[Patrick Turner]

In some circuits one might use a CT choke to prove a dc supply to a pair
of tubes for a balanced output.

The beauty of this is that the rejection of common mode signals applied
to
the grids good, and there is PP action, as seen in every normal PP
output stage with a CT OPT,
and while in class A each tube powers output from one side or the other
of the primary winding.

But if we don't want to use an OPT, and just want to have the eqivalent
of a choke with CT, and have the two outputs
off each anode via cap coupling, then what arrangement of tubes or SS
devices act exactly
the same way as a CT choke?


I tried having two separate CCS with MJE350 to a pair of triodes in 5687
but the balance wasn't too good.
And there wasn't any CMR.


Patrick Turner.

[Patrick Turner]

flipper wrote:

On Fri, 27 Jun 2008 01:31:38 GMT, Patrick Turner
wrote:

In some circuits one might use a CT choke to prove a dc supply to a pair
of tubes for a balanced output.

The beauty of this is that the rejection of common mode signals applied
to
the grids good, and there is PP action, as seen in every normal PP
output stage with a CT OPT,
and while in class A each tube powers output from one side or the other
of the primary winding.

But if we don't want to use an OPT, and just want to have the eqivalent
of a choke with CT, and have the two outputs
off each anode via cap coupling, then what arrangement of tubes or SS
devices act exactly
the same way as a CT choke?

I don't know what would be 'the same'. For one, with a CT choke the
two sides are coupled. I know you said as much above but clearly two
independent CCS plate loads are not.

When a choke is used, the two sides are magnetically coupled,
and as one anode goes positive, its voltage rises above the stages B+
supply,
while the other descends negative below the B+.
Its like a see saw.

Maybe its impossible to have a B+ at the same level as the CT of the
choke,
and it must be above the Ea as it is when two R are used on each side of
a longtail pair,

I suppose there might be some way to cross couple the thing to achieve
a similar result and I could swear I saw something like that somewhere
but can't find it at the moment.

A load connected to one anode must be reflected to the other anode,
as it is with a CT choke, by transformer action.



I tried having two separate CCS with MJE350 to a pair of triodes in 5687
but the balance wasn't too good.
And there wasn't any CMR.

Are you looking at a single ended out?

No, must be balanced, and have exactly the same properties as a choke,
ie, have enormous input impedance at each anode connection
at signal F for opposite phased a-a signal
but very low Z for common mode or same phased signals at each anode.
It can be low impedance a-a at DC, like a choke.

The proposed application is here..

http://turneraudio.com.au/schem-300w...ut+output.html

This works extremely well with wide BW, excellent
current levels available to overcome Miller C of an output stage,
and capable of a wide V swing with balance determined by the matching of
the
following stages grid biasing Rs.

If the choke could be replaced by a network of SS that behaved just the
same as the choke,
BW could be greater, and a-a impedance even higher, and perhaps no
iron caused distortion, which BTW is extremenely low in the case of the
above schematic,
maybe 0.02% maximum and below the already low levels produced by the
tubes.

Patrick Turner.







Patrick Turner.

[Patrick Turner]

flipper wrote:

On Fri, 27 Jun 2008 08:18:13 GMT, Patrick Turner
wrote:



flipper wrote:

On Fri, 27 Jun 2008 01:31:38 GMT, Patrick Turner
wrote:

In some circuits one might use a CT choke to prove a dc supply to a pair
of tubes for a balanced output.

The beauty of this is that the rejection of common mode signals applied
to
the grids good, and there is PP action, as seen in every normal PP
output stage with a CT OPT,
and while in class A each tube powers output from one side or the other
of the primary winding.

But if we don't want to use an OPT, and just want to have the eqivalent
of a choke with CT, and have the two outputs
off each anode via cap coupling, then what arrangement of tubes or SS
devices act exactly
the same way as a CT choke?

I don't know what would be 'the same'. For one, with a CT choke the
two sides are coupled. I know you said as much above but clearly two
independent CCS plate loads are not.

When a choke is used, the two sides are magnetically coupled,
and as one anode goes positive, its voltage rises above the stages B+
supply,
while the other descends negative below the B+.
Its like a see saw.

Right.


Maybe its impossible to have a B+ at the same level as the CT of the
choke,
and it must be above the Ea as it is when two R are used on each side of
a longtail pair,

Well, certainly if you want/need as much output swing.


I suppose there might be some way to cross couple the thing to achieve
a similar result and I could swear I saw something like that somewhere
but can't find it at the moment.

A load connected to one anode must be reflected to the other anode,
as it is with a CT choke, by transformer action.



I tried having two separate CCS with MJE350 to a pair of triodes in 5687
but the balance wasn't too good.
And there wasn't any CMR.

Are you looking at a single ended out?

No, must be balanced, and have exactly the same properties as a choke,
ie, have enormous input impedance at each anode connection
at signal F for opposite phased a-a signal
but very low Z for common mode or same phased signals at each anode.
It can be low impedance a-a at DC, like a choke.

The proposed application is here..

http://turneraudio.com.au/schem-300w...ut+output.html

This works extremely well with wide BW, excellent
current levels available to overcome Miller C of an output stage,
and capable of a wide V swing with balance determined by the matching of
the
following stages grid biasing Rs.

Oh, ok. You're doing a phase splitter.

yes, but not really, its simply an LTP driven at one grid only.

But its just as good to have an LTP at the input instead of
one triode, or a parallel pair to make one triode.
With the LTP at the input, input goes to one triode and NFB to the
other,
and there's a well balanced output of a few volts which is very linear.
These few volts can then be applied to the second LTP which need not
have a CCS cathode sink,
but a R taken to say -150V, and then the balance with equal R loads at
ther anodes
is excellent. What I like to do is bring the dc to this DRIVER LTP
via high impedance a-a but low impedance commom mode, hence the choke +
R arrangement.
The balanced class A anode output is loaded mainly by the following Rgs.
These can be a low value and the coupling cap a large value, and bass
response is good,
and biasing is very firm with Rg say only 47k or less, but the load is
still
over many times the triode Ra, and thus distortion is very low.
It is with EL84 as I have them, Ra = 2k2, and at 70Vrms at each anode
THD 0.5%,
so at normal listening levels its totally insignificant.

It'd be nice to have an active circuit on a board instead of the choke.
The choke works wonders, to keep RL seen by the triodes as high as
possible,
but if a bunch of bjts acting as slaves as see-saw current feed could be
arranged,
it means the circuit could be used with say just one 6SN7 with much less
Ia than a 6BQ5 etc.

I'm always looking to expand my range of circuit options.


The CMR confused me. You must mean B+ noise rejection because you're
feeding it a single ended signal so there's no 'common' on the input
to 'reject'.

In an ordinary LTP with dc brought to each triode via an R,
and fitted with a CCS cathode sink,
any noise in the B+ rail appears at the anodes because the common mode
input R to the anodes
is extremely high.




I'm murky on how you were doing the CCS 'choke' substitute because
you've got a CCS under the cathodes.

If it acts like a choke the Cathode sink is OK.
Having CCS at the cathode, and at each anode kinda doesn't work quite as
expected.



I did try to find the mystery article but no luck. I'll keep looking,
though, because my vague memory of it now imagines PSRR was part of
it.

The only thing I ran across, so far, was a 4 triode, cross cathode
coupled, phase splitter. First two as cathode followers DC coupled to
the next two with the second cathodes into an Rk and then crossed to
the previous, opposing, Rk.

If the choke could be replaced by a network of SS that behaved just the
same as the choke,
BW could be greater, and a-a impedance even higher, and perhaps no
iron caused distortion, which BTW is extremenely low in the case of the
above schematic,
maybe 0.02% maximum and below the already low levels produced by the
tubes.

I once toyed around with something akin to a 'self adjusting CCS' but
I don't recall if I ever made it work. The idea was to not have the
CCS 'fixed' but for it to 'seek' a current setting, sort of like how a
bypassed Rk does, except it would be on the anode. Purpose was to have
high impedance at signal but not a CCS, per see, because an anode CCS
(or a pair) driving into a cathode CCS (or a pentode) has the two ends
fighting each other. Something has to ;give', hence the 'self
adjusting' idea.

Conceptually, rather that a fixed Vref setting the CCS, Vref would be
taken from a suitable resistor divider off the anode and then bypassed
so it would DC settle but stay 'constant' at signal F. That's 'sort
of' choke like.

What I sort of want is like a current mirror where
if there is a change of current in one anode circuit
it is copied exactly in the other anode circuit
but flowing in the opposite direction.

The idea is that it should try to keep *voltage* balance.

And thus one app is that you could drive one side, and from the other
side
comes an opposite phased signal. It'd mean you'd have a phase inverter
with a bunch of active devices, not quite what I am aiming for, because
we want
to have tubes doing all the active signal handling and the
equivalant circuit acting as a passive slave, like a CCS.

Maybe the choke is *the best* way to go.

Its simple, its shunt L and shunt C is easily isolated from the anodes
with series R,
so that at extremes of BW the gain of the LTP reduces slightly without
the
ultimate phase shifts of L and C, which in fact is just what we want if
we want stability
with FB.

Any SS circuit would have to be simple too....

Patrick Turner.




Patrick Turner.







Patrick Turner.

[John Byrns]

In article ,
Patrick Turner wrote:

The proposed application is here..

http://turneraudio.com.au/schem-300w...ut+output.html

Please pardon the change of subject, but I noticed that your feedback
network, consisting of R21, R22, C8, contains a resistor R21 in series
with the lead compensation capacitor, while most amplifiers don't
include this additional resistor in the feedback network.

Iain and I have been having an email conversation about this resistor,
which is also used in the Radford STA25 that Iain likes. Could you
explain why you include this resistor in the feedback network, and how
you select its value?


Regards,

John Byrns

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

[Patrick Turner]

John Byrns wrote:

In article ,
Patrick Turner wrote:

The proposed application is here..

http://turneraudio.com.au/schem-300w...ut+output.html

Please pardon the change of subject, but I noticed that your feedback
network, consisting of R21, R22, C8, contains a resistor R21 in series
with the lead compensation capacitor, while most amplifiers don't
include this additional resistor in the feedback network.

Iain and I have been having an email conversation about this resistor,
which is also used in the Radford STA25 that Iain likes. Could you
explain why you include this resistor in the feedback network, and how
you select its value?

Regards,

John Byrns

Uts a fair question John.
I have a normal type Rfb = 1k2 and Ccomp = 700pF,
but there is 100 ohms in series with 1k2.

I found that when the original circuit for the amp was built
using plain 40% UL that OPT bandwidth was from 20Hz at onset of LF
core-sat to 270kHz -3dB, at 250W.

Using any global NFB extended the HF pole too far up the band and beyond
100kHz.
I would have found that there must have been a stabilty problem or poor
square wave shape at some HF,
and sometimes the Ccomp can cause oscillations at maybe 1Mhz because
open loop BW
is so high. Usually the Ccomp merely advances the phase a bit of signals
being fed back to compensate the
lag within the rest of the amplifier, thus helping to make the global FB
phase more coherent to the
grid input signal, even when a cap load is used without any other load.
The 700pF could eventually cause a near 90D phase lead but not if there
was a series R of
some value greater than the cathode FB R at V1, only 22 ohms.

I experimented to find the best series R value, not too large or else
you get not enough
phase shift and not too small or you get RF oscillations.

I normally don't have to ever use such a series R with amps with less
OPT bandwidth.

But the OPT BW isn't all there is too it.

I don't waste days and days trying to quantify every
single L, C and R equivalent element is within the amps I build or in
the ones I modify
or rebuild with my own circuits.

So I cannot tell you how I calculated because I never calculate such
things;
I just do it right, and if you've build as many amps as I have you just
find your way to stability by educated guesses and trials of these.
Then I test well to make sure it works and nothing saturates or
overloads due to some corrective
compensation network added.

If only every other damn maker did what I did, I'd have an easier life.

Leaks had 0.001uF across the cathode Rfb on their amps, and one might
think
this negates the phase advance of the Ccomp across Rfb, and maybe that's
partailly true;
I found it did nothing
except help instabiliy, so I like to get rid of such a C.
See my pages for strange details on how I make Leak amps unconditionally
stable.
What I do is necessary because of the woefully inadequate quality of
Leak OPTs.
Leak may have thought stray RF pick up from speaker leads couldn't get
back
to the input tube cathode if it was shunted by a C.

