copy of the input signal. The *residual* is then
subtracted from the input signal. So the signal in the forward path
appears to be 'pre-distorted' in such a way as to cancel the distortion
introduced in the later stages of the amplifier. But there is many a
slip 'twixt error subtraction and extreme output linearity.
--
So, bottom line, the difference is which point is defined as the summing
node, right? ;-)

Thanks,
Rich

smoking-amp wrote:

My reply from Sept 07 seems to have disappeared into cyber space. I
will try to remember what I said:

I posted the vacuum tube current mirror circuit on ABSE. Also is a
thread on this at the diyAudio tube forum too, in case someone can't
get ABSE. I use an assortment of fixtures with many diode tubes
connected in series, with selectable series tap points for
experimenting and designing with this idea. 6JU8 and 9006 diode tubes
are decent choices. Smaller diodes and higher gm pentodes in general
give higher current gain.

I was able to view the basic schematic of the dioses plus pentode
at abse, but what is the schematic for their applied use?



I don't try to reach extremely low levels of dist. in my designs
either, my .01% figures were just for illustration.

Saw your new class A error correction circuit on ABSE, similar to what
I had imagined would be required for a non-circlotron P-P with an
error corr. An LTP for each tube since the output xfmr. cannot be so
depended on to accurately cross couple signal between primary halves
in a normal P-P. The cross connected plates is interesting. If this
cross connecting of plates is left out, would it not work for class
AB? Most likely will want to use a trimpot on each error corr.
feedback network to set subtractor loop gain precisely, this is
something I think is needed for all P-P err. corr. designs in general,
since the output tubes will never be exactly matched.

I was thinking over your earlier comment about the cross coupling and
oscillator similarity in the circlotron like design, and maybe missed
your point somewhat in my earlier reply. My comments were on the
cathode outputs fully driving the primary between the sampling points,
so that no cross coupling thru the transformer was necessary or
likely, which is the major worry in a normal P-P err. corr. design.
But there is obvious cross coupling in the LTP error corr. circuitry
itself by design, which does make it look like a typical osc. circuit.
This similarity to an oscillator is actually inherent in all error
corr. designs, some just more obvious than others. The subtractor
circuit must be spot on in accuracy or some residual signal is left in
its output with either + or - polarity. This residual amounts to
additional positive or negative feedback added to a loop that is right
at unity gain, so oscillation is a constant threat. This is a generic
problem with all error corr. type circuits. This is somewhat like the
problem with bootstrapped load resistors, too much gain and its
positive feedback, too little gain and its not quite a current source
load but a high value resistor instead, and just the right gain makes
it look like a current source load as wanted. An err. corr. design
operating at .9999 loop gain for say .01% distortion residual (in
theory anyway) would be susceptible to any .01% variation in the
feedback loop gain, potentially causing oscillation. Hence my comment
about using error corr. circuitry to just lop off the bulk, say 90% of
distortion, but using conventional NFB for pushing the decimal point
beyond that.

We saw this post yesterday I think.
I addressedissues raised.

Patrick Turner.




Don

(John Byrns) wrote in message ...
Could someone please explain how this so called "error correction" system
is any different from ordinary negative feedback? It still seems to
depend on an error in the output signal, just as with negative feedback to
generate the "correction" signal, unlike for example a feed forward signal
which can theoretically cancel the entire error signal in the output.
Looking at the schematics that have been posted before all I see is
slightly complicated implementations of ordinary negative feedback, what
am I missing?


