This page explores single ended output stage properties with beam tetrodes and pentodes.
The aim is to produce the best audio fidelity which includes :-

1. Maintaining high enough gain of output tubes with local NFB while enabling drive
voltages at low at low distortion,
2. Minimizing any reliance on large amounts of global NFB,

3. Locally applied NFB within the output stage to reduce distortion and Rout to very
low levels to give better performance than triode strapped tubes,

4. Maintaining unconditional stability which means no possible combination of L, C
or R loads, or with no load at all can cause oscillations at any frequency,
5. Ensuring that stability is possible with critical damping R&C networks to reduce gain
and open loop phase shift below 20Hz and above 20kHz.
6. Maintaining bandwidth of 10Hz to 65kHz with pure resistance load at the -6dB output
voltage level, ie, 1/4 full rated power,
7. Ensuring that output power at -1dB below clipping is possible without any stage
of the amp being overloaded of forced into grid current draw between 20Hz and 35kHz.

Single ended output stages have been used for audio amps ever since the first amplifiers
were used for radio
and audio frequencies after the invention of the triode in 1903.
Up until about 1960 nearly everyone in the world listened to radios and TV sets which
had one tube devoted to
powering a loudspeaker.
Push Pull operation could give more power than most ppl ever needed, and were
initially more expensive to produce than using just one output triode. But once beam
tetrode and pentodes were invented, PP amps became popular because they could
operate in class AB with low bias current. Such PP amps required less power from
batteries or mains so construction costs declined which suited manufacturers.
However, if a single large tube or a number of paralleled tubes are used in pure
class A1 audio power, the sound produced
may be better than many PP amps.
I'd rather have one KT88 in triode making 8 Watts than having two 6V6 in PP
making 10 Watts in class AB.

Content of this page is based around schematics. 

Fig 1. Three most used basic SE amp stages, 13W to 10W, SE pure tetrode,
SE Ultralinear, SE triode, 1 x KT88/6550.

Fig 2. 20W+ amp with SEUL with 2 x EL34, KT66, KT88, KT90, KT120, etc.
Fig 3. The Equivalent Model of KT88 with g1 and g2 inputs treated as current

This allows understanding of operating properties of a KT88, and its Ra, gm g1,
gm g2, and to analyze all voltages
and currents in all electrodes to determine
voltage gain, and effect of local NFB.
The theory may be applied to all power tubes
including pure triodes which do not have a screen, such as 300B and 2A3.

Fig 4. 20W+ amp with CFB, with 2 x EL34, KT66, KT88, KT90, KT120, etc, using
same SEUL OPT in Fig 2.

Fig 5. 36W SE amp with CFB with 3 x KT88 etc, using SEUL OPT, 25% UL tap,
1k3 : 5r6 Z ratio.

This was designed to include "choke sink" for cathode current, and choke in anode
feed for driver tube.
Probably nobody
has ever built an amp like this because they need to source
good quality chokes.
The THD and Rout is much lower than
conventional SEUL amps.
Fig 6. 25W PSEUL + CFB amp designed around the Hammond 1640SEA output
transformer with mosfet CCS at KT88 cathodes,
Fig 7.
25W PSEUL + CFB amp designed around the Hammond 1640SEA output
transformer with choke at KT88 cathodes,

Fig 8. SE CFB output stage and SEUL output stages with OPT with 3 windings.
Fig 9. Choke Feed SEUL and SET output stages.
Fig 10. Choke Feed SET amp with 845.
Fig 11. Choke Feed SEUL with floating B+ supply.
Fig 12. Choke Feed SET with floating B+ supply.

There are a number or ways to arrange output tubes, OPTs, and driver stages
and NFB. It is impossible to always consider the output stage entirely separately
and without the interaction of the driver and input stages and the NFB loops.

Everyone should be familiar with the basic class A working of a single vacuum tube
( or a number of them paralleled )
in a circuit with tubes, OPT, and PSU all in series,
with an idle current which is varied between
zero idle
current and twice idle
, at maximum audio output power.

The operation of the tube will always remain predictable in terms of applied rail
voltages and applied signals to all electrodes.
The Ea, Eg2, Eg1 and Ia may all
be varied within a range to give ideal working with a chosen OPT.

Fig 1. Basic configurations of SE amp stages.

The main 3 varieties of "Single Ended" output stages used in many simple tube
amps are shown here.

I am using a modern Svetlana KT88 to show what is possible but other brands
of KT88 or 6550 could be used similarly.
The general principles apply to the range of power tubes now readily available,
could be
KT120, KT88, 6550, KT66, 6L6GC, 5881, 807, 6V6, 6CM5, 6CA7,
EL34, EL33, EL84, EL86, GU50, 13E1,etc, etc.


SE Beam Tetrode or Pentode.
See Far Left side of Fig 1.
The use of single ended amps with one beam tetrode or power pentode offers high class
A efficiency of up to 45% but with high THD with a dirty mix of both even and odd
numbered harmonic products from low levels to clipping level. The THD mix of H varies
much with load value. The IMD is appallingly high and Rout is high and to make such a
tube configured here work far better, at least 20dB of GNFB must be used from OPT
sec back to an input port of an input tube, usually its cathode. Many such amps have
been built with an output stage like the one shown and all seem to me to be rather
damned awful. Put it this way, if you use an EL84 to make 4 Watts in pure SE pentode
mode, and with 20dB GNFB, then its sound just is not as good as using a 2A3 to make
4Watts with say 12dB GNFB. Millions of AM radios made before 1955 had a lone 6V6
for the audio amp and no GNFB. Although the sound of music was awful the set was
cheap because most ppl were poor, and you could understand the nightly news and
cricket scores, and be manipulated by politicians and advertizing.
Beam or pentode tubes are easily driven because gain is highest, and drive for class
A1 never involves class A2, where grid voltage goes positive during input cycles so
grid current is then drawn, with huge reduction in grid input resistance. 

SEUL Beam Tetrode or Pentode.
See Center of Fig 1.
use of a tap along the anode primary winding was rarely ever used by any
manufacturer because it cost more money to put the tap there, and CEOs hated
innovation because it always cost money, and CEOs don't always agree with the
marketability of innovation. Besides, the idea of using an "Ultralinear" tap was not
invented until maybe just after WW2. With a tap at 40% to 65% of anode primary turns
the beam tetrodes or pentodes acquire triode like harmonic products, less IMD, while
maintaining anode efficiency at over 40%. Gain is less than pure beam / pentode and
grid drive remains free of grid current. There is nothing "ultra" about the linearity of
having a screen tap for "partial triode" operation, but in general the UL connection
makes the tube work very much better than otherwise, and measurements and
sound heard will verify it. Some GNFB is still needed, usually 15dB.

SET - Single Ended Triode.
See Far Right of Fig 1.
In the early days of electronics only one triode was used for audio amplifiers.
The idea remains well respected as we all see with 45, 2A3, 300B, 211, 845,
GM70 etc, for SET triode power
from 2Watts to 22Watts.
With real triodes listed, there was no screen grid connection shown above, and
OPTs never had a screen tap on OPT. All the common varieties of beam and pentode
power tubes may be configured as a triode by connecting screen to anode.
The screen then carries ALL the anode signal. This exerts electrostatic control of the
electron stream in similar manner to the control grid. The gm of the screen g2 is usually
between 1/20 to 1/6 of the gm of g1 control grid. The screen performs the task which
the anode in a real triode performs. This function is an application of local NFB.
However, the function is not a linear function, so despite the high amount of inherent
NFB in triodes or triode strapped multi grid tubes, there is usually about 5% THD at
full Po on the one load value which gives maximum Po. The efficiency of triodes or
triode connected tubes is between 15% to 33%, depending on the tube operating
conditions. Higher efficiency of up to 40% may be gained by using a direct coupled
cathode follower to drive the triode grid for class A2 operation. Usually class A2 is
more bother than its worth and if more triode Po is wanted, use more triodes and
stay in class A1. A renowned exception is Audio Note "Ongaku" amp with 211,
which has a 6SN7 cathode to drive its grid. Not all beam and pentode OP tubes
can be easily driven in Class A2 such as EL34. Grid current is just too high.