Anyway, the more you look at old schematics, the more you see changes
and differences that don't always make sense.
Nowdays, its common to find FB networks and compo all plain WRONG,
and placing a 0.22 across the output makes many modern amps oscillate
well at low RF.
Nobody actually spends the time to get such things right, its all now
done very slap dash
and lazy.

So what you seen used in one amp may not be usable in another.

So someone copying any schematic of mine must be wary that if they have
an OPT
different to the one I wound then the FB network will be different, and
they are on their own to
solve their problems.
I get 500 hits a day at my site, and a few ppl email me about their
bothers and I advise them where
I can, but sometimes I just have to advise them to study behavioural
phenomena of L,C&R, and
learn to apply such ideas intuitively.

If I had a dollar for every diyer who build an amp using a schematic
from some magazine
or website which resulted in them getting a decent oscillator they
didn't want, I'd be rich.

Patrick Turner.

[Patrick Turner]

following stages grid biasing Rs.

Oh, ok. You're doing a phase splitter.

yes, but not really, its simply an LTP driven at one grid only.

Why, for Pete's sake, do you feel the need to muddy things up for no
reason? Yes, I know an LTP can be used for other things but yes
"really" you're doing a phase splitter.

With all due respect to our overworked Pete, yes, in the app on my site
its an LTP phase splitter.

But the same issues arrise if you wanted to make a fully balanced preamp
or driver.

But its just as good to have an LTP at the input instead of
one triode, or a parallel pair to make one triode.
With the LTP at the input, input goes to one triode and NFB to the
other,
and there's a well balanced output of a few volts which is very linear.
These few volts can then be applied to the second LTP which need not
have a CCS cathode sink,
but a R taken to say -150V, and then the balance with equal R loads at
ther anodes
is excellent. What I like to do is bring the dc to this DRIVER LTP
via high impedance a-a but low impedance commom mode, hence the choke +
R arrangement.
The balanced class A anode output is loaded mainly by the following Rgs.
These can be a low value and the coupling cap a large value, and bass
response is good,
and biasing is very firm with Rg say only 47k or less, but the load is
still
over many times the triode Ra, and thus distortion is very low.
It is with EL84 as I have them, Ra = 2k2, and at 70Vrms at each anode
THD 0.5%,
so at normal listening levels its totally insignificant.

It'd be nice to have an active circuit on a board instead of the choke.
The choke works wonders, to keep RL seen by the triodes as high as
possible,
but if a bunch of bjts acting as slaves as see-saw current feed could be
arranged,
it means the circuit could be used with say just one 6SN7 with much less
Ia than a 6BQ5 etc.

I'm always looking to expand my range of circuit options.


The CMR confused me. You must mean B+ noise rejection because you're
feeding it a single ended signal so there's no 'common' on the input
to 'reject'.

In an ordinary LTP with dc brought to each triode via an R,
and fitted with a CCS cathode sink,
any noise in the B+ rail appears at the anodes because the common mode
input R to the anodes
is extremely high.

In other words, yes, you meant B+ noise.

yes.

With a CCS cathode sink and choke and some series R
to anodes, the same thing occurs; the B+ rail must be very well filtered
and low impedance lest the rail noise is applied in common mode to each
grid of a following stage.



I'm murky on how you were doing the CCS 'choke' substitute because
you've got a CCS under the cathodes.

If it acts like a choke the Cathode sink is OK.

A CCS doesn't act like a choke.

It does in that the impedance seen by the tube is high across a range of
AF
for both a choke load or CCS, so gain will approach µ and THD will be
minimised.
The CCS is better in that it does not generate iron caused distortions.


Having CCS at the cathode, and at each anode kinda doesn't work quite as
expected.

Is that what you were trying? Dual CCS anode loads along with a
cathode CCS?

I tried that once, but dc stability isn't as good as with R loads
everywhere.
But with an Rk from common cathodes to 0V of at least about 3k for
two 6SN7 halves, balancing and 2H cancelling is better than if Rk was
say 470 ohms.



What I sort of want is like a current mirror where
if there is a change of current in one anode circuit
it is copied exactly in the other anode circuit
but flowing in the opposite direction.

A current mirror doesn't do that. It mirrors the current in the same
direction. Not to mention the input side is low impedance and the
other is high.

Indeed.

The idea is that it should try to keep *voltage* balance.

And thus one app is that you could drive one side, and from the other
side
comes an opposite phased signal. It'd mean you'd have a phase inverter
with a bunch of active devices, not quite what I am aiming for, because
we want
to have tubes doing all the active signal handling and the
equivalant circuit acting as a passive slave, like a CCS.

Well, if you want a current mirror phase splitter look at my "Looking
Glass" amp. http://flipperhome.dyndns.org/Looking%20Glass.htm

Interesting what you have done there.
I'll have to analyse it.
I would have used a conventional tubed LTP with R loads only
because the R loads can be high enough to have the tubes operating
at low THD.

The output cathode dc control is intersting, and yes, this arrangement
does keep the Idc well matched. If the amp goes into class AB,
the cathode bias voltage will rise though due to charge up effects.


The phase inversion comes from tapping off the 'opposite end' of the
loads with the advantage being large available voltage swing.

Probably nothing you'd want to use but it works for a small 'economy'
amp.

Your idea has got me thinking though....

Using a current mirror on the anodes of a LPT, though, gets you a
single ended gain multiplier, but not a phase splitter (nor balanced
output). Think about it. One side is 'free' to swing current while the
other side is fighting the anode 'mirror CCS'.

Like in this thing.

http://www.welbornelabs.com/hyb.htm

Indeed...


Maybe the choke is *the best* way to go.

Might be and as much as I like SS I have an affection for the
simplicity of 'wire around iron'. It's so deliciously first
principles.

I just wish the stuff didn't cost as much as it weighs.

I guess the PSSR comes from the coupled inductors, just like coupled
inductors on a power input rejects common mode noise.


Its simple, its shunt L and shunt C is easily isolated from the anodes
with series R,
so that at extremes of BW the gain of the LTP reduces slightly without
the
ultimate phase shifts of L and C, which in fact is just what we want if
we want stability
with FB.

Any SS circuit would have to be simple too....

Well, the 'self adjusting' semi CCS would be about the same complexity
as a plain ole CCS. It just, in theory anyway, substitutes a resistor
divider and bypass cap for a hard reference. And it should have good
PSSR since, at hum frequencies, it would be a CCS.

I need to think more about it all.

Patrick Turner.


Patrick Turner.




Patrick Turner.







Patrick Turner.

[Patrick Turner]

flipper wrote:

On Fri, 27 Jun 2008 14:08:13 GMT, Patrick Turner
wrote:

snip

Ok, I spiced up a circuit to demonstrate the idea and posted it in
alt.binaries.schematic.electronic under the title "schematic for RAT."

I didn't work on optimizing things but spice says it works.

Cathode CCS is the one I normally use but any CCS will work.

The top side is setup so the PMOS gate is biased for the expected idle
current through the source resistor at the target anode voltage: I.E.
the resistor divider (plus gate threshold). The cap bypasses to B+ so
it will DC settle but be constant at signal F.

I originally buffered the plate with another P-MOSFET, because I had a
buffer when trying bipolar, but the 1Meg is high enough that there was
little difference, buffered or not, and increasing the resistor
divider by 10x gave less than a 1% increase in gain.

It works like you'd expect from a CCS plate load. LTP gain, with the
6BQ7, is 16, pretty close to 'ideal', and balance (remembering spice
uses identical tubes) is better than 0.5%. But that also held when I
tried subbing a 6N1P for one side to simulate 'unmatched' tubes.

The bad news is, 86% of B+ hum goes straight to the plates. Same on
both, though, so it should null in the PP but tube imbalance also
imbalances the plate hum. The CCSs don't help there.

But it's got great gain and balance.

It also works with bipolars but needs a plate buffer into the divider
because of the bipolar drive current\ and there's an odd 1/4 dB rise
(or dip, depending on which side) around 1.5kHz I didn't pin down
because it's not there with the MOSFET and that's simpler.

Finding high enough voltage P=MOSFETs might be a problem, though.

I can see the text only comments about your post at ABSE but no
schematic.

Maybe it'll turn up in a day or so, maybe it won't.

Patrick Turner.

[Patrick Turner]

flipper wrote:

On Sun, 29 Jun 2008 05:53:21 GMT, Patrick Turner
wrote:



flipper wrote:

On Fri, 27 Jun 2008 14:08:13 GMT, Patrick Turner
wrote:

snip

Ok, I spiced up a circuit to demonstrate the idea and posted it in
alt.binaries.schematic.electronic under the title "schematic for RAT."

I didn't work on optimizing things but spice says it works.

Cathode CCS is the one I normally use but any CCS will work.

The top side is setup so the PMOS gate is biased for the expected idle
current through the source resistor at the target anode voltage: I.E.
the resistor divider (plus gate threshold). The cap bypasses to B+ so
it will DC settle but be constant at signal F.

I originally buffered the plate with another P-MOSFET, because I had a
buffer when trying bipolar, but the 1Meg is high enough that there was
little difference, buffered or not, and increasing the resistor
divider by 10x gave less than a 1% increase in gain.

It works like you'd expect from a CCS plate load. LTP gain, with the
6BQ7, is 16, pretty close to 'ideal', and balance (remembering spice
uses identical tubes) is better than 0.5%. But that also held when I
tried subbing a 6N1P for one side to simulate 'unmatched' tubes.

The bad news is, 86% of B+ hum goes straight to the plates. Same on
both, though, so it should null in the PP but tube imbalance also
imbalances the plate hum. The CCSs don't help there.

But it's got great gain and balance.

It also works with bipolars but needs a plate buffer into the divider
because of the bipolar drive current\ and there's an odd 1/4 dB rise
(or dip, depending on which side) around 1.5kHz I didn't pin down
because it's not there with the MOSFET and that's simpler.

Finding high enough voltage P=MOSFETs might be a problem, though.

I can see the text only comments about your post at ABSE but no
schematic.

Maybe it'll turn up in a day or so, maybe it won't.

Well, that sucks.

I know it's at least there on my server because I downloaded it to
verify I didn't screw something up.

Ok, I made a quickie orphan page on my personal web server so you can
see it there.

http://flipperhome.dyndns.org/Self%20Adjusting%20CCS

'Orphan' in that there's no link to it from any other page on my site
but the above link should take you straight to it. That is, assuming
power don't glitch and bring the server down.


This looks just fine to me, one click, and it just displayed OK in
Nutscrape.


Its a nice circuit you have there, although I'd simplify the cathode
current sink
by using 1 x MJE340 between the two k and a negative rail, with say 4k7
as the emitter R etc.

What about the mosfet input capacitance?

As you have drawn it, the RL at dc at each anode = ( 1M/150k ) x 1k
approx, or about 6k ohms,
because the C becomes open at dc, and the 1m and 150k act as a divider
to
apply a slow F signal voltage to the gate and also to the 1k at the
mosfet source.

But if C is say 22uF, then at 1kHz, the R load seen by the anodes
becomes vitually 1M,
unless a following stage has cap coupled Rg.
These Rg can be a lot lower value than 1M, maybe 47k each, and if well
matched, the output will also
be well matched.

The only problem is if an output tube has a bit of input grid current
even at idle
due to its age or a fault, and the Rgrid-in is substantial enough to add
to the
47k biasing R so it becomes less than 47k, and voltage balance then
becomes poor.

But the effect of a crook output low Z grid input tube can unbalance
other varieties of driver/phase splitter
arrangements.

The way you have set up the CCS at anodes means that Vdc at the anodes
tends to stay constant
over a wide range of signals, because the load is low at dc, like a
choke,
where the load at dc is the wire resistance.

You can use your style of CCS, or rather high Z dc supply for a pentode
tube to enable it
to work with extremely high gain. The low load at dc stops the sway of
Vdc
you'll get with signal amplitude variation due to rectifying effects of
the 2H.

Open loop gains of 0ver 1,000 are possible, and if a tube with OLG gain
= 1,000
is reduced to say 10 for a line stage with shunt FB from a cathode
follower direct connected to the
pentode anode, the Rout is less than 20 ohms and THD is the open loop
value divided by 100,
so 0.2% at 2Vrms from the pentode becomes 0.002%, somewhat blame free.

And consider a darlington pair connected pair of MJE350, or MJE350
with some other smaller HV P bjt for the pair.
methinks maybe there is less parasitic C with the bjts.

The other thing is that the 1M and 150k act as a NFB network at dc to
give the low value RL at DC
and hence tend to regulate the anode voltages of the triodes or any
other tube in the LTP.

I use MJE350 instead of your mosfet; works fine.

