In 1997, I built the 22W SEUL amps using a single 13E1 in single ended
ultralinear mode. The details are well covered in my web page on the SEUL 22W.

The SEUL amp pleased anyone lucky enough to hear music piped through it.
 
Since 1997, I have increased my experience of using local negative feedback
in amplifier output stages. A customer of mine who bought a pair of
SEUL 22W amps had borrowed another customer's SE35 amps and he
thought the SE35 to be slightly more accurate and detailed.
It isn't uncommon for audiophiles to lend their amps to others occasionally.
I wondered if any better  sonic and technical performance could be had from
the 13E1, and I had suspected it to be possible ever since 1997 but had not
fully explored the possibilities and practicalities.
After the SEUL 22 with 13E1, the next high power SE amp I built was the SE35
using a quad of parallel 6CA7/EL34 and fully detailed at my page at
SE35W monoblocs.
My customer with SEUL 22 has always found that other projects I have built
for him resulted in a worthwhile and pleasing outcome so he went ahead with
the change from the ultralinear operation with screen feedback from  tap on
the OPT anode primary winding to having the primary divided into two windings
with 66% of turns for the anode and 33% of turns for the cathode for applied
cathode feedback.
He had also purchased a pair of my Sublime speakers, also described in my
website page on loudspeakers.
The original SEUL amp was in fact capable of about 22W into 8 ohms and about
25W into 4 ohms. But with 4 ohms there was more than twice the THD than
with 8 ohms and because the Sublimes had an impedance of about 5 ohms average,
it was thought a change to the output transformer ratio would give a much better
load match to the 13E1 and thus reduce the distortion and give a higher maximum
output power of 32 Watts because of increased anode efficiency with a much
lower screen dissipation.
The sound of the new amp circuit is very clear and natural, but never clinical or
blandly cold, and conveys the recorded warmth of a real live performance to
give high emotional engagement with music that is the hallmark of a good
tubed system. Bass is tight and gives the music its foundation, treble is sweet,
with midrange that is glorious without being euphonic.
Rather than wade through the changes to the SEUL22W schematic in a laborious
discussion, I will simply provide the SE32 schematic I used and explain how it
works, with some provisos and notes about limitations etc.
People are then free to compare the SEUL22W to the SE32 schematic,
and are free to adopt the principles of the operation.
Tube amp design is somewhat flexible.

Fig1.

Fig1 shows the audio circuit with input V1 6SL7, driver V2 EL34 in triode,
and output V3 13EI.
V1 input stage, 6SL7.
C1&R1 form a high pass filter with pole at 7.2Hz to keep out dc or very
low F signals. V1 Input stage signal is applied to the 6SL7 grids.
There is a very mild amount of 9dB of global negative feedback from OPT
secondary applied to the cathode via FB resistance divider, R5 and R11,R12.
The voltage difference between the grid input signal and cathode feedback
signal is amplified 47 times by the 6SL7 and applied to the network beginning
with C5.
The network after C5 has a shelved response at LF and HF to reduce the 6SL7
gain and phase shift at frequencies where otherwise oscillations might occur
below 10Hz or above 60kHz because of the use of the global NFB.
The 6SL7 is among the world's most linear triodes and easily produces the
15Vrms at very low THD required by the next EL34 driver stage.

V2 Driver stage, EL34.
The EL34 is triode connected and has a gain of about 8.7, close to the µ
of EL34. I had hoped to use a choke plus resistance to feed the EL34 with
anode dc so that this gave a high impedance dc feed to the tube but there
was no room to put any filter chokes, and very little time to do it.

The following grid bias R17, 47k, is bootstrapped to the cathode FB
winding at near 0V potential. This causes the its loading value on EL34
anode to effectively appear as approximately 203, and the total anode
load for EL34 becomes 25k in parallel with 200k in parallel giving
total of 22k. The EL34 has Ia = 16mA, and Ea = 320V approx, and
maximum anode signal = 180Vrms at about 2.5% of mainly 2H.

130Vrms is needed to drive the 13Ei grid to clipping level and at this
level the EL34 produces only about 1.8% 2H distortion and it could
not be made more linear easily. This 2H has a phase relationship with
fundamental frequency such that there is substantial cancellation of the
2H produced in the output stage, and most most effectively where loads
are less than rated nominal, when output stage distortion becomes highest.