It is difficult to make a bad sounding triode amplifier providing no tube
is driven into any overloaded operation. Triode mode is the "safest bet to good music".
The Ra of triodes is usually much less than the anode load driven. At low levels
THD and IMD can remain quite low enough so GNFB may not be needed.
THD at all levels is mainly 2H. Gain of triodes is perhaps 1/3 that of beam or pentode
tubes and the drive voltage level may be quite high, up to about 110Vrms with 845.
But providing enough drive is never difficult by use of another SE triode driver tube.
The 2H of the driver cancels part of the 2H produced in the output triode.
So at all levels up to clipping, some SET amps produce surprisingly low 2H which
gives less objectionable THD and IMD than from a PP amp of the same power.  

Few radio sets ever had slightly larger output tubes than 6V6 or 6F6 such as KT66,
807, 6L6, or later EL34, KT88, 6550. But fabulous sound quality
could be had from
these tubes strapped as triodes. In fact, in many radios I have re-wired, it is always
possible to get at least 3Watts from a single
EL34 strapped in triode and power is
the same as one 6V6, and fidelity
with trioded EL34 is far superior to a 6V6.


The above output stages are all based around Russian Svetlana KT88 and similar
Russian EH 6550, KT88.
Tubes such as KT90EH and KT120EH may always be used
with slightly higher Ia, cathode bias Rk to suit,
and a lower RLa load to be used while
obtaining the same Va, hence SE Po up to 15Watts is possible.
But an OPT meant for
4k2 : 4r,8r,16r and Ia = 100mA max cannot be pushed by increasing Iadc much beyond

Notice that for the same B+ supply, and similar Ea and Ia, the triode load value is lower
than for tetrode or UL.
See my pages on loadmatching to SE triodes and pentodes to
work out operating conditions different to the above.

To get better triode performance the B+ usually needs to be say 15% higher than for
the same tube with UL.
Then the same OPT can best be used with triode.
In the above example for SE Triode, the OPT can be 4k2 : 4,8,16r.

If the B+ is raised to +490Vdc, Ia +Ig2 reduced to 70mAdc, EK = +52V, and Pda+g2
remains 30W.
Triode Po then becomes 10W. KT88 or 6550 then give very similar
performance to 300B.
Fig 2.

Fig 2 shows a very ordinary 20Watt Ultralinear output stage using a pair of EL34
tubes in parallel, and based on using an available Hammond 1627SEA OPT with
ZR = 2k5 : 4r, 8r, 16r.

My comments below are based on using EL34 or 6CA7, but parallel pairs of other
tubes such as KT88, 6550
can be used.
There is no NFB applied in the EL34 output stage except for the local "ultralinear
screen tap" which is commonly
not considered local NFB, ( although it actually IS NFB ).
The UL tap allows the high power of class A1 pure pentode but with triode like
products, ie, less odd number H and more even number H. Ra is
reduced greatly so a better damping factor is possible than for pure pentode / tetrode.

All triodes or triode strapped pentodes or beam tetrodes have internal NFB by
means of the field effect of the anode
interacting with the field effect of the control
grid to give a resultant effect on the electron stream from cathode to anode.

See my equivalent models of multigrid tube basic operation with Ra in shunt
with g1 and g2 current  generators.

For multi-grid tubes such as EL34, KT88, having Eg2 a fixed Vdc hugely reduces
the signal field effect of
anode upon Ia current. The screen g2 may be connected
to a source of signal voltage which is a fraction of the anode
voltage and this
has an opposing action to whatever the g1 signal voltage tries to do.

A diagram about basic tetrode of pentode operation may explain more lest I
completely confuse everyone.

Please remember that the beam tetrode has no suppressor grid as found in the
pentode, but has beam forming plates
connected to the cathode internally which
function like a suppressor grid g3. There is no need for mention of the
suppressor grid or beam formers again because the relevant operation facts
are concerned with control grid g1 and
screen grid g2.

If you can follow and apply what you find in my Fig 3 diagram, you will find it
useful to analyze possible
signal voltages and currents like our ancestors did,
and all without PCs and simulation programs.
Understanding models frees you from any need  for a PC. But unlike our
ancestors who used a slide rule you may use a $5 pocket calculator.

Fig 3.


Fig 3 above shows an equivalent model for what is happening with 1 x KT88.
Other tubes such as 6550, KT90, KT120, 6L6GC, 5881, KT66, 6V6, EL34, EL84, etc,
will show similar basic operation,
and by understanding the model you can analyze
any output stage or design one. But you MUST start by knowing the
g1 gm and g2 gm,
and the Ra for the tube operating conditions of idle values of  Ea Vdc and Ia dc..

In the above Fig 3, the figures for signal current and voltage amplitudes have been
prepared from measurements
of transconductance of g1 and g2 with tested
samples of tubes under the conditions listed for Ea, Ia, Eg2.

The Fig 3 figures above don't apply to all output tubes but will agree with curves
shown for Svetlana KT88, and
other Russian makes of KT88 and 6550.

Testing tubes in a real world amplifier isn't difficult after you learn what you are
doing which may take days of practice.

Hum voltages present in output stages may spoil the attempts to measure small
signal voltages in an amp.
To avoid noise interfering with measurements, ensure
there is adequate filtering of the B+, and if not, apply much larger
filter caps and
perhaps a choke between rectifier and the existing B+ filter caps. This is very
necessary in SE amps where
there is no common mode rejection of noise in B+ rails.

But let us suppose you have an existing SE amplifier with a multigrid output tube,
and you have well filtered B+ rails for anode, and g2 and input stages.

To determine g1 gm :-
The g1 gm is the g1 transconductance which is the ability of the g1 grid to vary
the electron flow and is expressed in mA / Volt.
Connect 100r resistance between anode and OPT anode connection.

2. Connect wire shunt link with alligator clips from OPT anode connection and
B+ connection, to short circuit the primary of OPT.

3. Make sure the screen g2 has the intended fixed supply voltage.
4. Disconnect any global NFB and output speaker loads.
5. Connect audio signal gene to amp input and use a 1kHz sine wave and
increase until there is 1Vrms at the output tube g1 grid.

6. Measure the signal voltage across the 100r and have a CRO connected
to anode to make sure distortion is less than 2%.

Suppose you measure 0.64Vrms ( Vac on DMM ) across 100r.
7. Calculate g1 Gm. The current in 100r = 0.64V / 100r  = 6.4mAac. 
This means the 1V change at g1 produces 6.4 mA change at anode so
gm = 6.4mA / Volt. Simple?

The anode signal voltage variation will cause virtually no effect on the gm
measurement because the gain between g1 and anode is negligible
with a very
low RLa of only 100r. Hence there is no action of internal NFB within the tube. 

8. For triode connected tubes, the same set up works OK but g2 is connected
to anode, and the gm measured will
vary slightly, because g2 signal current is
generated in addition to the anode current. Triode Ra may also be neglected.

The load of 100r is a near vertical load line is drawn upon data sheet Ra
curves for KT88. We wish to understand the basic current
change behaviour
caused by grid voltage change.

Measuring g2 gm :-

The g2 gm is the screen g2 transconductance which is the ability of the g2 grid to vary
the electron flow and is expressed in mA / Volt.

1. The above set up is used with 100r anode to B+ load, and OPT primary
shunted by a wire link.
2. Ground g1 with 2uF so there is no g1 signal possible, but the wanted g1
Vdc bias is undisturbed.

3. Disconnect g2 screen is from wherever it is normally connected, and insert
1k0 in series with g2
and original g2 B+ supply,
so that g2 retains its Vdc bias supply.
4. Connect a signal gene between 0V and g2 but use a 2uF DC blocking
cap to enable up to 5Vrms at 1kHz to be applied to the g2.

5. Measure the signal voltage across the 100r anode to B+ resistor.
Suppose you measure 0.45Vrms at 100r. It means 5V at g2 produces 0.0045
amps of anode current.