So, at dc the loads offered by the actives are effectively 6k each.
If the cathode sink current were to alter say 1mA, there is only going
to be a small
3V change at each anode.

In my 845 amps, I have an SET driver with 3 x EL84 in triode all
paralleled
with total Ia = 36mA and Ea = 290V and 7k plus 60H in series to the
+620V +ve rail.
the cap coupled Rg of 23.5k dominates the loading.
I get 160Vrms max at less than 2%.
I need only 120Vrms max and that's at 1.4%.
at 12Vrms at normal listening, distortion is negligible, and it cancless
with output tube disotrtion and it reduced 8dB by the GNFB to quite
negligible levels.
I chose the choke because I don't trust 3 prong lil black varmints to
last long
in a box with hundreds of volts swinging around.

But I have an MJE350 CCS dc load at the paralleled 6CG7 input triodes.
I wanted its thd to be low as possible because it adds to that of the
output.
Operating voltages are mild, so the SS CCS is trusted in this situation.

The MJE350 base voltage is gained from a divider between the stage B+
and 0V,
so even at dc the Z is very high.
sway in anode Vdc with signal amplitude is very low because the 2H
distortion is low.

And BTW, ppl used to measure 2H with a voltage meter.
They'd set up a triode with all the RL used to DC, then measure the Vdc
across the R at idle to find the quiescent Idc, or Q point.
Then with say 20Vrms of wanted output signal but well below clipping,
Vdc across the RL is measured.
THD % = 100 x ( change in Idc x 0.707 ) / Idc at idle.

Well, I think that's the right calc to make, but the math whiz kids here
will have a better idea.

But I use a THD filter. No calculations. No worries about
load line shifts or rectification effects.

If I need a fast simple answer, I don't make calculations that take all
morning.


Patrick Turner.








Patrick Turner.

[Patrick Turner]

flipper wrote:

On Sun, 29 Jun 2008 05:41:41 GMT, Patrick Turner
wrote:

snip it down


yes.

With a CCS cathode sink and choke and some series R
to anodes, the same thing occurs; the B+ rail must be very well filtered
and low impedance lest the rail noise is applied in common mode to each
grid of a following stage.

Ok, I thought you were saying the CT choke rejected B+ noise.

It doesn't. It cannot.

And in an a PP output stage, if there is a lot of PS noise at the CT, it
is applied
to each anode of the output stage.
Peter Walker didn't think this mattered in Quad-II; there is 17Vrms
of Vripple at the OPT, bleedin terrible, but the Ra looking into
the anodes is high, and the change in tube gm is low,
so PS caused artifacts don't exceed THD. But they are just as much
though in class A and
worse in class AB.

In a triode output stage, if there is substantial noise at an OPT CT,
the LOW Ra of both tubes is subject to the Vripple,
and so the Iripple flow in each output tube is substantial
and this substantially changes the gm and the IMD is severe as a result.

Conclusion?

Use HUGE C values to anchor down the dc supply at CT chokes and OPTs
everywhere you use them.

Provide filtering to keep Vripple at CT at less than 30mV,
and at early input stages far lower levels.

SE amps need even stricter B+ filtering, expecially SET amps with little
NFB.



Well, my little MOSFET gimmick works a lot like independent chokes
except no inductive swing, of course.

No see saw effect.

But if the RL loads to each side of the LTP remain equal
dynamically then the circuit should balance very well.



http://flipperhome.dyndns.org/Self%20Adjusting%20CCS

I'm murky on how you were doing the CCS 'choke' substitute because
you've got a CCS under the cathodes.

If it acts like a choke the Cathode sink is OK.

A CCS doesn't act like a choke.

It does in that the impedance seen by the tube is high across a range of
AF
for both a choke load or CCS, so gain will approach µ and THD will be
minimised.
The CCS is better in that it does not generate iron caused distortions.

Well, I tend to think of a CCS as akin to a 'big value resistor',
rather than a choke, because it operates down to DC. That's the
'difference' in the new circuit.

Any CCS rarely has to actually be a pure CCS, with unmeasurably high
resistance.
Stray C within the device won't let you have huge Z.
But the CCS is an avenue where the ac power wasted in a dc supply R can
be avoided
almost entirely, and the tubes effectively loaded by a much higher load
value than if only R were used,
and Idc at idle can be higher, and a better working point used for
llower thd/imd.
And the Rg used fr the following stage can be much lower, thus much
better regulating
Eg, which all too easily can go too positive when you don't want it to
as the output or other tubes age.



Having CCS at the cathode, and at each anode kinda doesn't work quite as
expected.

Is that what you were trying? Dual CCS anode loads along with a
cathode CCS?

I tried that once, but dc stability isn't as good as with R loads
everywhere.
But with an Rk from common cathodes to 0V of at least about 3k for
two 6SN7 halves, balancing and 2H cancelling is better than if Rk was
say 470 ohms.

ok.

more trim

Well, if you want a current mirror phase splitter look at my "Looking
Glass" amp. http://flipperhome.dyndns.org/Looking%20Glass.htm

Interesting what you have done there.

Thank you. That means a lot to me.

I'll have to analyse it.
I would have used a conventional tubed LTP with R loads only
because the R loads can be high enough to have the tubes operating
at low THD.

Yes, and I did that on the 13FD7 amp I'm working on right now. Well,
not really an LTP, it's like a Williamson. Unit 1 in the 13FD7s are
the common cathode voltage multipliers and I added a 6BQ7 for the
front end triode and split load phase splitter.


So you have cascaded LTPs for the input / driver amp?

I totally rewired an ARC VT100 which is a pig of a thing normally,
with 5 x j-fets as CCS used in a horribly complex circuit.

I revised so there were two LTPs, with the input pair with cathode sink
= MJE450 CCS taken to a -ve bias supply.
Each side of the input LPT was a paralleled 6DJ8, with R loading at
anodes.
The input goes to one side, GNFB to the other.
The driver LTP became a pair of 12BH7 each one paralleled,
with 4 x 6550 output tubes cap coupled to BH7 anodes, an the fixed bias
applied to the output grids.

The BH7 have R loads to each anode, and have a common Rk taken to -127V
bias.
The large value of Rk means common mode amplification is negligible,
while balance is excellent.
LTPs driven at one side only NEED the CCS cathode tail to get balance,
but if the grid drive is balanced its much less important, even in
Williamsons.
Placing an extra 3k3 in the Willy tail R does wonders for balance
accuracy.


The Turnerized ARC soundeds magnificent, with firm foundation, creamy,
detailed, and all aspects if
distortion and stability improved and with less total GNFB used.
No more blown fuses and distressed client who is fed up with tube
replacements
and frequent servicing.

I did a whole shirt and trouser load more things to the bloomin ARC,
but that's another story.
The yanks have forgotten how to make good reliable simple amplifiers.
ARC, McIntosh, Manely Labs and others are far too optimistic
about the longevity of their designs, and full of awkward
comprimises to allow features such as balanced or non balanced inputs,
and easy choice of output loads. Don't get me started.

So the cascaded LTP set up works very well despite it being just
slightly
more complex than my simple LTP with SET input idea.
It used half the circuit parts that are in a VT100 circuit.
The cascaded LTP virtually eliminates all 2H which comes from input
stage
triodes operating on their own.
I've used a similar arrangement in totally re-wired Manley Snappers.
SET input stage 2H either adds or cancels slight 2H from imbalances in
an output stage
if the output tubes are not exactly matched, which is normal
in most PP amps in the real world.
So two channels can end up with very different thd and imd profiles.
I'd prefer less disimilarties between channels.



The current mirror phase splitter was to get it all in the two bottles
and, at the time, I was rather enamored with current mirrors.

The output cathode dc control is intersting, and yes, this arrangement
does keep the Idc well matched. If the amp goes into class AB,
the cathode bias voltage will rise though due to charge up effects.

Right, it has the same 'charge up' characteristic as plain ole cathode
bias but it's also self adjusting like plain old cathode bias so
there's no bias pots for the non technical to mess with.

Music signals rarely get large enough in their average value
to make Ek rise.
But if they do, you can dynamically bypass the excess
charge up signal currents as in
http://turneraudio.com.au/schem-300w...tabilizer.html


I have an opamp version for fixed bias that gets it down to 1 pot
(since the second side 'tracks' the first) but I haven't worked out a
completely self adjusting version for fixed bias yet.

Some form of ('automatic') output current balancing has become sort of
a 'trademark' in my PP designs with a current mirror being the most
common, so far, because of its self adjusting nature.

I explored such bias current equalization years ago
with an LTP using a pair of MJE350 in an LTP arrangement.
It worked well with two output tubes, until NFB was applied.
Then I had a good LF phase shift oscilator.
I abandoned the idea, and I sometimes use more than 2 outputs,
so the aim became have NO adjustments that will ALWAYS
confuse many owners, and yet maintain good enough regulation of Ek for
and hence Ik for all output tubes.

If a tube goes wrong, OK, an active fault detection circuit turns the
damn amp off
well before anything glows red hot.



Like the little 'PC Speaker' amp. It's in there.

http://flipperhome.dyndns.org/6AW8PCSpkr.htm

And the 6GK6 amp

http://flipperhome.dyndns.org/StealthAX.htm

The MJEs are on those boffo heatsinks

Speaking of charge up, I noticed that Broskie talked about your 'anti
charge up' circuit, congratulations, but was miffed why he spoke as if
the limited charge up was a 'problem'.

Maybe he didn't actually try out the circuit.

read my page quoted above and you'll see just how easy it is to apply to
any existing tube amp with cathode bias.

The phase inversion comes from tapping off the 'opposite end' of the
loads with the advantage being large available voltage swing.

Probably nothing you'd want to use but it works for a small 'economy'
amp.

Your idea has got me thinking though....

There's a variation on the theme that produces tons of gain but it
potentially takes adjusting (depending on just how ambitious one gets
with the gain) to center things up. Basically, run one CCS into the
plate and use the tube to 'subtract' from it. The 'left over' current
(which will be mostly signal) can then go through a very high load R
for increased gain (there's no plate feedback in the tertiary route).
But, as you can imagine, the CCS and tube have to be doing close to
the same current for things to bias up right.

That's this early one.

http://flipperhome.dyndns.org/13EM7CMPP.htm

I finally decided that, in the case of the 6EM7 anyway, it wasn't
worth the bother getting all that gain in 'one triode' since a pair of
them comes with two. hehe So, the 'simpler' current mirror phase
splitter.

Its quite easy to get HUGE voltage gain with bjt drivers.

But I like to have triodes in control of all voltage amplification.
The SS just acts as grovelling slaves to the tubes' every current or
voltage whim...



I did breadboard working versions of those, though. Still got the
little perfboard with. 'surprise', a trimmer pot to adjust the CCS.

more trim

Well, the 'self adjusting' semi CCS would be about the same complexity
as a plain ole CCS. It just, in theory anyway, substitutes a resistor
divider and bypass cap for a hard reference. And it should have good
PSSR since, at hum frequencies, it would be a CCS.

I need to think more about it all.

Let me know what you think about the MOSFET solution.

http://flipperhome.dyndns.org/Self%20Adjusting%20CCS

If time permits, there is endless juggling of possibilities....

But the clock screams at me to work, lest my bank mananger gets upset.

Patrick Turner.

[Patrick Turner]

flipper wrote:

On Sun, 29 Jun 2008 11:10:11 GMT, Patrick Turner
wrote:



flipper wrote:

On Sun, 29 Jun 2008 05:53:21 GMT, Patrick Turner
wrote:



flipper wrote:

On Fri, 27 Jun 2008 14:08:13 GMT, Patrick Turner
wrote:

snip

Ok, I spiced up a circuit to demonstrate the idea and posted it in
alt.binaries.schematic.electronic under the title "schematic for RAT."

I didn't work on optimizing things but spice says it works.

Cathode CCS is the one I normally use but any CCS will work.

The top side is setup so the PMOS gate is biased for the expected idle
current through the source resistor at the target anode voltage: I.E.
the resistor divider (plus gate threshold). The cap bypasses to B+ so
it will DC settle but be constant at signal F.

I originally buffered the plate with another P-MOSFET, because I had a
buffer when trying bipolar, but the 1Meg is high enough that there was
little difference, buffered or not, and increasing the resistor
divider by 10x gave less than a 1% increase in gain.

It works like you'd expect from a CCS plate load. LTP gain, with the
6BQ7, is 16, pretty close to 'ideal', and balance (remembering spice
uses identical tubes) is better than 0.5%. But that also held when I
tried subbing a 6N1P for one side to simulate 'unmatched' tubes.