In the SEUL, global NFB is about 16dB, so all distortions get reduced by a
factor of 0.159. So where there is no global NFB, there may be 6% THD,
including slight 2H cancelling between driver and output stage.
When GNFB is added, THD is then reduced to just under 1%. 

In the case of amps with substantial amounts of CFB such as the SE32 here
and SE35, the THD without GNFB varies with load value and is under
1.5% because the THD of the driver tube cancels the low THD of the
output stage. Therefore GNFB need only be 9dB and all distortion is
reduced by a factor of 0.36 so typical THD may be 0.54%, and about
equal to the SEUL amp but while using 1/2 the amount of GNFB.
Usually the CFB amp has lower Rout, ie, better damping factor.
Distortion measured much lower with CFB for low value loads.

However, it is now 2011, and the owner of this amp would like me to
try to put a choke into the EL34 anode circuit and while I am doing
this I can see there are further improvements I can make to the PSU.
The use of an anode choke for EL34 driver will allow me to abolish the
+700Vdc B+ rail and allow slightly higher Ia. This should improve what
is called the "micro-detail" in the sound quality which should then come
across s more natural, ie, higher fidelity, irrespective of any measurements
which never ever fully explain what we hear from SE amplifiers when
they sound well.

V3 Output stage has the 13E1 set up as a beam tetrode with a screen
Eg2 = +175Vdc, Ea = +475V, and Ia = 155mA, for a Pda = 73.6W.
The screen heat dissipation, Pdg2, is very low because the 13EI was
designed to operate with low screen voltages under +200Vdc with anode
voltages of up to 800V. With such low screen voltage the screen current
at idle is also low, and less than half what it is when using 13E1 in UL or
triode mode which is unsafe if Ea and hence Eg2 exceed +375Vdc.
I have an OPT cathode winding devoted to giving 33% of the total Va-k
signal as local cathode voltage feedback in series with the grid input signal.
So why was this 33% of primary turns chosen for local CFB
when 12% to 15% would be plenty?

When I wound the OPT for these amps in 1997, I used the following recipe :-

Core = double C-cores with strip width = 55mm, and build up = 36mm,
low grade GOSS which was all I could obtain locally in 1997. Max µ = 4,500
without a gap, but with a gap µe is about 350.
The air gap was set so 200mAdc would magnetize the core to about 0.6Tesla.

The Primary is 1,800 turns in 3 sections of 600 turns each with the center
section subdivided to give two 200 turn windings and two 100 turn windings
to allow a variation of screen connection points for UL and for future
arrangements.
The Secondary has 4 sections interleaved symmetrically with the 3 P sections,
giving an interleaving pattern of 4S x 3P, or S-P-S-P-S-P-S.
Each S section is a single layer of 57 turns each, with the last on section
divided into 3 sub sections of 19t each, and the arrangement allows :-
4 parallel 57t secs for 2k8 : 2.8 ohms,
3 parallel 76t secs, for 2k8 : 5.0 ohms,
2 parallel 114t secs, for 2k8 : 11.24 ohms.

The 2.8k to 5 ohm  match was selected for the above schematic,
1,800 P turns to 76 S turns.
It was decided that all of the center P section of 600 turns would be used for a
CFB winding which has one end taken to 0V. I could have used 1/2 the center
P section for 16.5% CFB and this would have resulted in only 50Vrms
cathode FB and an easier drive voltage of about 80Vrms at the grid.
But then I would have had a high Vdc potential between two adjacent P
layers of turns without enough P to P insulation thickness, and to avoid
the risk of dc arcing, I used the whole center section of P turns.
In any case, the amp is used at low levels for hi-fi where average signals
are 1/10 of the peak signals, and well away from high distortion levels.
The best screen arrangement took a day to work out. At first I just had
the screen going to a fixed voltage of +150Vdc above the cathode, as the
data on this tube says Eg2 at +150V is OK even though Ea might be 5
times this voltage. The 13E1 was designed at a time when designers tried to
produce beam tetrodes which did not need a high screen voltage or screen
current for mainly economic and efficiency reasons, but also for better
reliability with less voltage and current involved. It is mainly luck that the
13E1 works in triode mode or UL mode at all because in these modes the
screen is at the same potential as the anode and the limits for the Ea are
determined by the effect screen voltage has on its current draw. So Ea
of +375Vdc is the maximum for the 13E1 in triode or UL. With a high Eg2,
Eg1 must be increased to control the idle Idc, and with SEUL the Eg1
must be about -80Vdc, and any further increase of Ea&Eg2 beyond 375V
results in the likelyhood of the grid g1 losing control of the idle current.
With CFB, you could have Ea much higher, perhaps +800V which would
be useful in a push pull amps and then a pair could produce an output
power in class AB1 of well over over 200Watts with a few initial Watts
of pure class A. PP operation would be better with Ea no higher than
used for 4 x KT88/6550, ie, about 500Vdc, to give 100W max, with
at least 30 initial Watts of pure class A.