6. Calculate g2 gm = 0.0045A / 5.0V = 0.0009A/V = 0.9mA/V.

Measuring Ra :-

Ra is the internal dynamic resistance of the tube when operating in pure beam
tetrode or pentode, and it always exists in the current gene model and it varies
depending on Ia and to Ea. 
Ra is a difficult parameter to
measure accurately for beam and pentode tubes
because it is a high number of ohms. Ra will be easier to measure
there is a UL tap or triode connection, and then the Ra is the beam or pentode
Ra IN PARALLEL with the effect of the g2 current generator. When the g2 signal
voltage remains the same as the cathode voltage ( with a cap bypass between g2
and k ), then the g2 generator has infinite impedance between the anode and k.
1. To attempt to measure pure tetrode or pentode Ra, disconnect all shunts across
OPT winding and have g2 taken to the wanted B+
fixed voltage.
2. Remove any R&C zobel networks across the OPT secondary or primary.
Remove any secondary load or global NFB connected.

3. Make sure cathode is well bypassed to 0V.
4. Connect 1k0 between anode and a signal gene with DC blocking cap = 1uF.
((( This can be tricky if you have a solid state signal gene
because such test
gear can be fried to a useless crisp if their outputs are subject to external voltages
exceeding say +/-20V.
So you need to have adequate protection measures on
inputs and outputs of ALL TEST GEAR !!!!!
Using a tubed cathode follower
in a separate box is good practice for connections
between delicate SS test gear and tube gear.)))

5. Ground g1 of power tube with 2uF cap and ensure g2 is grounded without a
series screen stopper resistance.

6. Apply 10Vrms at 1kHz to input side of 1k0 resistance from signal gene.
Measure the anode to 0V Va-c, and record this.

7. Measure the Vac across the 1k0.
Suppose you measure 0.4Vac across 1k0. Then current flow from anode to
0V = 0.4V / 1k0 = 0.4mAac.

Suppose you measure Va-c to 0V = 9.6Vac.
8. Calculate Ra, dynamic anode resistance between anode and cathode
= 9.6Vac / 0.0004Amps ac = 24,000 ohms.

You will quickly realize that using the published data sheet values for gm and Ra
will lead you to gross mistakes in calculations
of what is going on in your tube.
A KT88 with Ia = 60mAdc may have g1 gm = 6.4mA/V, but the data sheets tell
us g1 gm = 11mA/V at Ia = 140mAdc. Nobody would ever have a KT88 at idle
with 140mAdc.

Let us suppose the OPT Lp = 25H, and at 1 kHz this has inductive reactance XLp
= 157,000 ohms, with perhaps 400pF of shunt C
which makes XC = 397,000 ohms.
The XLp, and XCsh will be found to have a negligible loading effect at 1kHz
because usually
you will find Ra is always below 50k for most large multigrid output
 tubes such as KT88, KT66, EL34.
So you may ignore the current flow from signal
gene through primary inductance and any shunt capacitance. ( But of course there
are some very badly made amps around and you can't
assume XLP and XC will be negligible. So measure these parameters as well.

Always avoid making any stupid assumptions.
Measuring triode Ra in the manner above but with g2 tied to anode, expect Ra to
be 1k1 for KT88.

To measure the amplification factor .
The amplification factor is the calculation of gm x Ra. It is in fact the voltage gain
of any tube where a grid causes anode voltage change without any current change.
This means the anode load = infinite ohms, and the loadline for an infinite load
is a horizontal line across the data Ra curve sheets for Ea and Ia characteristics.
The of beam and pentode tubes is the most constant parameter, with large
variations of gm and Ra. Triode operation makes much more constant for
wide Va swings. 
1. Set up the amp as it would be for normal operation, but with no global NFB loop,
no loads or zobel networks.

2. Apply 1kHz sine wave to output tube grid g1 via 2uF cap. Use DMM or hi-Z
input volt meter to measure Va to 0V.

3. Connect CRO to monitor the Va signal distortion.
4. Increase g1 signal to obtain Va 50Vrms. All output tubes should be able to
achieve this much voltage with no load,
and with THD < 5% for multigrids and
less than 1% for triodes.

5. Measure the grid input voltage.
Suppose you measure 0.312Vac.
6. Calculate voltage gain and without any load = Va / Vg = 50 / 0.312 = 160.
It is just a number, no units. It may be considered a negative number because
a negative going voltage at g1 or g2 produces a positive going voltage at anode.
The voltage gain without any signal load current = amplification factor, . 

7. Verify measurements and calculations. For all tubes and for any g1 or g2 input,
transconductance, gm, = / Ra.

If we knew gm, and Ra, we could calculate . From the above examples,
g1 gm = 6.4mA/V and Ra = 24k,
so for g1 = 0.0064 x 24,000 = 153.6, close
enough to the 160.

The Ra in Fig 3 is always present as a shunt resistance in Fig 3 and is in
parallel with the effect
of the g2 current generator if there is a signal voltage
applied between g2 and cathode. Thus the
g2 generator can become equivalent
to a resistance.

Consider Fig 3 with an OPT set up with a 25% UL tap for g2 so while there is
+100Va-c+ at anode, +25Vg2-c appears at
The +V at g2 increases current flow. it is calculated = Vg2 x g2 gm
= 25 x 0.9mA/V = 22.5mA.
If you went looking for the 22.5mA, you would not find it,
because it is a "useful imagined current" that
exists in a mathematical model to
help explain what we do measure in a real circuit.

Consider Fig 3 without the 4k0 RLa anode load. There would still be the imaginary
22.5mA with Va = 100Vac.
But there would be no
load current. The Ra of 25k is there, so the current in Ra
= 100V / 25k = 4mA.

Total Ia model current change = 22.5mA + 4mA = 26.5mA.
The effective Ra = Va / Ia = 100V / 0.0265A = 3,773 ohms.

The portion of this resistance due solely to g2 generator = 100V / 0.0225A = 4,444 ohms.

The effective Ra' for where a screen signal exists = Ra // ( 1 / UL fraction x g2 gm ),
where Ra is that for pure tetrode or pentode.

Example :- Effective Ra', 25% UL for KT88 = 25k // ( 1 / 0.25 x 0.0009 )
= 25,000 // 4,444 = 3,773 ohms.

Example :- Effective Ra', 50% UL for KT88 = 25k // ( 1 / 0.5 x 0.0009 ) = 25,000 // 2,222
= 2,040 ohms.

Effective Ra' for trioded KT88 = 25k // ( 1 / 1.0 x 0.0009 ) = 25k // 1,111 = 1,064 ohms.

From the calculation of Ra' for any % of UL, we can also calculate the UL g1 .
It will always be be less than pure beam tetrode / pentode and more that triode .
g1 = g1 gm x Ra, and if 25% UL Ra = 3,773r then g1 = 0.0064A/V x 3,773
= 24.15.
The screen has its own transconductance value g2 gm, based on having no g1 signal.
The pure beam or pentode Ra of the tube remains the same if the screen is used
alone for Ia change and Va change.
g2 = gm g2 x Ra = 0.0009A/V x 24,000 ohms = 21.6.
Screen gm is so much lower than g1 gm that screen drive is seldom ever used,
so pure screen driven output tubes are not discussed here. Feel free to experiment
to know more about screen drive properties. But try as you may, you will find the
model in Fig 3 very useful if you understand the concept of the basic input voltage
controlled current source which has infinite resistance looking into its output at the
top of the two circles on the drawing.  

The voltage gain, A, with a load for UL can also be calculated A = x RL / ( RL + Ra ).
For 25% UL, in this case, A = 24.15 x 4,000 / 7,773 = 12.42
For 100V into 4k0 load, the input voltage = 100 / 12.42 = 8.05Vg-c.
The only confusing thing remaining is the + / - sine for Vac. All tubes in "common cathode"
mode like in Fig 3
are inverting, ie, a + input voltage causes a - output voltage.
So as  said before, in effect, is negative, -, so -8.05Vg-c produces +100Va-c.

The damping factor is calculated DF = RLa / Ra. Typical values for a single KT88 without
any CFB or global NFB and as
shown in Fig 1,2,3, and for RLa = 4k0 :-
Pure tetrode, DF = 4,000 / 25,000 = 0.16 = very poor. 20dB NFB needed.
25% UL, DF = 4,000 / 3,773 = 1.06, much better, but 15dB NFB needed.
50% UL, DF = 4,000, 2,040 = 1.96, better, but 12dB NFB needed.
Triode, ( 100% UL ), DF = 4,000 / 1,064 = 3.76, 12dB NFB needed, but some can't hear
any difference with NFB.
CFB over 10%, fixed Eg2, DF generally as high as triode or higher.