The bad news is, 86% of B+ hum goes straight to the plates. Same on
both, though, so it should null in the PP but tube imbalance also
imbalances the plate hum. The CCSs don't help there.

But it's got great gain and balance.

It also works with bipolars but needs a plate buffer into the divider
because of the bipolar drive current\ and there's an odd 1/4 dB rise
(or dip, depending on which side) around 1.5kHz I didn't pin down
because it's not there with the MOSFET and that's simpler.

Finding high enough voltage P=MOSFETs might be a problem, though.

I can see the text only comments about your post at ABSE but no
schematic.

Maybe it'll turn up in a day or so, maybe it won't.

Well, that sucks.

I know it's at least there on my server because I downloaded it to
verify I didn't screw something up.

Ok, I made a quickie orphan page on my personal web server so you can
see it there.

http://flipperhome.dyndns.org/Self%20Adjusting%20CCS

'Orphan' in that there's no link to it from any other page on my site
but the above link should take you straight to it. That is, assuming
power don't glitch and bring the server down.


This looks just fine to me, one click, and it just displayed OK in
Nutscrape.


Its a nice circuit you have there,

Thanks.

although I'd simplify the cathode
current sink
by using 1 x MJE340 between the two k and a negative rail, with say 4k7
as the emitter R etc.

Yeah, as I said, any CCS will do down there. The double transistor is
just my 'standard' because I'm usually working without a negative
supply and it has the lowest voltage overhead to get under the
cathode.


What about the mosfet input capacitance?

Doesn't matter because its out of circuit at signal F.

As you have drawn it, the RL at dc at each anode = ( 1M/150k ) x 1k
approx, or about 6k ohms,
because the C becomes open at dc, and the 1m and 150k act as a divider
to
apply a slow F signal voltage to the gate and also to the 1k at the
mosfet source.

Yep. That's the 'trick' to it.

But if C is say 22uF, then at 1kHz, the R load seen by the anodes

Same with 1uF because the time constant there is the resistor divider.

becomes vitually 1M,

Yep

I tried 10x that and it made less than a 1% gain difference so 1 Meg
is 'enough'.

unless a following stage has cap coupled Rg.
These Rg can be a lot lower value than 1M, maybe 47k each, and if well
matched, the output will also
be well matched.

The only problem is if an output tube has a bit of input grid current
even at idle
due to its age or a fault, and the Rgrid-in is substantial enough to add
to the
47k biasing R so it becomes less than 47k, and voltage balance then
becomes poor.

But the effect of a crook output low Z grid input tube can unbalance
other varieties of driver/phase splitter
arrangements.

Precisely. Its not something 'special' to this one.

Besides, use good tubes

The way you have set up the CCS at anodes means that Vdc at the anodes
tends to stay constant
over a wide range of signals, because the load is low at dc, like a
choke,
where the load at dc is the wire resistance.

Yes, that was the idea. And there's nothing magic, per see, about the
1k except for MOSFET threshold variation. I.E. the higher that R is
(1k) the less percent threshold is of the total so it's effect is
less. I just arbitrarily made them roughly equal: 3mA through the 1k
and about 3V gate threshold, as a 'pick something' number. Well, it
was 3mA when I started but it's up to 3.7mA now.

On the other hand, if you make it 'real small' then the divider
becomes a large ratio, difficult to manage, C goes up, and a large
plate change would be a small divider change. 'Regulation' goes down.

I don't think any of it is real critical, though, unless you go to
extremes.

You can use your style of CCS, or rather high Z dc supply for a pentode
tube to enable it
to work with extremely high gain. The low load at dc stops the sway of
Vdc
you'll get with signal amplitude variation due to rectifying effects of
the 2H.

Open loop gains of 0ver 1,000 are possible, and if a tube with OLG gain
= 1,000
is reduced to say 10 for a line stage with shunt FB from a cathode
follower direct connected to the
pentode anode, the Rout is less than 20 ohms and THD is the open loop
value divided by 100,
so 0.2% at 2Vrms from the pentode becomes 0.002%, somewhat blame free.

Yeah, it should work for a pentode too.

Maybe I'll try that just to see a gain of 1000. hehe


A typical pentode with fully bypassed Rk such as a 6AU6
with say 4mA will have gm = about 3mA/V,
and Ra = about 500k. When I say "about", I mean approximately,
because figures are a bit variable.
µ = gm x Ra, so 0.004 x 500,000 = 2,000, so if the
anode load = say 1M,
then A = 2,000 x 1M / ( 1M + 0.5M ) = 2,000 / 1.5 = 1,333, or a lot.
If RL was closer to a a true CCS and 10M, gain approaches 2,000.

But if Ra = 500k, and RL was say 2M, then Rout = 400k.
If the stray C from anode to whatever else is coupled = 30pF,
then the HF pole is at 13 kHz.

Choke loads for pentodes don't work well at all unless large amounts of
NFB are used
because the shunt C and shunt L mean an arched shape to the F response
and there is a huge amount of iron distortion you get where the iron
coil
isn't shunted by low drive R.

The µ-follower with pentode and top tube a high µ triode such as a 12AT7
work well, and where you supply a fixed bias to the top triode
to get dc stability with changing signal levels due to
pentode 2H.
The bias R to the 12AT7 should be 2M, and the R between
triode k and pentode a should be maybe 20k, or even a transistor CCS,
and the pentode a is then cap coupled to the triode grid.
The pentode sees an anode load of 2M plus triode A x 20k in parallel,
or about 500k, and gain is still quite high at about 1000.

Shunt FB from the output side of the coupling cap off the triode
follower
is used, say R2 = 500k back to the pentode grid, with input R1 = 47k.
A' will be about 10.
The input terminal can be 100k, to bias the pentode, and Rin will
then be 47k//100k = 33k.

A triode pentode such as 6U8A can be used for a nice line stage.
But the loading of the top triode
is actually Triode Ra + ( triode µ + 1 x triode rk),
and so the the higher the triode µ, the higher the load on the pentode
below it,
and the higher the open loop gain and the greater is the amount of
applied NFB and the
more effect there is on Rout with NFB and thd/imd etc.

Just watch out for parasitic oscillations though.

A 6EJ7 plus 12AT7 paralleled makes a strong combination....



And consider a darlington pair connected pair of MJE350, or MJE350
with some other smaller HV P bjt for the pair.
methinks maybe there is less parasitic C with the bjts.

Well, I mentioned I tried bjts but got an odd 1/4dB 'bump' in the
frequency response at about 1.5k I haven't isolated yet. That was
using the plate buffer, though.

???


Darlingtons work just fine. I just used the MOSFET example because
it's 'simpler'.

Mosfets tend to have popcorn noise. depends where you use them though;
in high signal circuits, no worries, snr will be OK.
The shunt C between gate and source probably
isn't a worry as you suggest because if it was say 400pF,
and the Rsource = 1k, the pole is at 400kHz, and OK.
In a CCS where gate is bypassed to the rail well,
the 1k becomes a lower and reactive C load at the source,
so at extreme HF the CCS becomes a queer thing, a gyrator maybe?
Watch out for oscillations.



LF response is just a fractional 'hair' less with darlingtons because
you have the emitter resistor, darlingtoned up, in parallel, but
that's easy enough to make up for with a 'hair larger' C. HF response
is just a fractional 'hair' better but it's so close it's hard to pin
down why. Drain-source capacitance maybe.

In my CCS if found the MJE350 is fine on its own,
and no real need for a darlo because the hfe is around 100,
so if Ic = 5mA, Ib = 0.05mA, and not large enough
to upset normal set up much.


Actually, I just did a quick spice check and for the 6BQ7 circuit a
single MPSA92 works just as good as darlingtons but both have about 1%
imbalance vs under 1/2% for the MOSFET. I don't know why.

bjt matching?



I had just assumed a Darlington would be needed but the single PNP
working is just as simple as the MOSFET, it just takes a larger bypass
cap because of the effective emitter impedance. How well a particular
bjt would work probably depends on the beta. Or maybe not. I mean,
it's essentially 'out of circuit' too at signal F so as long as
collector impedance is high enough the 1 Meg dominates. Well, wait a
minute, there has to be enough gain for the 1Meg to bias it on. Ok,
that's the limiting factor. Which isn't too bad except for device to
device beta variation throwing off the effective emitter impedance and
affecting the resistor divider. So back to the Darlington unless you
want to 'trim' each one.

In my apps the gain is dominated by the following grid bias loads
which are well below any CCS or other R loads associated with CCS.

As the R used with the bjts or mosfets to bias them rise,
the balance from an LTP tends to drift a bit apart.

In my case I just like to see RL of each anode above 10 x Ra,
and once this is fullfilled, the thd becomes real low,
and making RL any higher doen't give much more thd reductions.

Even with a pure CCS load, a trioded 6BQ5 or a 6SN7
will still have some thd, and you will find the
distances between Ra lines close up as you move left across thre data
sheets.
And so there will always be some 2H no matter how high RL becomes.
And some 3H, at a low level, and when you have an LTP,
the differences in gm at different Ia levels between the two triodes
mean that you get some 3H generated, and some un-cancelled 2H,
so raising RL above 10 x Ra is somewhat pointless.

Having loads at over 10Ra is a heck of a lot better than having RL =
2Ra, or 3Ra,
like I have seen in many commercial amps.

33k for the dc carrying R to a 1/2 6SN7 is often used with Rg following
at 100k.
So RL = 25k only, and at 5mA, Ra = 10k, so RL = 2.5Ra, and thd is always
on the high side.



The other thing is that the 1M and 150k act as a NFB network at dc to
give the low value RL at DC
and hence tend to regulate the anode voltages of the triodes or any
other tube in the LTP.

Right.

The link to plate V, which gives the 'regulation', is the
'breakthrough' vs the first time I took a shot at it.

It works.

I use MJE350 instead of your mosfet; works fine.

So, at dc the loads offered by the actives are effectively 6k each.
If the cathode sink current were to alter say 1mA, there is only going
to be a small
3V change at each anode.

I think that's right but I hadn't done a detailed look at it past the
'different tube' simulation.


In my 845 amps, I have an SET driver with 3 x EL84 in triode all
paralleled
with total Ia = 36mA and Ea = 290V and 7k plus 60H in series to the
+620V +ve rail.
the cap coupled Rg of 23.5k dominates the loading.
I get 160Vrms max at less than 2%.
I need only 120Vrms max and that's at 1.4%.
at 12Vrms at normal listening, distortion is negligible, and it cancless
with output tube disotrtion and it reduced 8dB by the GNFB to quite
negligible levels.
I chose the choke because I don't trust 3 prong lil black varmints to
last long
in a box with hundreds of volts swinging around.

But I have an MJE350 CCS dc load at the paralleled 6CG7 input triodes.
I wanted its thd to be low as possible because it adds to that of the
output.
Operating voltages are mild, so the SS CCS is trusted in this situation.

The MJE350 base voltage is gained from a divider between the stage B+
and 0V,
so even at dc the Z is very high.
sway in anode Vdc with signal amplitude is very low because the 2H
distortion is low.

And BTW, ppl used to measure 2H with a voltage meter.
They'd set up a triode with all the RL used to DC, then measure the Vdc
across the R at idle to find the quiescent Idc, or Q point.
Then with say 20Vrms of wanted output signal but well below clipping,
Vdc across the RL is measured.
THD % = 100 x ( change in Idc x 0.707 ) / Idc at idle.

Well, I think that's the right calc to make, but the math whiz kids here
will have a better idea.

But I use a THD filter. No calculations. No worries about
load line shifts or rectification effects.

If I need a fast simple answer, I don't make calculations that take all
morning.

I like simple

So do me.

Patrick Turner.



Patrick Turner.

[Patrick Turner]

flipper wrote:

On Sun, 29 Jun 2008 12:08:12 GMT, Patrick Turner
wrote:



flipper wrote:

On Sun, 29 Jun 2008 05:41:41 GMT, Patrick Turner
wrote:

snip it down


yes.

With a CCS cathode sink and choke and some series R
to anodes, the same thing occurs; the B+ rail must be very well filtered
and low impedance lest the rail noise is applied in common mode to each
grid of a following stage.

Ok, I thought you were saying the CT choke rejected B+ noise.

It doesn't. It cannot.

Well, that's what I thought but then it's always possible I missed
something.