The 2k8 anode load for 13E1 was chosen to give a match for maximum
clipping power into 5 ohms, and then Ea adjusted from available taps on
the HT winding to suit the wanted load. 
Now for all beam tetrodes and pentodes:-
Load RLa for maximum power approximately = 0.9 x Ea/Ia.
Pda at the anode = Ea x Ia, so Ia = Pda / Ea,
so RL =  0.9 x Ea squared / Pda.
In this case the load was selected at 2,800 ohms.
So 2,800 = 0.9 x  Ea squared / 73.6, so Ea = 478.51Vdc.
With Pda = 73.6 maximum, Ia = Pda / Ea = 73.6 / 478.5 = 153 mA.

In practice, these Ea and Ia calculations proved to be very near correct.

At first I tried to have the screen supplied with a fixed Vdc voltage at
150Vdc above the cathode Vdc.
But  I found that with 33% of primary turns at the cathode and 66% at the
anode, the cathode voltage would swing upwards and so close to the fixed
screen voltage that the tube would go into cut off and the distortion became
high, and power limited to less than SEUL.

So I then connected the earthy end of the screen supply to available
tapping points on the cathode winding which was wound with these taps
to allow varied UL % taps.
The best outcome was when the screen was bypassed to the CT of the
CFB winding, or at 16.5% of the total primary turns. This meant the
minimum voltage between screen and cathode was well above the threshold
for Ia cut off caused by Eg2 becoming too low. Then as a double measure
I raised the Eg2 supply slightly to +175Vdc above the cathode and no
premature "cut off distortion" could occur at any load value.
The final result gives 32 watts and twice the power at clipping that anyone
gets when they try to use this tube in triode and slightly more than with
SEUL and with less THD and output resistance at all levels than for either
triode or UL.
So the screen connection method and Eg2 remains high enough at all
times to have its proper influence on the electron stream.

There are actually TWO local NFB circuits. Any distortion voltage between
anode and cathode appears at both anode and cathode but in a ratio of +2 : -1
respectively. So if we nominate the distortion voltage appearing at the anode
= -2d, then at the cathode it is +1d because of the relative transformer winding
phases, and it is as if -d appears at g1 and Ek was at 0V. The -2d and +1d
represent what we would measure with the cathode feedback action happening.
The open loop gain between grid and anode is about -10x, so a +10d
"correction signal" must exist at the anode, even though we measured -2d.
This seems to be impossible but there *was* already -12d produced at the
anode without any cathode feedback, and the generated correction signal
sums with the "open loop" -12d to leave -2d, the THD with "closed loop",
ie, with NFB applied.

But -d also effectively appears at the screen g2, and the screen has a gain
into the anode load of maybe -3x, so +3d also appears at the anode to give
an additional correction signal of +3d, and it also sums with whatever must
have been the distortion without any NFB, -15d.
So the distortion reduction with the 2.8k load as shown is from -15d to -2d,
or by a factor of 0.133, ie, -17.5dB.
The amount of distortion reduction is much more than provided by any
UL screen tapping, or by triode connection.