The maximum THD of SE output stages without local FB is highest for pure
pentode or tetrode and can be 15%
for where the RLa is that used to get
maximum possible Po and at 1dB below clipping. Many even and odd H
are present at all levels.
For class A1, the amount of UL % can be up to 70% for most multigrid tubes

and the Po max without grid current is nearly the same as pure tetrode or pentode.
Odd H are much reduced
and THD will be around 7% of mainly 2H. But where
the UL connected tube produces the same Po as lesser maximum for pure triode
operation, THD may be less than the triode.

The less THD one has before NFB is applied, the better is the music.
The FB network allows a fraction of the output distortion H to be applied to an
amp's FB input. The input tubes and output tubes are not perfectly linear,
and the signals fed back create distortion H, so intermodulation, or "IMD" H
are generated which are the sum and difference between any two frequencies.
If you have 3H fed back with fundamental test tone of 1H, then the IMD products
will be 3+1 = 4H and 3-1 = 2H. Their level will depend on the non-linearity of the
amp devices.
So by means of intermodulation, H products appear at the output which were
not present when the amp was tested without NFB. It is a very real problem
where one has a pure beam tetrode making 10% THD and where only 10dB of
global NFB is applied, and one will find some reduction of 2H but increase of
3H and 5H and perhaps other intermodulation IMD H that were not produced
by the beam tetrode. Where the signal is music with perhaps 20 frequencies
present, they all react to modulate each other even without any NFB applied
and when NFB is applied in small amounts the fidelity betterment is less than
expected if the THD for a signal tone was 10% to begin with. With 10% before
FB and with only 8dB NFB applied, the reduction of THD and IMD will not be 8dB,
and the sound may not seem any better. But a tube making 10% THD without
NFB will measure 1% THD with 20dB NFB.
The "second order" IMD H products may be less than 1%.
But as the H number rises, their audibility increases by factor of N squared / 4,
where N is the H number. Hence it takes very little 7H to make an amp bad where
the input signal = 100Hz, and the 7H = 700Hz. If you have 100Hz + 500Hz present
at high levels then you get IMD H at 400Hz and 600Hz and if these H are NOT
musical tones in the music scale their presence lessens perception of fidelity
because they are not in harmony with music.
The THD if considered alone make little difference to perceived fidelity if it is
below a few % because most musical notes have many H already present and
altering their levels by a few % makes little difference. But where the THD
does measure above 0.5% the perceived fidelity begins to suffer because of the
INEVITABLE production of IMD as a result of the non linearity which we measure
by using a single pure sine wave input to assess THD and the resulting H spectra.

With music, if you could listen to all the IMD products and THD products but
without the original undistorted music, you would hear a bad sounding non
harmonious noise rising and falling in level according to the music levels.
It is disgusting junk that poisons our enjoyment.

Applied FB should effective over a wide F range, and open loop gain of the amp
without any FB should be wide as possible, and phase shift minimized between
10Hz and 30kHz. What is needed to minimize production of IMD with output
tubes is to get them to be linear by NFB applied locally without involving global
NFB around 3 amp stages. The use of CFB and UL taps convert non linear beam
tetrodes and pentodes to being quite different in that the initial spectral complexity
is reduced and simplified more than NFB theory would predict.
The amount of global NFB applied can then be quite low without causing significant
"extra H" IMD products providing we make the input and driver stages linear.
In most tube amps, THD produced by driver tubes is much less than output tubes.
So multigrid tubes benefit with UL taps, and more with triode connection,
and even more with CFB with a tertiary winding on OPT. Adjusting the signal
level applied between screen and cathode is very important along with
setting the screen grid Vdc, Eg2, to be just right.
Usually, although most beam tetrodes and pentodes produce atrocious
THD levels of up to 15% in pure beam / pentode mode at just before clipping,
they can be cajoled into producing less than 2% with local NFB. It means their
gain is reduced from say 20 to 3, which means that instead of 10Vrms grid
drive we may need 70Vrms, but this is easy, and at low THD below 0.5%.
Is it possible to use the same tube and OPT as in Fig 1 but have
local CFB in the OP stage?

SE amps with CFB will normally have the OPT primary winding divided into
2 windings, with between 70% and 90% of the turns
in "the anode circuit" and
the remaining 30% to 10% of turns in "the cathode circuit".
The signal Ia flow and Idc flow is
equal in both windings except for where a
minor difference occurs with dc and ac screen current.

An example is at

There is no strict ideal ratio of anode turns to cathode turns and I have
used up to 2:1
successfully in my SE32
amps, 2012 version, with 13E1 tube, ie, anode turns
= 66.7% of total, cathode turns = 33.3%.

Hardly anyone has the patience, time, money or skill to make their own well
designed OPT to suit multiple paralleled
output tubes with CFB windings.
Few companies produce any OPTs with any CFB windings as a stock item you
buy off the shelf. So OPTs for CFB use usually have to be custom wound at
a far higher price than "normal" stock items, and you wait months and deal
with bullshit artists. There is little doubt my efforts to make the SE35 and the
SE32 creates sound quality to keep amp owners happy ( 2014 ).

But now let us question Fig 2 above.
Could the standard off the shelf Hammond OPT be used to provide local
cathode feedback in the
output stage, despite the fact there is only ONE
"anode" primary winding?

Is it possible to "build the tubes around the OPT"?

Here is my schematic......

Fig 4.
The above Fig 4 uses a floating B+ anode supply which in effect works like a 400V
battery between the tube anodes
and the anode connection of the OPT primary.
The UL tap is connected to 0V.

The normal connection for B+ is taken to cathodes and their biasing R&C networks.

This arrangement has anode Idc and signal Iac flowing around the whole series circuit
formed by tubes, primary winding, and PSU, just like any conventional amp.
But here I have the whole primary at 0Vdc potential with UL tap taken to 0V and the
non grounded power supply is moved to between tube anodes and anode connection
of OPT. The tubes are operating in exactly the same manner as they would in Fig 2
and relative signal voltages are the same.

There is a total of +229Vac across all turns of the primary. But -91Vac between
cathodes and 0V, and +138Vac between anodes and 0V.
The -91Vac and +138Vac
have opposite phases.

To make this this happen, -22Vac must be applied between grid and cathode,
and this must be added to to the cathode voltage of -91V t give -113Vac.
((( Some would ask, if you add -22V to -91V, don't you get -69V? or +69V?
All I can say to them is that they need to study more, observe more, measure more,
build more, and then finally what I am saying might just make sense. )))

A normal UL stage as in Fig 2 has the screens taken to a UL tap at same +Vdc as
the anode. In Fig 4 above the relative UL signal voltages remain the same,
so screens must be be connected to the UL tap. But the screens must be at a high
+Vdc potential so there must be an additional fixed B+ supply rail applied between
screens and UL tap. This B+ rail supply is just a normal grounded B+ supply
which will be also used for the input and driver stages.

The final result has an output stage with quite low overall gain compared to the
normal UL OPT stage.
The conventional UL stage has voltage gain of about
10, but the CFB stage has gain of only 2, with 113Vac
needed to produce 229Vac
across the primary of OPT.
There is a very worthwhile advantage because the
effective output resistance of each EL34 reduces to about 290 ohms.

The pair of tubes makes the total tube Rout = 145r, and damping factor with
RLa = 2k5 becomes 2,500 / 145 = 17.
The reduction of gain and distortion in the
OPT stage is a factor = 22 / 113 = 0.195.
This means that the typical UL THD
of say 6% at just below clipping is reduced to about 1.2%.