Regards,

John Byrns

As John W. described it. ...
I would call it sort of a half a--ed combination of feedback and
feedforward. The error ONLY (distortion, no signal component) is
hopefully derived by an accurate subtraction of a scaled down output
from the input signal, then the inverted error is summed back into the
input to correct the output, sort of like feedforward.
The difference from conventional NFB is in the details primarily. For
example, the attenuation factor to get the output scaled down to the
input level is a manually set factor by a trimpot and fixed attenuator
typically. If it is slightly off, then some actual signal will be left
in the residual, and this gets summed back into the input and goes
round the loop. Since the forward amplifier gain and the reverse
attenuation factor are supposed to give just a unity (well actually
just shy of unity, like .99...) gain result, this residual, depending
on polarity due to the attenuation error, either acts as a little
extra negative feedback, or acts as positive feedback. Positive
feedback into a unity gain loop is called an oscillator.
The advantages of this crazy scheme is that the amplifier does
not need all the excess forward (or open loop) gain that one needs in
conventional NFB. This can be very helpful for avoiding phase shift at
high frequencies. That in turn can enable higher bandwidth in the
amplifier and error correction circuitry than usual, making it a very
useful scheme for eliminating (at least a good part of) crossover
distortion in a class AB push-pull type amplifier. (Since crossover
distortion tends to be high harmonics of the signal, high bandwidth in
the error correction is important.) The downside of this scheme, is
instability. This is due to the manually set attenuation and the fact
that the loop gain is so near unity. Any drift in attenuation factor
or forward gain in the amplifier itself, can lead to either
oscillation or reduced error correction. For this reason, one does not
want to use this scheme to achive extremely low distortion numbers.
But it is a good complement to conventional N feedback in combination.
(Hmmm, I am surprised that someone from an analog electronics
outfit would be so negative about this idea. The SS amplifier guys are
the ones who basically invented this nutty idea. (Hawksford and others
before him on error feedback) We are just applying it to tube amps.
Also, see my posting on diyAudio about the current mirror recently,
the tube guys are the ones who really invented them! No IC would work
without this TOOOOBE technology. Not to mention that Titanium-O2
thermal emitters (0.1 eV thermal barrier) now make microlithographic
vacuum tubes quite practical and electron transit times in vacuum are
100s of times faster than in silicon. TeraHertz vacuum tube
microprocessors or Op amps anyone? )
While we are at it, let me tell you about an even nuttier, but
fantastic error correction scheme. (Star Wars - classified! Oops, just
kidding, I think) Its called pilot signal feedback. One can correct
distortion in an amplifier by predistorting the input signal in a
complementary way by using a multiplier cell. Now many amplifiers
conveniently have a diff. amp. stage with current source (LTP) for
input. By varying the current source you change the gain. By feeding a
high frequency pilot signal into the amplifier input along with the
audio, one can measure the forward gain of the amplifier in real time
and use that to control the current source for constant overall gain.
The high frequency signal gets filtered out at the output by a filter
and is above the audio band anyway. The amplifier itself needs no more
gain than the forward resultant gain (just like error correction
feedback). I could name a few more technologies, some of which haven't
even been officially invented yet, but someone would probably arrest
me. :-) Getting late, signing off.
Don

"Jim Thompson" wrote in message
...
The toooob types, however, are missing quite a few important brain
cells ;-)

Well, not this one anyway. I clearly see that, after you reduce distortion
from one "pass", the distortion is less, and thus the feedback correction is
less. So it has a limit, determined by the gain and distortion. Exactly
like global NFB.

As Rich said, the summing node is different but it still operates the same.

Tim

--
"I've got more trophies than Wayne Gretsky and the Pope combined!"
- Homer Simpson
Website @ http://webpages.charter.net/dawill/tmoranwms

John Byrns wrote:

Could someone please explain how this so called "error correction" system
is any different from ordinary negative feedback? It still seems to
depend on an error in the output signal, just as with negative feedback to
generate the "correction" signal, unlike for example a feed forward signal
which can theoretically cancel the entire error signal in the output.
Looking at the schematics that have been posted before all I see is
slightly complicated implementations of ordinary negative feedback, what
am I missing?

In a true error correction method, no signal voltage is fed back
at one of the two available ports at the input of an amp.

What is applied at the spare input is the the distortion,
after extracting it from a network of 2 resistors, R1 and R2, where R1 is from
a low distortion
source of oppositely phased voltage signal to the output, to which R2
connects.
If the values of R1&R2 are adjusted carefully, the signal voltage is nulled,
leaving
only the fraction of output distortion according to the ratio R1 / ( R1 + R2
).

This "error signal", really a fraction of the distortion which occurs when the
amp has the
the distortion reduction circuitry in place, can be also fed into the same
input
as the input voltage is the source impedance of the error signal is high,
say from the plate of a pentode tube. It has to be phased correctly, like the
error signal contained within a normal feedback signal voltage.

I suggest you carefully re-read my posts all over again to save me having to
yet again spell
out and spoon feed the info.
The schematics I have posted can and should be carefully analysed for the
intantaneous
working voltages and include distortion voltages and currents.

Patrick Turner.



Regards,

John Byrns

In article ,
(smoking-amp) wrote:

My reply from Sept 07 seems to have disappeared into cyber space. I
will try to remember what I said:

I posted the vacuum tube current mirror circuit on ABSE. Also is a
thread on this at the diyAudio tube forum too, in case someone can't
get ABSE. I use an assortment of fixtures with many diode tubes
connected in series, with selectable series tap points for
experimenting and designing with this idea. 6JU8 and 9006 diode tubes
are decent choices. Smaller diodes and higher gm pentodes in general
give higher current gain.

I don't try to reach extremely low levels of dist. in my designs
either, my .01% figures were just for illustration.