I won't bore everyone silly with all the complex reasoning behind why the
high beam tetrode Ra of the 13E1 is so much reduced from about 10.6k in
pure beam tetrode with no FB present, and wit the same Ea and Idc idle
conditions. But if the screen bypassed to the cathode instead of a tap along
the cathode winding, then the tube works in pure beam mode but with only
one loop of NFB around the grid to anode circuit, and Ra with FB is easily
calculated as Ra' = Ra / ( 1 + [ µ x ß ] ) where µ = amplification factor = 220,
and ß = faction fed back = 0.33.
So Ra' would be 10,600 / ( 1 +[ 220 x 0.33] ) = 144 ohms, a huge reduction.
But with the screen taken to a tap and fed with some signal of opposite phase
to the anode, the internal tube gain condition is equal to working with a 16.5%
ultralinear tapping, and this is enough to much lower µ and also the high tetrode
Ra. Nevertheless, regardless of mathematical explanations, the 33% of grid to
cathode CFB reduces Ra from 10.6k to about 400 ohms and near the 300 ohms
you get with triode connection so that with an OPT Z ratio of 2,800 : 5, or 560 : 1,
the anode resistance appears as 2,800 x 400/560 = 0.71 ohms at the OPT speaker
secondary connection.
Winding resistance of the OPT adds about 0.2 ohms and so output resistance
without global NFB is about 0.9 ohms.
The 9dB of global FB reduces this output resistance to 0.32 ohms giving a
damping factor of over 9 even with a 3 ohm load.

The easier and simpler way to set up the 13E1 tube is to have a fixed Eg2 at +175V
above the Ek at +33V developed with the cathode R&C biasing network,
ie, at +208Vdc. The CFB is then limited to less than 20% of the total anode turns.
People winding suitable OPT can have a total of 20 layers of wire for the anode
and cathode primary windings, and devote up to 4 layers to the cathode and 16 to
the anode, and arranged so the cathode winding is split into 2 windings and placed
among the other 16 layers for anode windings, with adequate insulation of 0.5mm
between any windings with 500Vdc potential difference. Speaker secondaries
should also be well interleaved with the primary in at least 4 or perhaps 5 sections
of 1 layer of wire in each. This will give still give you an output stage with less
THD/IMD than any UL or triode stage, and effective Ra near a triode,
and needing only about 75Vac maximum grid g1 drive voltage.
 
The single 13E1 stage will thus perform about the the same way as four 300B in
parallel, giving 32 watts output for a total of 75W input for anode plus screen pda.

If 300B were used, Pad total might be 120W, and the tube cost would be a lot
higher than a single 13E1 at present 2008 prices.
Insulation thickness minimum between any adjacent anode winding layers all at
+500V should be about 0.05mm.
Insulation thickness minimum between any anode layer and either cathode layer
or secondary layer should be 0.5mm.
For evenly distributed leakage inductance the insulation between any anode or
cathode layer and speaker secondary layer should be the same minimum 0.5mm.
Polyester should be used, well varnished, and never paper, which will have a
much too high dielectric constant when impregnated with varnish, thus increasing
effective parsitic and unwanted shunt cpacitances in the OPT.

With only 9dB of global NFB, at very loud listening, THD < 0.05% into any
load above 2.5 ohms. Noise is extremely low, even with a non-potted
"open frame" OPT and PT on the same chassis. But at least I have placed the
PT away from the OPT and orientated to prevent any significant stray magnetic
coupling. The local CFB and global NFB reduces whatever small amount of
stray magnetic coupling exists, but using mild steel boxes to pot the OPT
and PT will definately reduce any possibility of magnetic coupling.

The measured THD of the completed SE32 was very much like the results
I obtained with the SE35, and well below the SEUL22 levels and for the same
reasons I cited for the SE35 regarding natural unforced 2H distortion cancelation
between the driver stage and output stage. So there is little point to me publishing
the THD graphs I obtained for the SE32.
Distortion is quite low enough, and its all anyone really needs to know.

There is a less understood reason why local CFB works and sounds so well.
And it seems true even though a similar total amount of global and local applied
around 2 stages of triode gain and a single 13E1 acting in pure beam tetrode mode
would measure slightly better.

If there is local FB in a single gain block with one tube in class A, the distortion
correction signal does not have to travel through other stages on its way to correct
open loop distortion and thus generate other low level IMD products along the
way which also have to be then corrected. It becomes a never ending roundabout,
generating ever more high numbered harmonics at low levels.
Its better to have the slightly higher measured distortion of a series of stages each
with their own loop of NFB if that level is low enough.