But there's a price to be paid for the output tube performance betterment,
and the drive voltage applied to the
output stage must be 5 times greater, and this
must be created without much distortion. My usual method
is to use an EL34 in
triode mode with choke+R for the dc feed to easily make up to 140Vac at less
than 1.5% THD.
An example of this is at my SE32 amp at

There is an additional benefit not easily recognized. The rate of 2H production
in the output stage is similar to
the rate of 2H produced by the driving EL34 triode.
Therefore there may be considerable cancellation of the 2H
produced in the output
stage. This cancellation is load dependent, most reduction of THD by 2H
occurs in the middle range of load values, and the maximum
cancellation causes THD to become as low as that
found in a well designed
class A1 push pull amp.

There is no absolute need to have a floating PSU B+ supply for the OP tubes
as in Fig 4.
A normal grounded B+ supply may be used as I have it here....
Fig 5.
Fig 5 shows a complete amp using a conventional SEUL OPT with just one tapped
primary winding.
I designed this schematic for my friend Vali in Romania.
He wanted to make a 2 channel chassis and using 3 x KT88 for each channel.

He found a supplier for SE OPT for 30W+ into 1k8 : 8r0, with UL taps at 25%
and 50%.
The properties of the OPT
were analyzed and after some calculations, I thought
3 x KT88 could make 36Watts into primary load of 1,300, with sec = 5r8.
Thus he ought to get at least 30W at the 5r8 output load. The amp would
tolerate all loads between 3r0 and 20r0.

But I have since made a few improvements to the schematic and I know that
anyone else wanting to make such an amp may not be able to source the same
OPT which my friend Vali found. So therefore I have drawn the OPT having
UL taps at 25% and 50%, and with tapped secondaries for speakers nominally
4r, 8r, or 16r.

In Fig 5, there is NO FLOATING B+ anode supply as shown for Fig 4, and there
is just ONE B+ supply used for all B+ rails for the amp.

The B+ apply of +410Vdc is applied to the 25% UL tap so anode current flows
from B+ to the normal anode connection. The KT88 anodes have 75% of the
total primary voltage, in this case 162Vrms.
Having Idc flow in 75% of the primary winding
means the DC core magnetization
Bdc is only 75% of what it
would be if the DC flowed in all turns. Instead of say
Bdc = 0.7 Tesla, it is 0.53 Tesla.
Bac can be 133% higher at 0.93 Tesla, instead of 0.7 Tesla. So the frequency
of core saturation will be
0.75 x original design Fsat, so if Fsat was say
25Hz for full Po level it would be reduced to 19Hz, a considerable improvement.
OR, one could have Fsat at 25Hz but at a considerably higher signal voltage.
In other words, with the same Idc in fewer primary turns the output power can be

The signal current produced by KT88 flows in ALL the primary turns including
those without any Idc. The original B+ connection on the primary is taken to the
KT88 cathodes via C20, 470uF. So where does the Idc flow from cathodes?
It flows down to 0V and back to PSU via L2, 5H choke.

In this case, the OPT had a 25% UL tap which is an ideal % for deriving CFB.
Some available OPTs may have UL taps at somewhere between 25% and 50%.
The 50% UL tap is in fact the CT of the primary winding.
If the 50% UL tap was used for B+, then in this case the anode signal would become
108Vrms- and cathode signal would be 108Vrms+, and the the grid-to-cathode
drive signal would be about 20Vrms+ so the EL34 in triode mode would need to
make 128Vrms+ which is getting a bit high to achieve at low THD < 0.5%.
But in my SE32 with 13E1 I have an EL34 in triode which must make over
100Vrms to drive a tube with 33% CFB and it sounds just fine while measuring
just fine.

The KT88 in fig 5 are working with relative electrode voltages identical to a
conventional UL amp with normal 25% UL taps. The Fig 5 circuit has 25% of
the Va-k fed back to the cathode.
This cathode signal is in series with the input
grid signal. The amplitude of the cathode feedback is sufficient to be a very
effective amount of local
NFB which reduces the THD by a factor of more than
1/4, and reduces output resistance of the stage to about 1/2 that of
triodes. But the
maximum drive voltage needed by the output stage is still at a moderate level of
72Vrms and very easily
produced by an EL84 strapped in triode mode. 

The L2 5H and C20 470uF form an LC filter with a pole at 3.3Hz, but I believe this
F is so low and that cathode output resistance is so low that resonant effects of the
C&L are suppressed, ie, damped, and LF stability should not be threatened.

L2 choke should be at least 5H, and with Rw about 50r max. However, if Rw was
100r, then the top of the choke would be at +26.6Vdc and the cathodes would be
at +60.3Vdc, so the B+ supply would need to be raised from +410Vdc to +424Vdc.
Amps like this should always be built with a few taps 15Vrms apart at the end of the
winding to allow B+ to be varied depending on winding resistances and to
accommodate changes to the total load value seen by tubes to best suit an OPT.

L2 choke shunts the cathodes to 0V with its increasing low reactance as frequency
becomes lower.
This may seem to reduce the NFB and increase KT88. The effect
is also increased by C20 whose reactance increases as F gets lower. But at very
low F of say 3Hz, the OPT primary reactance shunts the tubes to an extent that
whatever weird behaviour occurs with C20 and L2, it is of no importance.

The C20 should be at least 470uF rated for 450Vwkg, or a pair paralleled. it
should be bypassed with 2uF to shunt the HF impedance of the electrolytic cap.
The choke will saturate at some low F below 14Hz. It needs to be designed to
take up to 100Vrms with 265mAdc present without core saturation above 14Hz.
many "off the shelf chokes" may not fulfill the condition, so perhaps a pair of
smaller 2.5H chokes in series would be better.

At 20Hz, the L2 5H has XL = 630r, and C20  XC = 17r. The inductance of the
whole OPT primary should be no less than 10H. 
The Lp acts to shunt cathodes
to anodes and reduce the Va-k. as F goes lower. The CFB tries to
maintain the
the Va to k with NFB action, but the tubes saturate at 20Hz at full Po so
output voltage is limited and the L2 and C20 have little effect on overload behaviour.
This SE amp, like all others, should have an C&R HPF at input plus LF gain shelving
R&C network after V1 to reduce the open loop gain to very low levels at below 5Hz.

The use of a choke in Fig 5 is in effect a "cathode current choke sink" of DC from
KT88 cathodes to 0V.

It is a similar technique to using the well known ( but seldom used ) choke feed from
B+ to the anodes with cap coupling
of anodes to OPT. It is also know as "parafeed".
The choke is feeding Idc to anodes and the choke inductance is in parallel with
capacitor coupled primary inductance. In Fig 5, the L2 5H is in parallel with the
25% of turns of the OPT, and inductance of 25% of turns will be about 1.3H,
so the extra L2 inductance has negligible extra loading effects at very low F.

The choke L2 is able to do some of the function
of an air gapped OPT and therefore
get more audio power from the OPT.

It also would be possible to have a CCS ( constant current sink ) using solid state
and all similar to what I have in Fig 6 below.
But the SS devices must be arranged very carefully because the + / - cathode V swing
means the cathodes go to a negative peak voltage well below 0V so there must be a
suitably designed negative voltage rail.

If the 36W CFB amp in Fig 5 seems like too much trouble, or you cannot source
the esoteric rarely ever available well wound OPT, then maybe something else
will sound well.....

Fig 6.
Fig 6 shows a "nearly conventional" SEUL amp and a Hammond 1640SEA with
40% UL tap.
The specifications for the Hammond 1640 tranny has P : S ratio 1,250r : 4r, 8r, 16r.
The secondary is one winding for 16r and 8r is a tap at 70% of turns and the 4r is
at 50%, ie, the secondary center tap.
The specified maximum Idc is 200mAdc. This means that peak current change in
class A = +/- 200mA peak.
So maximum Po can be calculated = 0.5 ( 0.2 x 0.2 x 1,250 ) = 25Watts.
This is a little bit less than what Hammond say, as it is a 30Watt rated tranny.
But we can forgive them this minor discrepancy.
With 25Watts, primary signal voltage = 176.7Vrms, and there is 20Vrms across
16r secondary.

Now 20Vrms is a large enough voltage to be usable for cathode feedback if
we wished because it is 11.36% of primary signal voltage. Anything over 10% is
useful. If we had just one KT88 producing 8.3Watts with OPT for 3,750r : 16r,
the primary voltage would be the same 176Vrms, but 16r sec would have
1.5Vrms, only 6.5% of primary voltage which is an ineffective amount of CFB.
The normal UL tap and GNFB loop would suffice.