And in an a PP output stage, if there is a lot of PS noise at the CT, it
is applied
to each anode of the output stage.
Peter Walker didn't think this mattered in Quad-II; there is 17Vrms
of Vripple at the OPT, bleedin terrible, but the Ra looking into
the anodes is high, and the change in tube gm is low,
so PS caused artifacts don't exceed THD. But they are just as much
though in class A and
worse in class AB.

In a triode output stage, if there is substantial noise at an OPT CT,
the LOW Ra of both tubes is subject to the Vripple,
and so the Iripple flow in each output tube is substantial

Yeah. having now switched to a triode, vs pentode, output stage on
this latest amp I realized that.

Ain't no 'free lunch' and triodes aren't a miracle cure for
everything.

and this substantially changes the gm and the IMD is severe as a result.

Conclusion?

Use HUGE C values to anchor down the dc supply at CT chokes and OPTs
everywhere you use them.

Provide filtering to keep Vripple at CT at less than 30mV,
and at early input stages far lower levels.

SE amps need even stricter B+ filtering, expecially SET amps with little
NFB.



Well, my little MOSFET gimmick works a lot like independent chokes
except no inductive swing, of course.

No see saw effect.

True, but I'm not convinced there's 'see saw' in the CT choke either.
The whole point is to have impedance so high that current becomes
negligible so what's to 'see saw' in the iron?

See-saw is the analogy of a centre tapped winding.
the voltages swing up and down and are locked together
magnetically, like the see saw beam.
While one tube turns on, more current flows in 1/2 the winding,
and the current decreases in the other 1/2, and both tubes
act together to swing their dc supply without having to make
ac power into R loads which could be wasted
This all happens in addition to the loop of ac current flow
through anode to anode loads and tubes and CCS cathode sink.
Its all like a nice little quintet all playing nice music.



The benefit is balanced, net zero, Idc.

snip,

So you have cascaded LTPs for the input / driver amp?

No, it's a Williamson. There's only 1 'pair' and they don't have a
long tail, just a common Rk. That does a little 'see saw' but the
primary benefit is obviating the need for Rk bypass caps so you save
one resistor and two caps (vs independent Rks).

The problem with the Willy balanced amp as originally designed is
its poor 2H cancelling ability.

The 2H currents in each tube are of the same phase, even though the
output voltages are opposite phases.

So the 2H in both triodes are common mode currents.
The single 'short tail' resistor isn't large enough in value
to give a large amount of common mode local current FB so most of the 2H
in each triode apears as a voltage at each anode.
If the anode loads are say 40k each total with following Rg,
then its 20k common load, and to reduce the 2H at each anode to
negligible levels
the Rk should be much more than 600 ohms, and perhaps 4k7 instead, at
least,
but 10k taken to a negative supply is better a CCS is the best.
Then you'll find that at the common cathodes there should be no signal
voltage
and the only voltage present is a sample % of the 2H that would have
otherwise
appeared at each anode.
The large Rk or CCS eliminates the 2H at each anode.

The pair are said to be much more linear, with a tiny amount of mainly
3H,
and IMD is much reduced.

The williamson is a little under engineered, ie, simpler than it should
be,
and a larger Rk does it some real good.

The front two are gain triode into a split load (concertina,
cathodyne, take your pick) phase splitter.

Now, on the 6GK6, triode mode, PP amp that I plan to get back around
to 'one of these days' I do have double, cascaded, CCS 'LTP' stages
(it's also fixed bias with MOSFET buffered +ve drive and the opamp
output current balance). It's 'PP' all the way through but I didn't do
that with the 13FD7 as it's supposed to be 'simple'



Btw, I did find at least one of the 'self balancing' circuits I was
thinking of. It's in RDH4 section 12 under phase splitters: "See-saw
self-balancing phase inverter" But it doesn't have any better balance,
about 7%. Plates are cap coupled to equal resistors (1Meg), the
joining node of which is taken to the opposing grid so any 'difference
between the two plate signals is applied to that grid, causing it to
bring things back to 'balance'. I.E. it's NFB. Except, of course, it
does nothing if there is no error so it can't go to 'zero' and the
amount of error is dependent on mu.

There isn't much in RDH4 that I'd want to use exactly as they show.

They didn't have cheap HV silicon bjts to use as CCS in 1953.
But by 1960, ppl were dumping their tube gear and switching to SS....
In Wireless World, after 1960 almost no articles appeared relating to
tube use.
Suddenly the music died.....

I don't like the Quad-II driver/input/phase inverter pair of EF86
either.
Its got positive FB and when analysed doesn't perform nearly as well as
a
pair of EF86 set up with a long tail Rk taken to -400V, and NFB applied
to one side of the pair
and input to the other.

Walker was great on speakers. His tube amps could have and should have
been better,
but they were for the masses at home mainly. Its remarkable the BBC
bought hundreds of Quad amp systems. But then noise and distortions
from amps was a minor problem compared to other sources of N&D.



Of course, if you don't have CCSs to work with it's better than
nothing but a good LTP works as good or better.

I totally rewired an ARC VT100 which is a pig of a thing normally,
with 5 x j-fets as CCS used in a horribly complex circuit.

I remember you talking about it and still have the schematic around
here somewhere.

I revised so there were two LTPs, with the input pair with cathode sink
= MJE450 CCS taken to a -ve bias supply.
Each side of the input LPT was a paralleled 6DJ8, with R loading at
anodes.
The input goes to one side, GNFB to the other.

Yeah, at the time we debated the virtues of having a distorting
device, the 6DJ8, in the feedback path.

The way ARC have 6DJ8 set up is hard for the ordinary mortal to
undertstand.
Its been done now for so many years, and ARC remain concreted in their
ways,
and they refuse to simplify. ARC is tube over-engineering.


The driver LTP became a pair of 12BH7 each one paralleled,
with 4 x 6550 output tubes cap coupled to BH7 anodes, an the fixed bias
applied to the output grids.

The BH7 have R loads to each anode, and have a common Rk taken to -127V
bias.
The large value of Rk means common mode amplification is negligible,
while balance is excellent.
LTPs driven at one side only NEED the CCS cathode tail to get balance,
but if the grid drive is balanced its much less important, even in
Williamsons.
Placing an extra 3k3 in the Willy tail R does wonders for balance
accuracy.


The Turnerized ARC soundeds magnificent, with firm foundation, creamy,
detailed, and all aspects if
distortion and stability improved and with less total GNFB used.
No more blown fuses and distressed client who is fed up with tube
replacements
and frequent servicing.

I did a whole shirt and trouser load more things to the bloomin ARC,
but that's another story.
The yanks have forgotten how to make good reliable simple amplifiers.
ARC, McIntosh, Manely Labs and others are far too optimistic
about the longevity of their designs, and full of awkward
comprimises to allow features such as balanced or non balanced inputs,
and easy choice of output loads. Don't get me started.

Yes, well, the 'modern world' revels in complexity.


So the cascaded LTP set up works very well despite it being just
slightly
more complex than my simple LTP with SET input idea.
It used half the circuit parts that are in a VT100 circuit.
The cascaded LTP virtually eliminates all 2H which comes from input
stage
triodes operating on their own.
I've used a similar arrangement in totally re-wired Manley Snappers.
SET input stage 2H either adds or cancels slight 2H from imbalances in
an output stage
if the output tubes are not exactly matched, which is normal
in most PP amps in the real world.
So two channels can end up with very different thd and imd profiles.
I'd prefer less disimilarties between channels.



The current mirror phase splitter was to get it all in the two bottles
and, at the time, I was rather enamored with current mirrors.

The output cathode dc control is intersting, and yes, this arrangement
does keep the Idc well matched. If the amp goes into class AB,
the cathode bias voltage will rise though due to charge up effects.

Right, it has the same 'charge up' characteristic as plain ole cathode
bias but it's also self adjusting like plain old cathode bias so
there's no bias pots for the non technical to mess with.

Music signals rarely get large enough in their average value
to make Ek rise.

Yep.

But if they do, you can dynamically bypass the excess
charge up signal currents as in
http://turneraudio.com.au/schem-300w...tabilizer.html

I've seen it

I've been thinking along similar lines but these little amps I'm doing
at the moment don't really justify the added complexity... and it's
not as simple as it looks with low cathode Vs..

I have an opamp version for fixed bias that gets it down to 1 pot
(since the second side 'tracks' the first) but I haven't worked out a
completely self adjusting version for fixed bias yet.

Some form of ('automatic') output current balancing has become sort of
a 'trademark' in my PP designs with a current mirror being the most
common, so far, because of its self adjusting nature.

I explored such bias current equalization years ago
with an LTP using a pair of MJE350 in an LTP arrangement.
It worked well with two output tubes, until NFB was applied.
Then I had a good LF phase shift oscilator.

Hehe Yeah, funny things can sometimes happen

Two bypassed CCS under the cathodes, to keep balance, sounds like a
good idea till you run them into the B. Cap charge up is almost
instantaneous, and much larger, because them CCS suckers just ain't
gonna let that average current increase or decrease..

I abandoned the idea, and I sometimes use more than 2 outputs,
so the aim became have NO adjustments that will ALWAYS
confuse many owners, and yet maintain good enough regulation of Ek for
and hence Ik for all output tubes.

If a tube goes wrong, OK, an active fault detection circuit turns the
damn amp off
well before anything glows red hot.



Like the little 'PC Speaker' amp. It's in there.

http://flipperhome.dyndns.org/6AW8PCSpkr.htm

And the 6GK6 amp

http://flipperhome.dyndns.org/StealthAX.htm

The MJEs are on those boffo heatsinks

Speaking of charge up, I noticed that Broskie talked about your 'anti
charge up' circuit, congratulations, but was miffed why he spoke as if
the limited charge up was a 'problem'.

Maybe he didn't actually try out the circuit.

I dunno but, as I said, it had me miffed. He thought it was 'great'
but then 'complained' about the whole point to it. Maybe it was just a
bad hair day.

Nobody else has thought of a better way to simply allow excess Ik to
take an easy bypass route and thus prevent the Ek rise in class AB.



read my page quoted above and you'll see just how easy it is to apply to
any existing tube amp with cathode bias.

The phase inversion comes from tapping off the 'opposite end' of the
loads with the advantage being large available voltage swing.

Probably nothing you'd want to use but it works for a small 'economy'
amp.

Your idea has got me thinking though....

There's a variation on the theme that produces tons of gain but it
potentially takes adjusting (depending on just how ambitious one gets
with the gain) to center things up. Basically, run one CCS into the
plate and use the tube to 'subtract' from it. The 'left over' current
(which will be mostly signal) can then go through a very high load R
for increased gain (there's no plate feedback in the tertiary route).
But, as you can imagine, the CCS and tube have to be doing close to
the same current for things to bias up right.

That's this early one.

http://flipperhome.dyndns.org/13EM7CMPP.htm

I finally decided that, in the case of the 6EM7 anyway, it wasn't
worth the bother getting all that gain in 'one triode' since a pair of
them comes with two. hehe So, the 'simpler' current mirror phase
splitter.

Its quite easy to get HUGE voltage gain with bjt drivers.

But I like to have triodes in control of all voltage amplification.
The SS just acts as grovelling slaves to the tubes' every current or
voltage whim...

Yes, and the triode is doing the gain in that one too. It's all in the
gm, not mu, because that one bipolar holds the plate at constant V.
The CCS then subtracts out idle current, leaving just signal (plus a
little 'extra' to bias up the load R) to go across a large load R.
The gain is all 'triode gm' signal across that load R.

The current mirror then does a phase inversion but it's not part of
the gain circuit.

People would argue about that.
Its giving dumb slaves some power over a process....

A bypassed CCS under the triode might take care of 'tweaking' the
current balance and, in keeping with the 'modern' Rube Goldberg trend,
an output V to cathode bias DC servo would probably work too.

Or just use a pentode, or two triodes, and to hell with it

Endless possibilities.

Patrick Turner.

I did breadboard working versions of those, though. Still got the
little perfboard with. 'surprise', a trimmer pot to adjust the CCS.

more trim

Well, the 'self adjusting' semi CCS would be about the same complexity
as a plain ole CCS. It just, in theory anyway, substitutes a resistor
divider and bypass cap for a hard reference. And it should have good
PSSR since, at hum frequencies, it would be a CCS.

I need to think more about it all.

Let me know what you think about the MOSFET solution.

http://flipperhome.dyndns.org/Self%20Adjusting%20CCS

If time permits, there is endless juggling of possibilities....

But the clock screams at me to work, lest my bank mananger gets upset.

Patrick Turner.