As everyone should know, all triodes have inbuilt and unavoidable natural
electrostatic shunt feedback loops. The action of the triode FB is maximal when
the gain of the triode equals its amplification factor, µ, which can only occur
when there is zero current change even though there is a high signal voltage change.
Such a mysterious thing is seldom understood by anyone, but it is the nature of the
triode and it makes it the most naturally linear device in the universe when operated
with some external loop FB from resistance networks or transformer windings.   

There is plenty of electrostatic shunt NFB in the input and driver triodes of the amp
I have described here because the gain of all triodes has been kept high because I
have arranged their loads to be high so gain approaches the µ.

Therefore the SE32 will work well without the global NFB if it is really not wanted,
especially where the speaker load was perhaps 8 ohms or more when the damping
factor would be fine without the global NFB. The sensitivity would increase by a
factor of 2.8, ie, full power could be developed from only 0.32Vrms
In my case I am adding only 9dB of global NFB, a tiny amount compared to the
typical 60dB around a typical solid state amp. The numerical difference is between
3 times to 1,000 times.
The amount of global FB I apply is around a substantially linear circuit to begin with,
so few IMD low level products are formed. The extra global NFB lowers the Rout
to get a very good damping factor which translates to good speaker driver control
even if the the speaker Z dips to 2.5 ohms. Any damping factor increase would
be inaudible, IMHO.
If ever anyone were to try to use the 13EI as a pure beam tetrode without any
FB but with the above load and dc operation, they may well be shocked at the
non-linearity, with 2H, 3H and other H reaching above 10% at the onset of clipping.
But all beam tetrodes and pentodes are like this. 13EI open loop gain would be maybe
40 though, so there is lots of gain that can be easily be reduced with external and
linear NFB mechanisms such as provided by tight magnetic coupling in an OPT.
The linear NFB path around the tetrode here is a more linear path than exists in
all triodes which don't have a screen to interrupt their NFB action internally.
Triodes are fine, but their NFB delivery path is one obeying a rate of current
change proportional to a cube root of a constant squared, and triodes only really
become very linear when there is minimal Ia change. But in a power tube we
want a lot of Ia change because there is real work to be done at a speaker.
So we can use a beam tube, and apply the local FB, and drive it with a triode
which has minimal Ia change, and as long as the driver tube doesn't go anywhere
near clipping, the total outcome will produce low distortion. I'd never use more
than 33% CFB if I began from scratch because at 50%, drive voltage needed
leaps to over half the total Va-k on the output tube, or about 180Vrms,
and then the driver tubes begin making more distortion % than the output stage
and few benefits are gained.

Those wanting to use all 9 pin tubes instead of octals for the driver amp should
consider the input and driver amp I have used in my Deep Space 845 amps detailed
elsewhere on this site.
It uses one 6CG7 plus three EL84 to make about the same drive voltage needed
for an 845.

Never be tempted to let 13E1 Pda exceed 73W. The anode will begin to
glow red at 80W, and the sound becomes mud, even though data says
Pda max is 90W.
The heater generates over 30 watts of heat, and radiates this heat at the
anode which passes it on through the glass. The old data Pda rating of 90W
is a design rating for maximum signal generated Pda which in amplifiers
always varies from a low average figure and with the tube biased for maybe
20 Watts at idle for push-pull AB1 use. I have the two 13E1 which I
originally fitted to this pair of amps back in 1997 and in 2008 and after an
estimated 7,000 hours they still measured as well for maximum power
as when new, but did develop some positive idle voltage at their grids
which means there are a few stray positive gas ions in the tube which
cannot be absorbed by the gettering. The use of low value biasing resistors
not exceeding 47k does tend to prevent the positive grid current even at
idle from becoming too high, thus turning on the tube which makes it
hotter, thus generating even more positive grids. 

Stability must be checked as always with local or global NFB, and the
R24 + C19 worked fine for my output stage, but maybe different R&C
values would be required in something made by someone else, and with
slightly different amounts of leakage inductance and capacitances in the
output circuit with OPT.

Fig2.