Saw your new class A error correction circuit on ABSE, similar to what
I had imagined would be required for a non-circlotron P-P with an
error corr. An LTP for each tube since the output xfmr. cannot be so
depended on to accurately cross couple signal between primary halves
in a normal P-P. The cross connected plates is interesting. If this
cross connecting of plates is left out, would it not work for class
AB? Most likely will want to use a trimpot on each error corr.
feedback network to set subtractor loop gain precisely, this is
something I think is needed for all P-P err. corr. designs in general,
since the output tubes will never be exactly matched.

I was thinking over your earlier comment about the cross coupling and
oscillator similarity in the circlotron like design, and maybe missed
your point somewhat in my earlier reply. My comments were on the
cathode outputs fully driving the primary between the sampling points,
so that no cross coupling thru the transformer was necessary or
likely, which is the major worry in a normal P-P err. corr. design.
But there is obvious cross coupling in the LTP error corr. circuitry
itself by design, which does make it look like a typical osc. circuit.
This similarity to an oscillator is actually inherent in all error
corr. designs, some just more obvious than others. The subtractor
circuit must be spot on in accuracy or some residual signal is left in
its output with either + or - polarity. This residual amounts to
additional positive or negative feedback added to a loop that is right
at unity gain, so oscillation is a constant threat. This is a generic
problem with all error corr. type circuits. This is somewhat like the
problem with bootstrapped load resistors, too much gain and its
positive feedback, too little gain and its not quite a current source
load but a high value resistor instead, and just the right gain makes
it look like a current source load as wanted. An err. corr. design
operating at .9999 loop gain for say .01% distortion residual (in
theory anyway) would be susceptible to any .01% variation in the
feedback loop gain, potentially causing oscillation. Hence my comment
about using error corr. circuitry to just lop off the bulk, say 90% of
distortion, but using conventional NFB for pushing the decimal point
beyond that.

Don

Surf my web pages at, http://users.rcn.com/jbyrns/

Jim Thompson wrote:

On Wed, 08 Sep 2004 11:51:51 -0500, (John Byrns) wrote:


Could someone please explain how this so called "error correction" system
is any different from ordinary negative feedback? It still seems to
depend on an error in the output signal, just as with negative feedback to
generate the "correction" signal, unlike for example a feed forward signal
which can theoretically cancel the entire error signal in the output.
Looking at the schematics that have been posted before all I see is
slightly complicated implementations of ordinary negative feedback, what
am I missing?


Regards,

John Byrns


[snip]

You are missing nothing.

The toooob types, however, are missing quite a few important brain
cells ;-)

Jim is like the guy who would never enjoy owning a yacht.
He'd always say that the sailors in the harbour were brainless,
and hadn't they heard of ferries, or the bus route around the bay.

The guy who flies to work in his helicopter is another variation of human
who may/may not have brain cells, depending one's mental value system.
Maybe a director of Enron who flies to work that way does challenge
our perceptions of brainpower somewhat, when we al know he ought to catch
the bus ;-)

Patrick Turner.



...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona Voice480)460-2350 | |
| E-mail Address at Website Fax480)460-2142 | Brass Rat |
| http://www.analog-innovations.com | 1962 |

I love to cook with wine. Sometimes I even put it in the food.

On Thu, 09 Sep 2004 10:25:20 +1000, Patrick Turner
wrote:



Jim Thompson wrote:

On Wed, 08 Sep 2004 11:51:51 -0500, (John Byrns) wrote:


Could someone please explain how this so called "error correction" system
is any different from ordinary negative feedback? It still seems to
depend on an error in the output signal, just as with negative feedback to
generate the "correction" signal, unlike for example a feed forward signal
which can theoretically cancel the entire error signal in the output.
Looking at the schematics that have been posted before all I see is
slightly complicated implementations of ordinary negative feedback, what
am I missing?


Regards,

John Byrns


[snip]

You are missing nothing.

The toooob types, however, are missing quite a few important brain
cells ;-)

Jim is like the guy who would never enjoy owning a yacht.
He'd always say that the sailors in the harbour were brainless,
and hadn't they heard of ferries, or the bus route around the bay.

The guy who flies to work in his helicopter is another variation of human
who may/may not have brain cells, depending one's mental value system.
Maybe a director of Enron who flies to work that way does challenge
our perceptions of brainpower somewhat, when we al know he ought to catch
the bus ;-)

Patrick Turner.


What a pathetic set of analogies.

...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona Voice480)460-2350 | |
| E-mail Address at Website Fax480)460-2142 | Brass Rat |
| http://www.analog-innovations.com | 1962 |

I love to cook with wine. Sometimes I even put it in the food.