[Patrick Turner]

flipper wrote:

On Mon, 30 Jun 2008 03:27:27 GMT, Patrick Turner
wrote:



flipper wrote:

On Sun, 29 Jun 2008 11:10:11 GMT, Patrick Turner
wrote:



flipper wrote:

On Sun, 29 Jun 2008 05:53:21 GMT, Patrick Turner
wrote:



flipper wrote:

On Fri, 27 Jun 2008 14:08:13 GMT, Patrick Turner
wrote:

snip

Ok, I spiced up a circuit to demonstrate the idea and posted it in
alt.binaries.schematic.electronic under the title "schematic for RAT."

I didn't work on optimizing things but spice says it works.

Cathode CCS is the one I normally use but any CCS will work.

The top side is setup so the PMOS gate is biased for the expected idle
current through the source resistor at the target anode voltage: I.E.
the resistor divider (plus gate threshold). The cap bypasses to B+ so
it will DC settle but be constant at signal F.

I originally buffered the plate with another P-MOSFET, because I had a
buffer when trying bipolar, but the 1Meg is high enough that there was
little difference, buffered or not, and increasing the resistor
divider by 10x gave less than a 1% increase in gain.

It works like you'd expect from a CCS plate load. LTP gain, with the
6BQ7, is 16, pretty close to 'ideal', and balance (remembering spice
uses identical tubes) is better than 0.5%. But that also held when I
tried subbing a 6N1P for one side to simulate 'unmatched' tubes.

The bad news is, 86% of B+ hum goes straight to the plates. Same on
both, though, so it should null in the PP but tube imbalance also
imbalances the plate hum. The CCSs don't help there.

But it's got great gain and balance.

It also works with bipolars but needs a plate buffer into the divider
because of the bipolar drive current\ and there's an odd 1/4 dB rise
(or dip, depending on which side) around 1.5kHz I didn't pin down
because it's not there with the MOSFET and that's simpler.

Finding high enough voltage P=MOSFETs might be a problem, though.

I can see the text only comments about your post at ABSE but no
schematic.

Maybe it'll turn up in a day or so, maybe it won't.

Well, that sucks.

I know it's at least there on my server because I downloaded it to
verify I didn't screw something up.

Ok, I made a quickie orphan page on my personal web server so you can
see it there.

http://flipperhome.dyndns.org/Self%20Adjusting%20CCS

'Orphan' in that there's no link to it from any other page on my site
but the above link should take you straight to it. That is, assuming
power don't glitch and bring the server down.


This looks just fine to me, one click, and it just displayed OK in
Nutscrape.


Its a nice circuit you have there,

Thanks.

although I'd simplify the cathode
current sink
by using 1 x MJE340 between the two k and a negative rail, with say 4k7
as the emitter R etc.

Yeah, as I said, any CCS will do down there. The double transistor is
just my 'standard' because I'm usually working without a negative
supply and it has the lowest voltage overhead to get under the
cathode.


What about the mosfet input capacitance?

Doesn't matter because its out of circuit at signal F.

As you have drawn it, the RL at dc at each anode = ( 1M/150k ) x 1k
approx, or about 6k ohms,
because the C becomes open at dc, and the 1m and 150k act as a divider
to
apply a slow F signal voltage to the gate and also to the 1k at the
mosfet source.

Yep. That's the 'trick' to it.

But if C is say 22uF, then at 1kHz, the R load seen by the anodes

Same with 1uF because the time constant there is the resistor divider.

becomes vitually 1M,

Yep

I tried 10x that and it made less than a 1% gain difference so 1 Meg
is 'enough'.

unless a following stage has cap coupled Rg.
These Rg can be a lot lower value than 1M, maybe 47k each, and if well
matched, the output will also
be well matched.

The only problem is if an output tube has a bit of input grid current
even at idle
due to its age or a fault, and the Rgrid-in is substantial enough to add
to the
47k biasing R so it becomes less than 47k, and voltage balance then
becomes poor.

But the effect of a crook output low Z grid input tube can unbalance
other varieties of driver/phase splitter
arrangements.

Precisely. Its not something 'special' to this one.

Besides, use good tubes

The way you have set up the CCS at anodes means that Vdc at the anodes
tends to stay constant
over a wide range of signals, because the load is low at dc, like a
choke,
where the load at dc is the wire resistance.

Yes, that was the idea. And there's nothing magic, per see, about the
1k except for MOSFET threshold variation. I.E. the higher that R is
(1k) the less percent threshold is of the total so it's effect is
less. I just arbitrarily made them roughly equal: 3mA through the 1k
and about 3V gate threshold, as a 'pick something' number. Well, it
was 3mA when I started but it's up to 3.7mA now.

On the other hand, if you make it 'real small' then the divider
becomes a large ratio, difficult to manage, C goes up, and a large
plate change would be a small divider change. 'Regulation' goes down.

I don't think any of it is real critical, though, unless you go to
extremes.

You can use your style of CCS, or rather high Z dc supply for a pentode
tube to enable it
to work with extremely high gain. The low load at dc stops the sway of
Vdc
you'll get with signal amplitude variation due to rectifying effects of
the 2H.

Open loop gains of 0ver 1,000 are possible, and if a tube with OLG gain
= 1,000
is reduced to say 10 for a line stage with shunt FB from a cathode
follower direct connected to the
pentode anode, the Rout is less than 20 ohms and THD is the open loop
value divided by 100,
so 0.2% at 2Vrms from the pentode becomes 0.002%, somewhat blame free.

Yeah, it should work for a pentode too.

Maybe I'll try that just to see a gain of 1000. hehe


A typical pentode with fully bypassed Rk such as a 6AU6
with say 4mA will have gm = about 3mA/V,
and Ra = about 500k. When I say "about", I mean approximately,
because figures are a bit variable.
µ = gm x Ra, so 0.004 x 500,000 = 2,000, so if the
anode load = say 1M,
then A = 2,000 x 1M / ( 1M + 0.5M ) = 2,000 / 1.5 = 1,333, or a lot.
If RL was closer to a a true CCS and 10M, gain approaches 2,000.

But if Ra = 500k, and RL was say 2M, then Rout = 400k.
If the stray C from anode to whatever else is coupled = 30pF,
then the HF pole is at 13 kHz.

Choke loads for pentodes don't work well at all unless large amounts of
NFB are used
because the shunt C and shunt L mean an arched shape to the F response
and there is a huge amount of iron distortion you get where the iron
coil
isn't shunted by low drive R.

The µ-follower with pentode and top tube a high µ triode such as a 12AT7
work well, and where you supply a fixed bias to the top triode
to get dc stability with changing signal levels due to
pentode 2H.
The bias R to the 12AT7 should be 2M, and the R between
triode k and pentode a should be maybe 20k, or even a transistor CCS,
and the pentode a is then cap coupled to the triode grid.
The pentode sees an anode load of 2M plus triode A x 20k in parallel,
or about 500k, and gain is still quite high at about 1000.

Shunt FB from the output side of the coupling cap off the triode
follower
is used, say R2 = 500k back to the pentode grid, with input R1 = 47k.
A' will be about 10.
The input terminal can be 100k, to bias the pentode, and Rin will
then be 47k//100k = 33k.

A triode pentode such as 6U8A can be used for a nice line stage.
But the loading of the top triode
is actually Triode Ra + ( triode µ + 1 x triode rk),
and so the the higher the triode µ, the higher the load on the pentode
below it,
and the higher the open loop gain and the greater is the amount of
applied NFB and the
more effect there is on Rout with NFB and thd/imd etc.

Just watch out for parasitic oscillations though.

A 6EJ7 plus 12AT7 paralleled makes a strong combination....

Well, as I said. I'll try it someday. Got 6AU6s and plenty of triode
pentode pairs to play with. Maybe try a 6KT8, or maybe a pipsqueak
6JW8. I keep trying to think of 'something' to use them for.

My original 'plan' for the 6KT8s was to use them in my sub watt guitar
amp but then it occurred to me it might be a bitch selecting non
microphonic pairs and there'd be no place for the ones that didn't
pass. But with two 6BQ7s I can put the 'quiet' one on the front and
use the others for the power stage.

There's a good story in that, though. When I took the amp to a local
recording studio for testing there was a loose metal plate on the
'make a pretty picture' combo cab that rattled so I removed the
chassis and stuck it on a whatever was nearby, which happened to be a
piano bench. Guy hit a chord and, good god, the feedback squeals and
howls damn near killed us.

I sent nephew off to get the 'non microphonic' tube stash in the car
but the culprit was that dern piano bench acting like one hell of a
'drum head' shaking the whole chassis to hell and back. Made even
worse because, in my haste, I hadn't bothered to put the rubber foot
plate back on.

That'll learn ya....

And consider a darlington pair connected pair of MJE350, or MJE350
with some other smaller HV P bjt for the pair.
methinks maybe there is less parasitic C with the bjts.

Well, I mentioned I tried bjts but got an odd 1/4dB 'bump' in the
frequency response at about 1.5k I haven't isolated yet. That was
using the plate buffer, though.

???


Darlingtons work just fine. I just used the MOSFET example because
it's 'simpler'.

Mosfets tend to have popcorn noise. depends where you use them though;
in high signal circuits, no worries, snr will be OK.

Hmm. Well, I hadn't considered that. No matter, both MOSFET and
bipolar work.

The shunt C between gate and source probably
isn't a worry as you suggest because if it was say 400pF,
and the Rsource = 1k, the pole is at 400kHz, and OK.
In a CCS where gate is bypassed to the rail well,
the 1k becomes a lower and reactive C load at the source,
so at extreme HF the CCS becomes a queer thing, a gyrator maybe?
Watch out for oscillations.

Remember, this isn't a 'plain' CCS. Where the gate capacitance 'pole'
is doesn't matter because there's a honker 1uF glued to it. 'HF'
response is nonexistent.

But the mosfet C is between the gate and source, so if there is a 1uF
gate to
V rail and end of source R, the Cgs begins to shunt Rs as F rises,
and C becomes a dominant reactive low Z instead of Rs.

Its not a big deal because it all happens at such a high F.




LF response is just a fractional 'hair' less with darlingtons because
you have the emitter resistor, darlingtoned up, in parallel, but
that's easy enough to make up for with a 'hair larger' C. HF response
is just a fractional 'hair' better but it's so close it's hard to pin
down why. Drain-source capacitance maybe.

In my CCS if found the MJE350 is fine on its own,
and no real need for a darlo because the hfe is around 100,
so if Ic = 5mA, Ib = 0.05mA, and not large enough
to upset normal set up much.

Well, in your normal MJE CCS it's not being biased by anode current
through a resistor divider.

Actually, I just did a quick spice check and for the 6BQ7 circuit a
single MPSA92 works just as good as darlingtons but both have about 1%
imbalance vs under 1/2% for the MOSFET. I don't know why.

bjt matching?

Well, unless I turn on Monte Carlo they're all 'identical' in spice.
And it's not turned on because I have enough problems getting
circuitmaker to finish a sim without scrambling device parameters.

I just build and measure and tweak and analyse what I have.

My brain simulates when it sleeps.

Well, when its not off rooting Kylie or someone.



Only thing I can think of is if maybe base current, small as it is,
affects balance.

hfe need to be matched for best results if there are two CCS.

I had just assumed a Darlington would be needed but the single PNP
working is just as simple as the MOSFET, it just takes a larger bypass
cap because of the effective emitter impedance. How well a particular
bjt would work probably depends on the beta. Or maybe not. I mean,
it's essentially 'out of circuit' too at signal F so as long as
collector impedance is high enough the 1 Meg dominates. Well, wait a
minute, there has to be enough gain for the 1Meg to bias it on. Ok,
that's the limiting factor. Which isn't too bad except for device to
device beta variation throwing off the effective emitter impedance and
affecting the resistor divider. So back to the Darlington unless you
want to 'trim' each one.

In my apps the gain is dominated by the following grid bias loads
which are well below any CCS or other R loads associated with CCS.

Yes, but it's not a gain issue here, it's the DC biasing. Base drive
throws off the divider ratio, which could be compensated for in the
values if it weren't for the wide variation in Hfe that changes how
much base current it pulls.

But if you take the trouble to pick a pair of bjts with close hfe,
it gets good enough for a tube LTP.



Well, it could be turned into a gain issue by fixing it 'the other
way': lowering the divider resistance so base current is negligible.

If I have a divider, I use lower than 1M and 150k

As long as the final design leads to the triode being loaded
by more than 10Ra, then the thd/imd will be as low as you'd want it
and going for better isn't worth the bother.


As the R used with the bjts or mosfets to bias them rise,
the balance from an LTP tends to drift a bit apart.

??

