I'd try AC heating first.

With 813 in SE, perhaps you could run
a cathode feedback winding for local FB, using between 12% and 20%
of the primary turns in a winding devoted to the cathode.
I suspect the tetrode class A performance probably won't be very linear.

I had thought about that after reading your earlier article/post on
that matter.

I have found someone in Auckland who can wind transformers for me. I
have a pair of SE 6k trannies with an Ultralinear tap on it.



Ordinary UL would be difficult because of the screen rating, so better to take
the screen
to the lowest fixed voltage without compromising the anode voltage swing.
This means you may have 900v anode supply, but
perhaps only 350v screen supply, and this means the grid bias voltage needed is
lower.

I have been looking at the specs for the 813 and the max screen
voltage is 750V [ this is from the AM modulator /power amp ratings].
so using the UL should not be such a problem.

The curves in pentode mode the curves look ugly, and if listening to
the breadboard model that i have at the moment is anything to go by,
they are really ugly to listen to as well.

So i think that I will be following pete Milletts lead on this and
going with triode strapping. If you go to diyaudio.com and do a
search on 813 in the tube section you will come across a set of triode
curves which are absolutely gorgeous. Mind you the power goes down,
but who cares, 21W vs 30 odd, it is stil too loud.


For SE, I quite like cathode bias which sets itself reliably, and in use it
does not drift.

I quite like the idea of cathode bias as well, However, I was not
keen on having too high a voltage hanging around the hum pots etc.

I have doodled using two pots, one across the filament to null the ac
ripple, then taking the wiper to ground through a resistor and
through another 100 ohm pot so i could vary the bias. I will strap
the wiper to the top of the pot so if it ever comes off, the cathode
will lift higher, driving the tube out of conduction. Time will tell,
it is much simpler and this method ages well as it were. You lose
valuable volts tho. I have plenty of time to try both.


With my SE amps using 4 x 6CA7 in parallel, I have 4 separate RC, one for each
tube's cathode circuit,
consisting of 240 ohms and 1,000 uF bypass cap.
These four RC networks are joined commonly to the top of the feedback winding
from the OPT.

In your case, there is nothing to stop you connecting the CT of the filament
winding
to the top of an RC network for biasing which can be taken to a FB winding at
the bottom,
or just ground the bottom of the RC network if pure terode is tried.

The FB winding will reduce the gain of the 813 considerably, so more drive is
required,
but the benefit is that the tube with CFB acts like triode, ie,
thd will be dramatically reduced, and its spectral content shoulod be better
than in pure tetrode, and the Ro should be a couple of ohms at the OPT sec.

The cathode FB does not only work to provide series voltage negative feedback
in series with the grid voltage.
The distortion of the tube is not only fed back inverted in phase to the grid,
but also to the screen, since there is a signal difference between screen and
cathode with
CFB, unless the screen is bypassed with a large cap to the cathode
instead of to ground.

Basically, using CFB the way I am humbly suggesting is similar to
having 1/2 a Quad II circuit.

Where the screen is bypassed to the cathode, then the cathode circuit
sees the impedance of the dropping resistor from the B+ to the
screen supply, so the value of R dropper shouldn't be too small.
Anything over a kohm would be OK.

Where the screen is fully bypassed to the cathode, the tube is operating in
pure beam
tetrode mode.
Where CFB is used in this case, the spectra of harmonics produced is similar to

plain beam tetrode without any CFB, although its reduced by the CFB,
to an amount about equal to the amount of gain reduction.
The Ro due to CFB is usually slightly more with a bypassed screen to cathode.
But the simplest way, and the best way, imho, is to use the screen
bypassed to 0V.

You should also find that the driver tube may have to work a little harder with
CFB,
but then you also find that the 2H of the driver will more likely
tend to cancel the 2H of the output tube, which is a bonus, imho.


I guess you may be aiming to use about 8k for a load, and get a swing of nearly
800v peak,
which would give you just on 40 watts.
Grid bias seems to be only about -15v required for 90mA at 900v.
Since Gm is about 4 mA/V, then the tetrode gain into
8 k would be about 32 times, so for 565vrms of plate signal,
some 17.65 volts at the grid is required, and this exceeds the
negative bias voltage, so only class A2 seems possible for full power.

Youi know i thought that the grid bias on the breadboard that i have
is all wrong, tis currently sitting at about -90v, which means that
the signal does not have far to go when the tube goes out of
conduction. in fact when the signal level goes too high you can see
something is wrong, it distorts terribly, and the ammeter in the
cathode, [voltmeter across a sense resistor] starts to wobble. ie.
no longer class A.


If there was 12.5% CFB, ( 1/8 of the total primary turns, or three layers of 24
layers of wire
in the primary, ) then the CFB voltage would be 70v, and plate voltage 494v,
and the grid input would be about 90 vrms, which sounds like a lot,
until you work out a decent way to make 90vrms at low thd.

Now it is funny you should say that, I am/have also played with some
IXYS high voltage constant current reglators, theser are trickty
things, once used one as a load of a SE 807 with a parafeed PP
oputput tranny. Wasn't very loud, only a watt or two but it proved
the point. Too high plate voltages for such measly power tho. the CC
reg that i am working with can withstand 900v, conduct 250ma [these
are at the extremes you realise] but they can only dissipate 20W. and
that does not bear close scrutiny either, you would need a heatsink
the size of the queen mary. possibly water cooled. anyway, i am
looking to use the good oly 6sn7 with a ccs plate load, according to
pete milletts site, it is good for 90 volts rms, with distortion
obviously, but mainly 2H.