In Fig 2 above, there is a total of 4,700 uF for the main 500V B+ supply
filtering. There are no filter chokes, and they are not needed in this case
because if you increase C enough then the R values for R&C filtering
become low enough not to dissipate much heat, and yet attain a high
enough ripple attenuation factor.
Ripple voltage Vr at top of C9 = 1.6Vrms, 100Hz, and is reduced by
a two stage RC filter with  an attenuation factor of 0.0017, so Vr at the
OPT = 2.8mV, and quite low enough.
R12 and R16 are mounted on a heat sink to keep their temp low as they
dissipate 4.5Watts each.
They each consist of 5 x 820 x 10W all in parallel.
The +780Vdc at the top of C3 is developed by means of a 1/2 wave
voltage doubler working from the +500V main doubler rectifier for the
anode supply current. The +780V is made by the doubler formed with 
C11, and two 1N5408, and feeds C3 through R15, and peak charge
currents are low, and don't affect the switching of the anode diodes for
the main anode supply.
If anything in the EL34 shorts to 0V, the cheap R will burn open before
the circuit produces smoke from the PT.
A short in the main 515Vdc anode supply will blow the mains fuse.

Active protection has been fitted to the SE32 circuit to guard against
excessive Ia in 13EI, but has not been drawn up yet.
It has a simple RC filter using 4.7k from the cathode to a 470uF cap
to reduce the ac voltage but allow the Vdc at the cathode to be divided
down further by a resistance network and applied to a C106D sensitive
gate SCR.
If the cathode Vdc rises to 50Vdc, Idc in the tube would be 217mA,
and Ea would drop by about 25Vdc, making
Pda = about 93Watts, and the tube would show some red and be over
stressed, but able to cope for a short time. At Vdc at cathode = 50V,
the SCR is arranged to turn on, causing a relay to open in the HT winding
on the PT so that the whole anode supply is turned right off, and no
damage is sustained. With such a small Ia change involved between correct
operation and a fault condition, active protection which has precision which
ordinary fuses cannot provide. Owners are notorious for fitting the wrong
value of fuse after a fuse blows, and therefore causing much more
expensive damage. My protect circuits can be triggered if there is a
shorted speaker load connected, or if bias failure or tube failure from
any reason occurs, and the amp may be re-set by turning off, then
back on. Repeating fault conditions mean the amp needs a visit to a
capable technician.
The amps now have a blue "on" LED, and a red LED turns on when
a fault occurs. 

The 6SL7 has a dc supply to its heater as shown to minimize its hum.
Those wanting a similar gain and Ra and wonderful sound and less hum
but from a 9 pin tube could use a 12AY7, or 12AT7.

If you wanted me to build you a pair of these amps, I have plenty of 13EI
which are no longer made, but the price would be around aud $8,000,
and at 2008 exchange rates.

There are few SE 30W mono SE class A amps offered on the Net for
less than what I want, and those that are offered may not have as much
simple sophistication included, nor be genuinely hard wired point to point.
In 2008, there was a SE VA350 made by KR Audio available from the
local Duratone Hi-Fi  shop in Canberra. It is an integrated amp using
2 x KRT100 output triodes similar to 845 for 30W per channel and
with solid state driver circuits and the price is aud $18,000.

These days I have moved away from using 1.2mm brass for any chassis
material. I found it better to use all steel at least 1.6mm thick or have a
steel rectangular channel frame with a 2mm aluminium top plate like
I use in my 300W monoblocs.
The SE32 amps come with a sturdy steel grille over the tubes to allow
good ventilation and so you can see them at night, but act to prevent
the inevitable "oops" when something falls onto the amp and breaks a
tube or pushes one sideways in its socket.

Fig 3, 2012 Improvement for SE32 from 2008 :-

This schematic has almost identical operation to Fig SE32 from 2008 above.
The differences are :-
+750Vdc rail is deleted, 25k RL carrying dc to EL34 is deleted,
68k dc carrying RL to 6SL7 is deleted.
Choke added to anode circuit EL34, CCS added to anode circuit
of 6SL7.

Fig 4. SE31 schematic, similar to SE32, but different OPT, lower B+.

The major difference is the fixed Eg2 slightly higher because the amount
od CFB is 17.25%, thus there is no risk of Ia cut off with the low Eg2,
which is raised anyway to +200Vdc. The intended OPT for this amp
has much lower RLa of 1k2 than 2008 amp with 2k8.

Fig 5. PSU for SE31 with lower B+.

And for those who might be tempted to use 13E1 for Push Pull operation,
with comfortable operation giving plenty of class A :-

Fig 6.

Fig 7. The power supply for PP 60W amp with 13E1 :-

Happy listening!

Back to Index.