In my case I just like to see RL of each anode above 10 x Ra,
and once this is fullfilled, the thd becomes real low,
and making RL any higher doen't give much more thd reductions.

ok

Even with a pure CCS load, a trioded 6BQ5 or a 6SN7
will still have some thd, and you will find the
distances between Ra lines close up as you move left across thre data
sheets.
And so there will always be some 2H no matter how high RL becomes.
And some 3H, at a low level, and when you have an LTP,
the differences in gm at different Ia levels between the two triodes
mean that you get some 3H generated, and some un-cancelled 2H,
so raising RL above 10 x Ra is somewhat pointless.

Well, in that case then 100k should be more than enough for a 6BQ7 but
when I tried that in spice gain went down about 14%.

Having loads at over 10Ra is a heck of a lot better than having RL =
2Ra, or 3Ra,
like I have seen in many commercial amps.

33k for the dc carrying R to a 1/2 6SN7 is often used with Rg following
at 100k.
So RL = 25k only, and at 5mA, Ra = 10k, so RL = 2.5Ra, and thd is always
on the high side.



The other thing is that the 1M and 150k act as a NFB network at dc to
give the low value RL at DC
and hence tend to regulate the anode voltages of the triodes or any
other tube in the LTP.

Right.

The link to plate V, which gives the 'regulation', is the
'breakthrough' vs the first time I took a shot at it.

It works.


Let me know how it works out if you build one because I'm sure you're
a lot better at testing them than I am.

I will,

Patrick Turner.

trim

[Ian Iveson]

"flipper" wrote in message
...
On Sun, 29 Jun 2008 12:08:12 GMT, Patrick Turner
wrote:



flipper wrote:

On Sun, 29 Jun 2008 05:41:41 GMT, Patrick Turner
wrote:

snip it down


yes.

With a CCS cathode sink and choke and some series R
to anodes, the same thing occurs; the B+ rail must be
very well filtered
and low impedance lest the rail noise is applied in
common mode to each
grid of a following stage.

Ok, I thought you were saying the CT choke rejected B+
noise.

It doesn't. It cannot.

Well, that's what I thought but then it's always possible
I missed
something.


And in an a PP output stage, if there is a lot of PS noise
at the CT, it
is applied
to each anode of the output stage.
Peter Walker didn't think this mattered in Quad-II; there
is 17Vrms
of Vripple at the OPT, bleedin terrible, but the Ra
looking into
the anodes is high, and the change in tube gm is low,
so PS caused artifacts don't exceed THD. But they are just
as much
though in class A and
worse in class AB.

In a triode output stage, if there is substantial noise at
an OPT CT,
the LOW Ra of both tubes is subject to the Vripple,
and so the Iripple flow in each output tube is substantial

Yeah. having now switched to a triode, vs pentode, output
stage on
this latest amp I realized that.

Ain't no 'free lunch' and triodes aren't a miracle cure
for
everything.

and this substantially changes the gm and the IMD is
severe as a result.

Conclusion?

Use HUGE C values to anchor down the dc supply at CT
chokes and OPTs
everywhere you use them.

Provide filtering to keep Vripple at CT at less than 30mV,
and at early input stages far lower levels.

SE amps need even stricter B+ filtering, expecially SET
amps with little
NFB.



Well, my little MOSFET gimmick works a lot like
independent chokes
except no inductive swing, of course.

No see saw effect.

True, but I'm not convinced there's 'see saw' in the CT
choke either.
The whole point is to have impedance so high that current
becomes
negligible so what's to 'see saw' in the iron?

The benefit is balanced, net zero, Idc.

But if the RL loads to each side of the LTP remain equal
dynamically then the circuit should balance very well.



http://flipperhome.dyndns.org/Self%20Adjusting%20CCS

I'm murky on how you were doing the CCS 'choke'
substitute because
you've got a CCS under the cathodes.

If it acts like a choke the Cathode sink is OK.

A CCS doesn't act like a choke.

It does in that the impedance seen by the tube is high
across a range of
AF
for both a choke load or CCS, so gain will approach µ
and THD will be
minimised.
The CCS is better in that it does not generate iron
caused distortions.

Well, I tend to think of a CCS as akin to a 'big value
resistor',
rather than a choke, because it operates down to DC.
That's the
'difference' in the new circuit.

Any CCS rarely has to actually be a pure CCS, with
unmeasurably high
resistance.
Stray C within the device won't let you have huge Z.

That's not a 'low impedance' at LF issue.

But the CCS is an avenue where the ac power wasted in a dc
supply R can
be avoided
almost entirely, and the tubes effectively loaded by a
much higher load
value than if only R were used,
and Idc at idle can be higher, and a better working point
used for
llower thd/imd.

Yes, I know. But, see? You speaking of replacing a 'big R'
(and a 'big
V' so you can have the 'big R') with the CCS.

And the Rg used fr the following stage can be much lower,
thus much
better regulating
Eg, which all too easily can go too positive when you
don't want it to
as the output or other tubes age.



Having CCS at the cathode, and at each anode kinda
doesn't work quite as
expected.

Is that what you were trying? Dual CCS anode loads
along with a
cathode CCS?

I tried that once, but dc stability isn't as good as
with R loads
everywhere.
But with an Rk from common cathodes to 0V of at least
about 3k for
two 6SN7 halves, balancing and 2H cancelling is better
than if Rk was
say 470 ohms.

ok.

more trim

Well, if you want a current mirror phase splitter
look at my "Looking
Glass" amp.
http://flipperhome.dyndns.org/Looking%20Glass.htm

Interesting what you have done there.

Thank you. That means a lot to me.

I'll have to analyse it.
I would have used a conventional tubed LTP with R loads
only
because the R loads can be high enough to have the
tubes operating
at low THD.

Yes, and I did that on the 13FD7 amp I'm working on
right now. Well,
not really an LTP, it's like a Williamson. Unit 1 in the
13FD7s are
the common cathode voltage multipliers and I added a
6BQ7 for the
front end triode and split load phase splitter.


So you have cascaded LTPs for the input / driver amp?

No, it's a Williamson. There's only 1 'pair' and they
don't have a
long tail, just a common Rk. That does a little 'see saw'
but the
primary benefit is obviating the need for Rk bypass caps
so you save
one resistor and two caps (vs independent Rks).

The front two are gain triode into a split load
(concertina,
cathodyne, take your pick) phase splitter.

Now, on the 6GK6, triode mode, PP amp that I plan to get
back around
to 'one of these days' I do have double, cascaded, CCS
'LTP' stages
(it's also fixed bias with MOSFET buffered +ve drive and
the opamp
output current balance). It's 'PP' all the way through but
I didn't do
that with the 13FD7 as it's supposed to be 'simple'

Btw, I did find at least one of the 'self balancing'
circuits I was
thinking of. It's in RDH4 section 12 under phase
splitters: "See-saw
self-balancing phase inverter" But it doesn't have any
better balance,
about 7%. Plates are cap coupled to equal resistors
(1Meg), the
joining node of which is taken to the opposing grid so any
'difference
between the two plate signals is applied to that grid,
causing it to
bring things back to 'balance'. I.E. it's NFB. Except, of
course, it
does nothing if there is no error so it can't go to 'zero'
and the
amount of error is dependent on mu.

Of course, if you don't have CCSs to work with it's better
than
nothing but a good LTP works as good or better.

I totally rewired an ARC VT100 which is a pig of a thing
normally,
with 5 x j-fets as CCS used in a horribly complex circuit.

I remember you talking about it and still have the
schematic around
here somewhere.


I revised so there were two LTPs, with the input pair with
cathode sink
= MJE450 CCS taken to a -ve bias supply.
Each side of the input LPT was a paralleled 6DJ8, with R
loading at
anodes.
The input goes to one side, GNFB to the other.

Yeah, at the time we debated the virtues of having a
distorting
device, the 6DJ8, in the feedback path.

The driver LTP became a pair of 12BH7 each one paralleled,
with 4 x 6550 output tubes cap coupled to BH7 anodes, an
the fixed bias
applied to the output grids.

The BH7 have R loads to each anode, and have a common Rk
taken to -127V
bias.
The large value of Rk means common mode amplification is
negligible,
while balance is excellent.
LTPs driven at one side only NEED the CCS cathode tail to
get balance,
but if the grid drive is balanced its much less important,
even in
Williamsons.
Placing an extra 3k3 in the Willy tail R does wonders for
balance
accuracy.


The Turnerized ARC soundeds magnificent, with firm
foundation, creamy,
detailed, and all aspects if
distortion and stability improved and with less total GNFB
used.
No more blown fuses and distressed client who is fed up
with tube
replacements
and frequent servicing.

I did a whole shirt and trouser load more things to the
bloomin ARC,
but that's another story.
The yanks have forgotten how to make good reliable simple
amplifiers.
ARC, McIntosh, Manely Labs and others are far too
optimistic
about the longevity of their designs, and full of awkward
comprimises to allow features such as balanced or non
balanced inputs,
and easy choice of output loads. Don't get me started.

Yes, well, the 'modern world' revels in complexity.


So the cascaded LTP set up works very well despite it
being just
slightly
more complex than my simple LTP with SET input idea.
It used half the circuit parts that are in a VT100
circuit.
The cascaded LTP virtually eliminates all 2H which comes
from input
stage
triodes operating on their own.
I've used a similar arrangement in totally re-wired Manley
Snappers.
SET input stage 2H either adds or cancels slight 2H from
imbalances in
an output stage
if the output tubes are not exactly matched, which is
normal
in most PP amps in the real world.
So two channels can end up with very different thd and imd
profiles.
I'd prefer less disimilarties between channels.



The current mirror phase splitter was to get it all in
the two bottles
and, at the time, I was rather enamored with current
mirrors.

The output cathode dc control is intersting, and yes,
this arrangement
does keep the Idc well matched. If the amp goes into
class AB,
the cathode bias voltage will rise though due to charge
up effects.

Right, it has the same 'charge up' characteristic as
plain ole cathode
bias but it's also self adjusting like plain old cathode
bias so
there's no bias pots for the non technical to mess with.

Music signals rarely get large enough in their average
value
to make Ek rise.

Yep.

But if they do, you can dynamically bypass the excess
charge up signal currents as in
http://turneraudio.com.au/schem-300w...tabilizer.html

I've seen it

I've been thinking along similar lines but these little
amps I'm doing
at the moment don't really justify the added complexity...
and it's
not as simple as it looks with low cathode Vs..


I have an opamp version for fixed bias that gets it down
to 1 pot
(since the second side 'tracks' the first) but I haven't
worked out a
completely self adjusting version for fixed bias yet.

Some form of ('automatic') output current balancing has
become sort of
a 'trademark' in my PP designs with a current mirror
being the most
common, so far, because of its self adjusting nature.

I explored such bias current equalization years ago
with an LTP using a pair of MJE350 in an LTP arrangement.
It worked well with two output tubes, until NFB was
applied.
Then I had a good LF phase shift oscilator.

Hehe Yeah, funny things can sometimes happen

Two bypassed CCS under the cathodes, to keep balance,
sounds like a
good idea till you run them into the B. Cap charge up is
almost
instantaneous, and much larger, because them CCS suckers
just ain't
gonna let that average current increase or decrease..

I abandoned the idea, and I sometimes use more than 2
outputs,
so the aim became have NO adjustments that will ALWAYS
confuse many owners, and yet maintain good enough
regulation of Ek for
and hence Ik for all output tubes.

If a tube goes wrong, OK, an active fault detection
circuit turns the
damn amp off
well before anything glows red hot.



Like the little 'PC Speaker' amp. It's in there.

http://flipperhome.dyndns.org/6AW8PCSpkr.htm

And the 6GK6 amp

http://flipperhome.dyndns.org/StealthAX.htm

The MJEs are on those boffo heatsinks

Speaking of charge up, I noticed that Broskie talked
about your 'anti
charge up' circuit, congratulations, but was miffed why
he spoke as if
the limited charge up was a 'problem'.

Maybe he didn't actually try out the circuit.

I dunno but, as I said, it had me miffed. He thought it
was 'great'
but then 'complained' about the whole point to it. Maybe
it was just a
bad hair day.


read my page quoted above and you'll see just how easy it
is to apply to
any existing tube amp with cathode bias.

The phase inversion comes from tapping off the
'opposite end' of the
loads with the advantage being large available
voltage swing.

Probably nothing you'd want to use but it works for a
small 'economy'
amp.

Your idea has got me thinking though....