The anode resistance of the pure tetrode 813 looks very high from the valve
curves,
because the plate resistance line is nearly horizontal at 90 mA of bias
current,
and my guess the Ra = approximately 40 kOhms.
Because Gm = about 4 mA/V, the tube U is around 160 at the bias point
proposed.
If the screen was bypassed directly to the cathode, you'd
have 18vrms needed between G1 and K to make the tube produce the
565 A to K signal voltage, and there would be the 70v at the K.
The fraction of the output fed back in series with the G1 signal
is 12.5%, or 1/8, ie, Beta, B, is 0.125.


yup, not nice, so i will be sticking with the triode curves.


From this, knowing U is 160, and knowing Ra = 40,000, we can work
out the effective Ra after feedback is applied.
Ra' = Ra / ( 1 + [U x B] ) = 40,000 / ( 1 + [ 160 x 0.125 ] ) = 40,000 / 21
= 1.9k.

If the OPT has a Z ratio of 8k to 5 ohms, or 1,600 to 1,
then the effective Ra' appears at the secondary as
1,900 / 1600 ohms, plus the secondary winding resistance,
so expect to see Ro at the sec = 1.5 ohms.
Bypassing G2 to 0V instead of to the K will give a figure slightly
higher than the pure tetrode with FB case, but methinks the spectra
with bypassing to 0V is better.
Even 10% of CFB would be very beneficial .

The effectively applied amount of CFB varies with load value,
since thre gain of a tetrode varies almost directly as the load value
so that where the load is halved, so is the gain,
but with the CFB, the variations in applied grid voltage are
very much reduced for different load values.
So the driver does not have to produce sucha wide range of voltages
for different loads, and perhaps there is less imd produced in the driver as a
result if global FB is also applied to the amp to reduce Ro
below 0.5 ohms.
I found that in my amps with 6CA7, only about 8 dB
of global FB was needed in addition to the CFB.

In my case the 4 paralleled 6AC7 tubes
gave me a composite tube where Gm = about 40 mA/V,
Ra = 3k, U = 130, so with
B = 0.125, Ra' = 173 ohms, and since the opt has a Z ratio
of 246 :1, Ro = about 0.7 + Rw = 1ohm, approx,
for where G2 was bypassed to the K, but
I have bypassed G2 to 0V, and Ro is about 1.6 ohms.
With global FB, it is about 0.6 ohms.

In my amps the OPT has a primary load of 1,200 ohms,
so the leakage inductance needs to be 6.6 times lower than a tranny
with a load of 8k, for the same HF response.
This is easy to achieve with interleaving, and there are only 1,904 turns on
the primary
on a 60 stack of 44 tongue material.
The maximum signal voltage across the P winding from A to K is about 220vrms.
With an 813, there will be about 565 vrms, so
the P turns would have to be 4,889, if the core was the same size.
In other words, 2.56 times the turns I am using.
With the same winding interleaving geometry, this means the LL
will be 6.6 times that inmy tranny.
There will also be 6.6 times the primary inductance, which is a good thing,
because the relationship between P inductance to load
should be maintained so ZLp = RL at 20 Hz.
In my case, RL is 1.2k, and I have over 10H of Lp.
But with 813, and RL = 8k, you'd need 66H to get the same LF pole I am getting.

I would use a 75 stack of 44 tongue for the OPT for tubes like the 813,
or 845, GM70, 211, etc.
This allows the turns to be reduced, and thus the wire to be thicker,
and the winding losses to be lower, since getting copper losses below
10% isn't all that easy with SE amps.

Thanks for this, I have been playing with transformer designs as well,
but nearly always gave up as i did not have the confidence to build
one. Now that I have found someone and seen there work [only last
night on the breadboard] i will probably do something about it. I am
not confident on the OP{ tranny that i have at the moment now that I
understand the design process that went into it. [goes something like
this, VA rating of the core for equivalent power tranny used, windings
something like an old design, no-one knows anything about the maths,
only the construction]. So my part in this is to get the maths right,
then the design and get someone else to build it. Once i have my
breadboard rebuilt, if the transformer sucks, i will bounce my
design ideas off you, ok?


I am not advising anyone on what they must do, or ought to do,
but just trying to explain some of the issues involved, and leaving options
open.

For example, an SE amp with 813 in pure tetrode could be set up
with nothing but global FB, and to equal the above figures using local CFB
and a small amount of global, some 20 dB of global FB has to be used
to get Ro down from 25 ohms without any FB whatsover
to about 0.5 ohms.
If this is done, the thd for the 20 dB of GFB will be lower than the combined
FB method
because the driver stage is also fully enclosed by the FB loop,
and the drive voltage applied to the output tube is much lower.

I had been looking to use the 811A [the real one], and global feedback
would most certainly have been necessary, the curves are pretty
cramped. the second deal killer was the need for input power as the
811 operates in A2. the upside is that it only needs about 20Vrms to
get full signal out of it, approx 16W, so it is not that bad. again
the transformer was peculiar, approx 4k, i would have a CFB in that
to help with the distortion taming.

thanks for the post.

will keep you posted on progress.

This breadboard i have at the moment belongs to another chap who has
built one for himself, I am going to construct one from scratch
myself this weekend, using his op transformers.

I will try and take pictures.

kind regards

bill





Whether it sounds as well is a moot point.

Patrick Turner.





Kind regards

bill ramsay