What would happen if we used the 1640 secondary as a CFB winding AND
speaker winding?
The P:S turn ratio will change so that the transformer P:S ratio = 1,549r : 16r,
because now the signal current of tubes flows in secondary turns.
This current is much smaller than the speaker current so the thicker sec turns can
cope easily.

With primary load at 1,549r, it suits the use of 3 x KT88 with each tube seeing
4,647r, a very nice load for KT88 operating very comfortably with Ea at +310V
and Ia at 60mA. EL34 would also work very well. Pda + Pg2 per tube
= 18.6W + 1.55W = 20.15W.
The Idc total for 3 KT88 is 195mAdc, nearly Hammond's maximum allowed.

If the 195mAdc flows in 16r winding which may have Rw = 0.8r, the Vdc across
winding is about 0.16Vdc, and be applied to a speaker, unless a DC blocking cap
is used, which we will not to because there is insufficient Vdc to polarize a large
value electrolytic cap. In addition, the Idc in sec raises the Bdc of the core which
will saturate at a higher frequency. So there's two good reasons NOT to have
the tube Idc flowing in the sec. But the tube signal current "helps" things to happen.

Individual cathode biasing R&C networks are good for paralleled tubes always
operating in class A. But we could have constant current sinks instead of resistors.
The Fig 6 shows IRF610 used for each output tube cathode. But other CCS with
Darlington pair connected
bjts could be used because the base input resistance is
very high and voltage across R27, R31, R36 remain constant.
The cathode "bypass" caps are still required for each cathode, but instead of being
taken to 0V they are taken to the top of the 16r secondary winding.
The three CCS each have effective collector resistance > 50k so having 3 in parallel
makes a very high cathode load which we may consider has negligible effect in any
considerations. The three CCS act like a choke in Fig 5.

Notice that the 3 KT88 cathodes will settle at about +35Vdc. Just exactly what Ek
will be may vary but I expect =35V, but samples may vary. Now the cathodes have
20Vrms at clipping, so the V-swing is +/- 28V peak. The mosfet drain connections will
also have +/- 28V peak, and the minimum voltage across the mosfets should not
become less than about 10V. The Vg-s should never go negative.
I have the mosfets set up with gates at -14Vdc, and IRF610 data says gate bias will
be -4Vdc approx at low Id. So sources should be at -10V, which is -45V below Ek,
enabling +/- 28Vpk cathode swing. I've chosen to have the required negative rail for
mosfets at -24Vdc. Its not too hard for anyone to make an unregulated -24Vdc rail
and for 200mA. I should not have to spoon feed you such a detail, and I just won't.

The Fig 6 has a large total amount of applied NFB in 4 "loops". Each can be
considered, and perhaps discarded if deemed unnecessary.

Loop 1.
It is the use of the UL tap from primary winding to supply KT88 screens
with what is normal Ultra Linear screen FB used now since about 1955.
The effect of this connection makes the effective Ra of each KT88 = approximately
3k0, a huge reduction from the pure beam tetrode Ra of 24k. Odd number H are
reduced and spectra is brought closer to triode operation.
Loop 2.
Local Cathode Feedback from OPT is used to reduce the Ra to less than
KT88 triode value. The CFB  is an external loop involving linear working transformer
windings so that all even and odd number H are reduced by the same amount.
One could say nobody needs to have any more NFB in such an amp but the
grid input to KT88 requires 38Vrms, so we MUST use a driver tube because a preamp
or PC sound card cannot make a linear 38Vrms.
Loop 3.

Screen FB to EL34 is used from the OPT secondary.
Instead of connecting
the EL34 to anode for simple triode operation, it is bypassed with C12 to the CFB
from OPT secondary. One might ask why, well, it can be done, so let us consider
the results. The EL34 has to make 38Vms at anode and although triode strapping
would work just fine, we could have its screen fed with a signal voltage that is less
than its anode signal so the EL34 is working in Ultra Linear mode. There is a
convenient signal available and of the correct phase which is the OPT secondary
signal of 20Vrms that is applied to KT88 cathodes. So EL34 has 52% UL operation.
The load for EL34 is the R17 and L1 in series which becomes high impedance load
for most of the audio band. The EL34 will be found to be very nearly as linear as
triode strapped with such a load. Gain g1 to anode will be found to be twice that of
triode. Now within the OPT secondary signal there will be distortion products
generated by KT88 AND those produced by the EL34. All these distortion products
are amplified x about 10 by the gain between g2 and anode to create an "error" signal
that when applied to the KT88 grids they are then amplified to oppose their own
production. The screen NFB is effectively about 20dB NFB, although not a most
perfect form of FB, because the EL34 cannot ever provide less THD than when
triode strapped and with the high RLa load value I have used. But a typical EL34
in triode mode with RLa > 20Ra can make 100Vrms at 1% THD, and at 38Vrms
perhaps 0.3%, mostly 2H and with 52% UL operation it may be 0.6%. But the KT88
may produce 2% THD, much more than the THD of EL34. The effect of this screen
FB will reduce THD from 2% to about 0.4% at least, a huge reduction.

OK, so now the amp makes 25Watts at 0.4%, and we need only 2.2 Vrms input to
the EL34 grid so anyone may try all this without adding yet another input tube V1.
At normal listening levels of 1 Watt, THD can be expected to be 0.07% which is
an excellent measured result for most SE amps, and I suggest ppl try it out.

What other SE amp amplifier has just TWO active devices with good linearity?
You say there are 4 tubes total, but the three KT88 act as one because they are
paralleled, and they could be replaced by 13E1, or 4 x EL34, and all I'm doing
is exploiting the screen properties of an EL34 so it can be both input and driver.
This is possible because the ratio of g1 gm : g2 gm is quite low. Put another
way, the g2 gm is a useful high value which can actually do a lot. Suppose we
used a 6BX6/EF80 instead. The g1 gm could be 5mA/V, quite useful, but g2 gm is only
0.083mA/V, and the fed back THD content is not amplified many times. The EL34
performs far better, even though it is a power tube. Other suitable tubes for screen
FB applications are 6CA7, EL84, EL86, possibly EL36 / 6CM5.

Loop 4.
Global NFB from OPT sec to V1 cathode. V1 is a 12AU7 low twin triode with = 17.
Gain with CCS anode supply via MJE350 is about 15x. The sound of this tube is usually
just worth dying for. Here is has little to do, but the amount of NFB = 14dB, so the 0.4%
I spoke of above is reduced by about 1/5 to 0.08%, say 0.1% at 25 Watts, and at 1Watt
THD may be 0.014%. Such figures are typical of very well made pure class A1 PP triode
with 2 x KT88 in triode and with 20dB global NFB.

However, I have to say, "GEE, what a huge amount of NFB!"
Is it all really needed? Could it ever really be used? Well, in fact, possibly because I
have NOT BUILT THIS AMP, but I can see already that there could be oscillations 
at both LF and HF just outside the audio band. The fact is that the Hammond 1640SEA
does not have extremely low leakage inductance. Its barely low enough for general
conventional use let alone for the "sophisticated" schematic I have proposed here.
But if anyone made an OPT with twice the amount of interleaving used by Hammond,
they may surprised by what might be done. In general, I have found the 16xx series
SE Hammond OPTs to be very useable and good sounding.

In any amp, as the total amount of open loop gain without any FB increased, any application
of NFB tries to extend over a wider range of F thus extending bandwidth beyond the
open loop bandwidth. For example, if OL BW is from 30Hz to 20kHz, 10dB applied GNFB might
extend BW from 10Hz to 40kHz. 30dB GNFB may increase BW to be 2Hz to 150kHz.
But the phase shift of the open loop amp at LF and HF will cause the applied FB to become
positive and hence cause oscillations unless gain shelving networks are used to reduce the
phase shift and reduce the open loop gain outside the audio band. There is no need at all to
have a high amount of NFB applied outside the audio band.
In the Fig 6 amp, probable open loop F1 pole = 40Hz and F2 pole = 5kHz to get LF and HF
This means that the high amount of NFB only applies to the band of 40Hz to 5kHz.
However, perhaps music is better and amp more stable to have less open loop gain and
less total NFB, with slightly more THD and slightly lower damping factor and open loop
F1 and F2 further apart at 15hz to 20kHz, resulting in final bandwidth of 7Hz to 65kHz.