There's a variation on the theme that produces tons of
gain but it
potentially takes adjusting (depending on just how
ambitious one gets
with the gain) to center things up. Basically, run one
CCS into the
plate and use the tube to 'subtract' from it. The 'left
over' current
(which will be mostly signal) can then go through a very
high load R
for increased gain (there's no plate feedback in the
tertiary route).
But, as you can imagine, the CCS and tube have to be
doing close to
the same current for things to bias up right.

That's this early one.

http://flipperhome.dyndns.org/13EM7CMPP.htm

I finally decided that, in the case of the 6EM7 anyway,
it wasn't
worth the bother getting all that gain in 'one triode'
since a pair of
them comes with two. hehe So, the 'simpler' current
mirror phase
splitter.

Its quite easy to get HUGE voltage gain with bjt drivers.

But I like to have triodes in control of all voltage
amplification.
The SS just acts as grovelling slaves to the tubes' every
current or
voltage whim...

Yes, and the triode is doing the gain in that one too.
It's all in the
gm, not mu, because that one bipolar holds the plate at
constant V.
The CCS then subtracts out idle current, leaving just
signal (plus a
little 'extra' to bias up the load R) to go across a
large load R.
The gain is all 'triode gm' signal across that load R.

The current mirror then does a phase inversion but it's
not part of
the gain circuit.


I feel it's time to question again the idea that, if a
component or sub-circuit contributes no gain, it can have no
effect on the signal content. I know you haven't quite said
that, and maybe it's a bad way of stating what seems to me
to be a widespread notion. Perhaps you could express your
own underlying logic yourself? Maybe I should start a thread
on what might be acceptable limits to hybridisation in the
context of a valve group?

At a guess it would depend on what you mean by "gain". For
example, if I posited a non-linear unity-gain buffer as a
counter-example, I suppose you would argue that if it is
non-linear, then it cannot be unity-gain. But then it seems
to me that the original notion becomes a tautology.

Anyway, the difficulties you are experiencing in defining
the requirements of Patrick's idea reflect a real nonsense.
However defined, I think you will find it has serious
shortcomings.

What Patrick appears to think he wants is a device which
will reflect a change of current down one leg, producing an
equal and opposite change in current up the other. This is
good for PSRR but disastrous for CMRR. Seems to me that a
with a differential mode signal the anodes would see no
load. If you add load resistors then surely you have to
wonder why not just use the resistors on their own?

OTOH, if you use a current mirror that reflects changes of
current down one leg, down the other leg in the same
direction, then the circuit would have excellent CMRR and
differential mode gain, load the valves lightly, but have no
CMRR. How that would work in conjunction with a CCS at the
common cathodes might take some thought.

In both cases, when you factor in the impedances presented
by next stage, the apparent advantages may disappear. The
output impedances of a LTP, under various load and drive
conditions, aren't especially simple at the best of times.

Perhaps if Patrick were to state his objectives in terms of
performance criteria for the LTP as a whole, we could all
help him design something simple?

Somewhere you've said that transformers reflect change in
voltage. True as far as it goes, but voltage-controlled
voltage sources won't work here. What you also need is
reflected impedance, across which a voltage may be
developed. An impedance and a change in voltage together go
to make a change in current, so maybe that's what needs
reflecting: di/dt. At least in Patrick's mind. I don't think
it makes sense anyway.

Ian

A bypassed CCS under the triode might take care of
'tweaking' the
current balance and, in keeping with the 'modern' Rube
Goldberg trend,
an output V to cathode bias DC servo would probably work
too.

Or just use a pentode, or two triodes, and to hell with it


I did breadboard working versions of those, though.
Still got the
little perfboard with. 'surprise', a trimmer pot to
adjust the CCS.

more trim

Well, the 'self adjusting' semi CCS would be about
the same complexity
as a plain ole CCS. It just, in theory anyway,
substitutes a resistor
divider and bypass cap for a hard reference. And it
should have good
PSSR since, at hum frequencies, it would be a CCS.

I need to think more about it all.

Let me know what you think about the MOSFET solution.

http://flipperhome.dyndns.org/Self%20Adjusting%20CCS

If time permits, there is endless juggling of
possibilities....

But the clock screams at me to work, lest my bank mananger
gets upset.

Patrick Turner.

[Patrick Turner]

Ian Iveson wrote:


snip,



I feel it's time to question again the idea that, if a
component or sub-circuit contributes no gain, it can have no
effect on the signal content. I know you haven't quite said
that, and maybe it's a bad way of stating what seems to me
to be a widespread notion. Perhaps you could express your
own underlying logic yourself? Maybe I should start a thread
on what might be acceptable limits to hybridisation in the
context of a valve group?

At a guess it would depend on what you mean by "gain". For
example, if I posited a non-linear unity-gain buffer as a
counter-example, I suppose you would argue that if it is
non-linear, then it cannot be unity-gain. But then it seems
to me that the original notion becomes a tautology.

Anyway, the difficulties you are experiencing in defining
the requirements of Patrick's idea reflect a real nonsense.
However defined, I think you will find it has serious
shortcomings.

What Patrick appears to think he wants is a device which
will reflect a change of current down one leg, producing an
equal and opposite change in current up the other. This is
good for PSRR but disastrous for CMRR. Seems to me that a
with a differential mode signal the anodes would see no
load. If you add load resistors then surely you have to
wonder why not just use the resistors on their own?

OTOH, if you use a current mirror that reflects changes of
current down one leg, down the other leg in the same
direction, then the circuit would have excellent CMRR and
differential mode gain, load the valves lightly, but have no
CMRR. How that would work in conjunction with a CCS at the
common cathodes might take some thought.

In both cases, when you factor in the impedances presented
by next stage, the apparent advantages may disappear. The
output impedances of a LTP, under various load and drive
conditions, aren't especially simple at the best of times.

Perhaps if Patrick were to state his objectives in terms of
performance criteria for the LTP as a whole, we could all
help him design something simple?


I began this thread asking what SS device circuit acts just like a
ct choke, ie, like a circuit in a black box with 3 terminals,
and which measures and behaves just like a choke with a CT, but which
isn't a choke.

Nobody has yet come up with a solution.

It doesn't matter if there are power supply feeds to the box.

So whatever devices are in the box, they have to mimic the
magnetic coupling between the two winding halves each side of the CT.


The LTP should with CT choke eqivalent circuit or with a real choke
should have wide bandwidth from 5 to 55kHz at least
and can have shelved responses extending further is series R
are added from anodes to the live ends of CT choke/equivalent.
Loading by the equvalent choke circuit should be
like a very high impedance anode to anode and equivalent
to say 1M a-a at 1kHz.
DC must be able to be brought to the anodes from the B+ through the
equivalent circuit.

Balance is maintained partially because of :-
1. The magnetic coupling or equivalent
coupling action of the equivalent circuit action.
2. the accuracy of the capacitor coupled grid bias resistors
in the following stage which will be the dominant load resistance value
so that the
total load value on each side of the LTP at 1kHz is at least 10 x Ra of
one of the
LTP triodes.

Balance will be good with both actions occuring.

The LTP should be able to swing 100Vrms at less than 1.0% THD, nearly
all 3H.

The CMRR of noise from the rails does not have to be good because
it is assumed that rail noise will be less than .05mV,
and the impedance anchoring the B+ above the anodes
is 235uF at least. ( 6.8 ohms at 100Hz )

The LTP should not amplify common mode signals applied to each LTP grid
input and this will be ensured by use of a CCS from the commoned
cathodes to a negative supply voltage rail.

I think I might have covered the aims behind the questions
so that everyone here might benefit from
an applied solution, and get less THD and IMD and better music
and without reliance on global NFB.

Patrick Turner.

[Ian Iveson]

"Patrick Turner" wrote in message
...


Ian Iveson wrote:


snip,



I feel it's time to question again the idea that, if a
component or sub-circuit contributes no gain, it can have
no
effect on the signal content. I know you haven't quite
said
that, and maybe it's a bad way of stating what seems to
me
to be a widespread notion. Perhaps you could express your
own underlying logic yourself? Maybe I should start a
thread
on what might be acceptable limits to hybridisation in
the
context of a valve group?

At a guess it would depend on what you mean by "gain".
For
example, if I posited a non-linear unity-gain buffer as a
counter-example, I suppose you would argue that if it is
non-linear, then it cannot be unity-gain. But then it
seems
to me that the original notion becomes a tautology.

Anyway, the difficulties you are experiencing in defining
the requirements of Patrick's idea reflect a real
nonsense.
However defined, I think you will find it has serious
shortcomings.

What Patrick appears to think he wants is a device which
will reflect a change of current down one leg, producing
an
equal and opposite change in current up the other. This
is
good for PSRR but disastrous for CMRR. Seems to me that a
with a differential mode signal the anodes would see no
load. If you add load resistors then surely you have to
wonder why not just use the resistors on their own?

OTOH, if you use a current mirror that reflects changes
of
current down one leg, down the other leg in the same
direction, then the circuit would have excellent CMRR and
differential mode gain, load the valves lightly, but have
no
CMRR. How that would work in conjunction with a CCS at
the
common cathodes might take some thought.

In both cases, when you factor in the impedances
presented
by next stage, the apparent advantages may disappear. The
output impedances of a LTP, under various load and drive
conditions, aren't especially simple at the best of
times.

Perhaps if Patrick were to state his objectives in terms
of
performance criteria for the LTP as a whole, we could all
help him design something simple?


I began this thread asking what SS device circuit acts
just like a
ct choke, ie, like a circuit in a black box with 3
terminals,
and which measures and behaves just like a choke with a
CT, but which
isn't a choke.

Nobody has yet come up with a solution.

It doesn't matter if there are power supply feeds to the
box.

So whatever devices are in the box, they have to mimic the
magnetic coupling between the two winding halves each side
of the CT.


The LTP should with CT choke eqivalent circuit or with a
real choke
should have wide bandwidth from 5 to 55kHz at least
and can have shelved responses extending further is series
R
are added from anodes to the live ends of CT
choke/equivalent.
Loading by the equvalent choke circuit should be
like a very high impedance anode to anode and equivalent
to say 1M a-a at 1kHz.
DC must be able to be brought to the anodes from the B+
through the
equivalent circuit.

Balance is maintained partially because of :-
1. The magnetic coupling or equivalent
coupling action of the equivalent circuit action.
2. the accuracy of the capacitor coupled grid bias
resistors
in the following stage which will be the dominant load
resistance value
so that the
total load value on each side of the LTP at 1kHz is at
least 10 x Ra of
one of the
LTP triodes.

Balance will be good with both actions occuring.

The LTP should be able to swing 100Vrms at less than 1.0%
THD, nearly
all 3H.

The CMRR of noise from the rails does not have to be good
because
it is assumed that rail noise will be less than .05mV,
and the impedance anchoring the B+ above the anodes
is 235uF at least. ( 6.8 ohms at 100Hz )

The LTP should not amplify common mode signals applied to
each LTP grid
input and this will be ensured by use of a CCS from the
commoned
cathodes to a negative supply voltage rail.

I think I might have covered the aims behind the questions
so that everyone here might benefit from
an applied solution, and get less THD and IMD and better
music
and without reliance on global NFB.


I have posted transformer equivalent circuits here before a
couple of times, together with some explanation and
demonstration simulations. You didn't like them then so I
don't suppose they will help you now. They were in any case
constructed from ideal components, coz I don't know how to
do that SS stuff.

What you want in each leg is a current or voltage-controlled
*voltage* source, and a current sensor of low resistance.
Connecting the current sensor of one side to the voltage
source of the other, you need a differentiator, which can be
done using either a coil or a capacitor. Alternatively, you
may sense the rate of change of current directly with a
small coil in each leg, and AC-couple that signal to control
the voltage source in the other leg; an arrangement that
could be described as an amplified choke (common or
differential mode, depending on whether or not you invert
the control signals), and might be a good bet.

Note that the use of voltage sources and current sensors
gives you the desired common-mode next-to-no-impedance, and
the circuit has no problem with adjustments to the tail CCS
DC current.

Naturally you need some gain to get the dependency of the
right magnitude and reduce error with lots of local
feedback. That raises questions about the stability of the
arrangement, particularly within the constraints imposed by
the CCS at the tail.

However, once you've got what you think you want (not too
hard from
the above description...maybe, er, "flipper" could knock one
up in a jiffy), you will find it doesn't do everything you
hope, and does some things you may not yet have feared: my
contention remains that it's a daft idea
from the start, and was when you made that amp you sold,
which didn't work very well. Your prescription reads like an
election leaflet for the Party of Sweetness and Light.

Ian