John Woodgate wrote:

I read in alt.binaries.schematics.electronic that John Byrns
wrote (in
act.com) about 'Who needs NFB when there is error correction?', on Wed,
8 Sep 2004:
Could someone please explain how this so called "error correction"
system is any different from ordinary negative feedback?

Negative feedback feeds back the *whole* output signal, which is then
subtracted from the input signal. Error feedback has a first subtraction
that eliminates the component of the feedback signal which is an exact
(we hope) scale copy of the input signal. The *residual* is then
subtracted from the input signal. So the signal in the forward path
appears to be 'pre-distorted' in such a way as to cancel the distortion
introduced in the later stages of the amplifier. But there is many a
slip 'twixt error subtraction and extreme output linearity.

You got it.

Except that the *whole* output signal is rarely ever all fed back
in most audio power amps using a "global NFB loop."
Usually only 1/20 of the output signal is fed back globally.
The output signal is divided by a 1 : 20
resistor divider to make a smaller version of the original output voltage
and that is
what is fed back.

However feeding back all the output signal complete with its distortions
is quite common in the output stages of SS amps where all the emitter
voltage
is in series with applied base voltage, in emitter follower output stages.
This is a case of maybe 30 dB of series voltage NFB.
Amps like this also use 50 dB of series voltage NFB applied from the output
emmitters to the input diffpair, in a global loop.
So a total of 80 dB of NFB is not uncommon in SS amps.

Also in preamp outputs which use cathode followers, the output from the
cathode is
in series with the grid voltage.

In the error correction scheme I have proposed, or with NFB application,
the error voltage is amplified to produce a current in the output of the amp

which attempts to oppose the distortion current already there without the
error signal present.

Any amplifier will have a distortion voltage, Dv, at its output.

If the amp has NFB or error correction connected, the Dv
is a result of *two* factors.
One is the natural tendency of the amp to produce whatever Dv occurs without
NFB or EC
This is the "open loop distortion".
The other is the circuit's ability to create a current at the output which
cancels the open loop distortion
thus leaving a much smaller resultant current, and hence a smaller Dv at the
output.

Nothing is perfect, and its impossible to error correct fully, or without
having such schemes rendered useless at extremes of frequency where phase
shift
of the voltages and currents concerned cause instability, ie, reinforcement
of the
distortions, and the extreme frequencies, and oscillations.

Patrick Turner.



--
Regards, John Woodgate, OOO - Own Opinions Only.
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

Jim Thompson wrote:

On Thu, 09 Sep 2004 10:25:20 +1000, Patrick Turner
wrote:



Jim Thompson wrote:

On Wed, 08 Sep 2004 11:51:51 -0500, (John Byrns) wrote:


Could someone please explain how this so called "error correction" system
is any different from ordinary negative feedback? It still seems to
depend on an error in the output signal, just as with negative feedback to
generate the "correction" signal, unlike for example a feed forward signal
which can theoretically cancel the entire error signal in the output.
Looking at the schematics that have been posted before all I see is
slightly complicated implementations of ordinary negative feedback, what
am I missing?


Regards,

John Byrns


[snip]

You are missing nothing.

The toooob types, however, are missing quite a few important brain
cells ;-)

Jim is like the guy who would never enjoy owning a yacht.
He'd always say that the sailors in the harbour were brainless,
and hadn't they heard of ferries, or the bus route around the bay.

The guy who flies to work in his helicopter is another variation of human
who may/may not have brain cells, depending one's mental value system.
Maybe a director of Enron who flies to work that way does challenge
our perceptions of brainpower somewhat, when we al know he ought to catch
the bus ;-)

Patrick Turner.


What a pathetic set of analogies.

And patheticism, or pathos, has been around for so long now....

Patrick Turner.



...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona Voice480)460-2350 | |
| E-mail Address at Website Fax480)460-2142 | Brass Rat |
| http://www.analog-innovations.com | 1962 |

I love to cook with wine. Sometimes I even put it in the food.

Does anyone know how to fix the stock markets?
That would be a real contribution when compared to what is being
discussed here for the last while!!! JLS

John Byrns wrote:

Could someone please explain how this so called "error correction" system
is any different from ordinary negative feedback? It still seems to
depend on an error in the output signal, just as with negative feedback to
generate the "correction" signal, unlike for example a feed forward signal
which can theoretically cancel the entire error signal in the output.
Looking at the schematics that have been posted before all I see is
slightly complicated implementations of ordinary negative feedback, what
am I missing?