There is a simple answer, just leave out the screen NFB and use the EL34 strapped
as a triode. EL34 triode grid signal will be 4.5Vrms. 12AU7 can have a higher FB signal
applied to its cathode = 1.5Vrms. Total Va to Vk = 6Vrms, so Vg-k = 0.4Vrms, so input
signal to 12AU7 grids = 1.9Vrms. This seems high, but is OK, and the amount of GNFB
= 13.5dB, and the amp's THD at 25Watts = 0.4%, quite good.

Now take note that in all my amps I have supplied to customers, I have provided
protection circuits to prevent damage to OPTs and other parts if one or more output
tubes decides to conduct far too much Idc because for one reason or another,
the grid bias voltage ceases to control the Ia flow. Hence the note on the drawing
about tp1, tp2, tp3. A KT88 which becomes a short circuit could damage the cathode
CCS mosfet. I show no cathode fuse because before it would blow, the mosfet
may fail with excessive Vd-s and enough current to fuse the mosfet innards.
There is a zener diode in the mosfet, but it could fry easily. 

OK, now you have seen the "traitors way to use Squalid State devices" in a tube amp.
How could I ? Didn't I know they don't belong ?. Well, OK, I hear the complaints,
but constant current sources or current sinks using SS ARE OK, because they are
so good at providing an extremely high impedance source of current and as such
cannot have any effect on the signal. The SS devices are friendly slaves, totally
under control of tubes, and they allow tubes to work better than they other wise might.

Now, to appease the "Society Against Solid State", I have another similar schematic
below which is similar to Fig.
3 x KT88 may be used with Hammond 1640SEA but with a single choke in cathode
circuit as seen in Fig 7 below.....
Fig 7.
Fig 7 is the same as Fig 6 but has SS current sinks replaced with a choke.
Ah, so simple! - until you start thinking about a choke, and go shopping for it!

If you build this type of circuit using an available-off-the-shelf-stock OPT
then all I said above about how it works applies.
The Fig 7 amp like those above have a high total amount of NFB.
But to prevent oscillations with a high amount of NFB you need to be able to
Nyquist and Bode graphs and theory intuitively. Its too difficult for me to
write a book about it right now and have you read 2 pages to understand,
So, do your own Googling with Bode and Nyquist, and lose a week of your life
trying to understand WTF it means.
Basically, as one increases the amount of NFB and or the amount of open loop,
gain, ie, the voltage gain from input to output without any NFB applied, then the
amp becomes more prone to oscillations due to the phase shift caused by
reactive circuit elements L and C reacting with R. For most tube amps there is
ONE F where there is no phase shift and it usually is between 200Hz and 2kHz.
But below 15Hz and above 15kHz, and where the open loop gain has become
attenuated by R&C Miller effects, C&R couplings, OPT primary L, leakage
inductances, shunt C, etc, and where phase shift reaches 180 degrees and where
gain exceeds 1.0, or "unity", then the amp will oscillate.
The only way to prevent this is to use Zobel networks at output and perhaps in
output anode circuits and at V1 output.

In Fig 7, the R&C values affecting stability ARE A GUIDE ONLY, and YOU have
to figure out the best values for unconditional stability.
The critical R&C parts are R12&C8, R16&C9, C6&R13, C19&R35, C21&R37.

If you find you just cannot stop oscillations, then abolish the screen FB to EL34.
Connect bottom of C12 to 0V instead of to the FB from OPT.
Reduce value of R16 to 1k0 to increase GNFB.
But you will still have to optimize the bandwidth with a pure R load and
get a good square wave without severe ringing or HF oscillations with any
pure C load between 4uF and 0.047uF.


Cathode FB with a custom wound OPT.

It is time to mention my favorite output stage configuration for both SE and PP tube amps.
Such stages have 2 primary windings and the usual secondary. One of the primary
windings has up to 1/3 of the tuns of the other and is called a cathode feedback winding,
or tertiary winding.
Conventional local CFB in SE output stage may be used if you can find an OPT which
has a separate cathode feedback winding
but really it is just part of the whole primary
winding because the tube signal current
flows in both windings. But the CFB winding
is usually at an earthy Vdc potential, while anode winding is at the B+ Vdc.

Fig 8.

Fig 8 has two examples of CFB in output stages.
Fig 8 Left side.
The total signal voltage across both primary windings = 184V + 46V = 230Vrms.
20% of the total turns are in the CFB winding and 80% in the anode winding.
I show the formula for working out the effective UL % as
UL% = 100% x ( V ULtap + Vk ) / ( Va + Vk ) where the signal Vrms are measured
between each of the screen, anode and cathode terminals to 0V.
In this case the V UL = 0V because there is no UL tap. But the KT88 is still
operating as though there is a UL tap because a signal of 46Vrms exists between
screen and cathode. Pure beam tetrode or pentode with CFB is achieved by
bypassing the screen 100uF to the cathode so there is no signal voltage between
screen and cathode. It is usually found that the operation of pure beam or pentode
with CFB merely reduces the complex THD spectra of the pure beam of pentode.
The use of an effective UL tap by means of bypassing the screen to 0V changes
the THD spectra towards triode with less odd H. Usually, where the CFB % of
total primary turns is close to 10%, the effective Ra of most beam tetrodes and
pentodes to be about equal to the triode Ra. When the CFB % is raised above
10%, the Ra can become considerably lower than triode Ra. Optimum CFB %
will be about 20% but can be from 10% to 40%.

The higher the CFB%, the higher the drive voltage needed so the driver should
have low THD.

Usually, the driver is a trioded EL34 or EL84, but could be 45 or 3A3, although
real triodes have very low gain so a high gain input triode is needed. So in my
view the EL34 or EL84 are winners. Inevitably, driver triodes produce some
2H. If the driver has to make 75Vrms, expect 0.7% 2H, and the output tube
may make 1.5%. The 2H generated will cancel, and 2H total at output becomes
1.5% - 0.7% = 0.8%.
However, the 2H of all such CFB use with KT88, EL34 varies with load, and there
is an RLa value just above the RLa value for maximum possible Po where 2H = 0%.
Below this RLa value the 2H cancels, above the RLa value it adds, because the
relative phase of the 2H of the output tube is the opposite phase !
It is more fully explained at my pages on the SE35 amp.

Fig 8 Right side.

The operation of the right side has effective UL % = 40% because there is a UL tap
at 20% of total primary turns and there is 20% CFB.
With class A operation there could be a UL tap at 40% of total turns which would
increase the effective UL total to 60%. Going beyond this % is pointless because it
restricts the Po available because operation becomes  too much like a triode
where the Ea negative swing is restricted by grid current onset.
But at over 40% effective UL, the odd H are very much reduced leaving triode like
THD which is mainly 2H, and the phenomena of having 0% 2H at a higher RLa
diminishes and you get good cancelling of 2H produced by the driver tube.

For SE class A1 and with most beam tetrodes and pentodes the operation with
about 20% CFB and UL seems to give extraordinarily good sound, so I have
been told.
"Shunt feed", aka "parafeed" output stages.
Shunt feed output stages have seldom ever been used for hi-fi because you
need a large air gapped choke with high Idc flow, high inductance and not likely to
saturate with a high anode signal voltage at above 14Hz. This choke provides a high
impedance source of Idc to the tube while not consuming any audio power produced
by the tube. The size and its weight may be larger than the OPT used.

The audio signal power of the tube is conveyed to a primary winding on an ungapped
OPT via a coupling cap. The OPT primary may be connected to 0V at one end,
and the cap has a large Vdc voltage across it. The OPT can be a normal PP OPT
which is very easy to source, and its Lp inductance is usually far more than is actually
needed, and it usually does not saturate above 20Hz at full Po. The coupling cap
capacitance value must be high enough to have a resonance with OPT LP at below
3Hz. So if Lp = 30H at low signal levels the C should be 100uF. One might use a
number of C in series with resistance dividers to ensure equal Vdc is across each
cap and Vdc never exceeds 2/3 the Vdc rating for the cap. Electrolytic caps are
needed, but each must be bypassed with 1uF plastic film caps.