Regards,

John Byrns

A few days ago if my memory has not failed someone made
the point in this thread that the SE guys would not tolerate NFB.
That may very well be with good reason.

A while ago I made some measurements of THD while simultaneously
measuring the resulting distortion spectrum. The results showed THD
was usually reduced with NFB as we expect but the spectral
components in the result included several of higher order, all
of which had not appeared with no FB.

Higher order distortion products are far more audible
than those of say the 2nd or 3rd. Another result I observed
was more IM products. PP does not solve the problem.
Local FB in the output stage is a great help.

Cheers, John Stewart

Patrick Turner said:

What a pathetic set of analogies.

And patheticism, or pathos, has been around for so long now....

Perhaps mr. Thompson needs some humor correction? ;-)

--
Sander deWaal
"SOA of a KT88? Sufficient."

I read in alt.binaries.schematics.electronic that John Stewart
wrote (in )
about 'Who needs NFB when there is error correction?', on Thu, 9 Sep
2004:
Does anyone know how to fix the stock markets?

Yes, I could fix them but I haven't got a union card. It would be even
easier if anyone had a schematic to lend me.
--
Regards, John Woodgate, OOO - Own Opinions Only.
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

I read in alt.binaries.schematics.electronic that Sander deWaal
wrote (in rbd0k09qusee3ja4109d3hlc6uffgf0
) about 'Who needs NFB when there is error correction?', on
Thu, 9 Sep 2004:

"SOA of a KT88? Sufficient."

They will thermally run away if the control grid gets too hot due to
restricted ventilation. Guess how I know.
--
Regards, John Woodgate, OOO - Own Opinions Only.
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

John Woodgate said:

"SOA of a KT88? Sufficient."

They will thermally run away if the control grid gets too hot due to
restricted ventilation. Guess how I know.

Grid resistors over 100 kohms? ;-)

--
Sander deWaal
"SOA of a KT88? Sufficient."

I read in alt.binaries.schematics.electronic that John Stewart
wrote (in )
about 'Who needs NFB when there is error correction?', on Thu, 9 Sep
2004:
A while ago I made some measurements of THD while simultaneously
measuring the resulting distortion spectrum. The results showed THD was
usually reduced with NFB as we expect but the spectral components in the
result included several of higher order, all of which had not appeared
with no FB.

Higher order distortion products are far more audible than those of say
the 2nd or 3rd. Another result I observed was more IM products. PP does
not solve the problem. Local FB in the output stage is a great help.

Well, this is not exactly new! Under **ideal**, as opposed to real,
conditions, the magnitudes of these effects can be calculated fairly
simply, although the IM equations get pretty unwieldy. In a real
amplifier, the magnitudes are usually different, sometimes a lot
different. See IEC/EN 60268-2 and -3.
--
Regards, John Woodgate, OOO - Own Opinions Only.
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

I read in alt.binaries.schematics.electronic that Sander deWaal
wrote (in adf0k0p1o9ht0b83nsjvbqjl4jnkuln
) about 'Who needs NFB when there is error correction?', on
Thu, 9 Sep 2004:
John Woodgate said:

"SOA of a KT88? Sufficient."

They will thermally run away if the control grid gets too hot due to
restricted ventilation. Guess how I know.

Grid resistors over 100 kohms? ;-)


Small enclosures, outdoor operation, very hot weather.
--
Regards, John Woodgate, OOO - Own Opinions Only.
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk

John Woodgate wrote:

I read in alt.binaries.schematics.electronic that John Stewart
wrote (in )
about 'Who needs NFB when there is error correction?', on Thu, 9 Sep
2004:
A while ago I made some measurements of THD while simultaneously
measuring the resulting distortion spectrum. The results showed THD was
usually reduced with NFB as we expect but the spectral components in the
result included several of higher order, all of which had not appeared
with no FB.

Higher order distortion products are far more audible than those of say
the 2nd or 3rd. Another result I observed was more IM products. PP does
not solve the problem. Local FB in the output stage is a great help.

Well, this is not exactly new!

Well, that's nice, but sometimes people need to be reminded.
Seems like that may have worked for you as well!!!

Cheers Anyway, John Stewart

Under **ideal**, as opposed to real,
conditions, the magnitudes of these effects can be calculated fairly
simply, although the IM equations get pretty unwieldy. In a real
amplifier, the magnitudes are usually different, sometimes a lot
different. See IEC/EN 60268-2 and -3.
--
Regards, John Woodgate, OOO - Own Opinions Only.
The good news is that nothing is compulsory.
The bad news is that everything is prohibited.
http://www.jmwa.demon.co.uk Also see http://www.isce.org.uk