The Shunt Feed a
dvantages are :-
The OPT has no huge Vdc potential between primary and earthy secondary or
0V or chassis.

The OPT may have no air gap. Laminations may be maximally interleaved.
OPTs meant for PP amps may be used.

For the maximum Po, the OPT may be smaller than if the OPT was a conventional
air gapped OPT with Idc flow. 

The Shunt Feed disadvantages are :-
A large choke or 2 or 3 series chokes which may have the greater size and weight as
the OPT must be used between
anode and B+ to provide a high impedance feed of
Idc to anode.
Capacitor coupling from anode to OPT must be used which introduces yet another
time constant filter behaviour which can affect the LF stability of the amp when NFB
is used.
GNFB application with a choke feed amp may be more difficult because
unconditional stability must be assured. The amount of GNFB is limited by the number
of C&R and C&L couplings and L&R shunts. But with triode output tubes 12dB GNFB
with no output load is usually enough, and possible, when
LF oscillations are most likely.
More careful arrangement of open
loop gain shelving R&C networks are needed.
But for HF stability, there is no extra stability problem compared to using
a simple air
gapped OPT and conventional GNFB arrangement.

The capacitors from anode to OPT must be chosen carefully and used with respect to
their voltage ratings.

Fig 9.
Fig 9 shows how choke feed allows the use of a PP OPT without DC flow through
the primary winding.
SEUL Choke feed, left.
I show the anode choke L1 = 55H at 80mAdc, and estimate Rw at 200r. You
cannot purchase a Hammond 55H choke equal to what I say is needed.
Hammond have have 193C 20H with Rw 180r so 3 in series are needed for 60H.
But then Rw total = 540r giving 46Vdc and B+ must be raised from +430V to +460Vdc.
So before you copy what I show, be sure you know ALL about what you are doing!

The screen is cap coupled with 100uF to the CT of OPT primary giving 50% UL.
The screen requires
low Idc of 4mAdc choke feed through L1, perhaps a Hammond
155C, with Rw = 2k7, L = 60H.
I estimate 50H is plenty, and to get the Eg2 nearly equal to Ea, and to prevent fusing
the L1 choke if screen shorts to cathode or 0V the series R = 1k0, 0.25W rated.
If 300Vdc appears across 1k0, it fuses quickly, and a new R costs 10c.
All electrolytic caps used should be rated for 350Vdc, and each bypassed with 2uF,
plastic film types rated for 630V.

The C value for coupling anode to OPT seems quite high, 110uF in fact.
The LF pole formed by 110uF and anode load 4k2 is 0.34Hz, but more important is
the pole of HPF formed by the 110uF and OPT primary L which may become 50H
at low levels of signal typically used for listening. This F pole would be 2.14Hz,
and the peak in response is prevented by the very low impedance of choke
and OPT at 2.14Hz. Therefore enough global GNFB should be able to be applied
while maintaining unconditional LF stability with R&C critical damping networks
in input - driver stages.
Triode Choke feed.
The triode use of KT88 can have B+ at a higher voltage for best triode performance.
a 300B could also be used instead of the KT88, although the B+ has to be +40Vdc
higher because the Ek bias will be about +88Vdc. The triode use does not need
any screen choke, but anode choke needs to have high L value as for SEUL.
All the same comments made about SEUL apply about the L1 and caps.
Triode is easier because the screen is simply strapped to anode via its stopper
resistance of 220r. 

845 Choke feed.
An 845 may happily work with an easy to buy Hammond 1650P OPT rated for
60W, but used to make about 21 Watts. The load match is 6k6 : 4r,8r,16r.
Fig 10.
To make an air gapped OPT for the Fig 10 output stage with an 845 is extremely
difficult for 99% of DIYers. Making their own chokes would also be very difficult.
But they could purchase
1650P OPT and three Hammond 193C chokes, and the
required capacitors.  Possibly MUCH better chokes and PP OPTs could be found
than made by Hammond, but appraisals of whatever brands are used demands the
full understanding of how the basic item properties affect the performance.
I won't suggest how an air gapped OPT or choke may be designed right now
but the method and examples can be found in my other OPT design pages.

I must mention the analysis behind Choke feed.
A choke is defined as a coil of enameled wire. Its basic properties are described as
an amount of pure inductance in Henrys in series with resistance = winding wire ohms.
There is also capacitance between turns resulting in a summed effect of an amount C
shunting the L. So all chokes have a parallel resonance between the L and the C.
For a single choke of 55H with an iron core, shunting C might be only 300pF.
The reactance XL of a 55H choke at 10Hz = 3,454r, and far less than the C.
The XL rises with F to a maximum of perhaps 500k ohms at the theoretical resonance
Fo at 1,240Hz. Above Fo, the XL reduces because the XC declines with rising F and
= 3,454 ohms at 153kHz. To extend the high XL the choke can be wound on a bobbin
divided into say 3 sections physically 2mm apart so that the 100pF shunt C of each
is in series with the other sections thus reducing C total to < 50pF.
In practice, the C has little effect when the choke is in the anode circuit of a tube
output stage.
Chokes oppose the flow of AC because the voltage across the choke sets up a
magnetic field which acts to oppose the flow. Therefore low current flows in the
choke across the audio band. The small amount of audio power that is lost
= Iac squared x Rw, and is usually negligible. The same applies to the inductance
of any OPT where the power lost as heat in the OPT = Iac squared x Rw.
The OPT and choke Rw losses are greatest at bass frequencies where Iac becomes
highest and XL is lowest.
The shunt feed choke may be two or 3 chokes in series to make up the wanted total
choke inductance as I have shown in Fig 10.
The total L is simply the sum of the individual L values. The L values don't have to
be equal, one could have a 40H choke plus 20H choke to make 60H.
But they MUST not saturate with the presence of large signal voltages at LF
and combined with high Idc flows.

For Shunt feed SE amps, one aims to ensure core saturation of choke or OPT does not
occur at F higher than 20Hz and with a signal voltage at the maximum Po level for the RLa
value which gives the highest Po at the clipping level.
For an 845, one might use RLa = 6k6, and get at least 21Watts so Va = 372Vrms.
The 1650P OPT is rated for 60W to 6k6 and OK for 629Va-a and saturation could be
at 30Hz. ( I am not exactly sure. ) But saturation F is a voltage dependent phenomena
and occurs independently to loading and currents. So at 372vrms, expect the Hammond
OPT to saturate at 18Hz. The maximum primary inductance at high Va-a is probably more
than 300H at say 1Tesla, typical with GOSS core material with high .
The choke has to cope with 372Vrms at 20Hz without total Bac and Bdc summing to
more than say 1.2Tesla which is a maximum for medium grade iron. Top grade GOSS
lams or C-cores may take only 1.4Tesla  before onset of core saturation and poor old iron
from 1950 might take only 1.0Tesla. If you make the3 choke, assume the maximum iron
is below 2,000, and that it saturates at 1.0Tesla. So one may find one has to wind TWO
chokes and connect them in series. One may buy chokes, but whatever is manufactured
or bought MUST satisfy the engineering design requirements, and whatever is used must
be tested carefully.

The total L shunting RLa is the choke L plus OPT Lp in parallel. In all shunt feed SE amps,
the air gapped choke will have much less L than the cap coupled OPT.
The total L should have reactance = RLa at 20Hz or lower F.
Minimum Choke L value = RLa-a / ( 20 x 2pye ).
For RLa = 6,600r, L should be 6,600 / ( 20 x 6.28 ) = 52.5H.
During the amp's life it will be used at less than 1 Watt, so the Vac across the L will be
less than 1/5 the clipping level. The inductance of iron wound coils varies with the
applied voltage, ie, the lower the Bac, the lower the permeability . Air gapped chokes and
air gapped OPTs have the least variation and reduction of L at low signal levels. So
we don't need to aim to make the calculated choke L higher to compensate the drop in
at low Vac levels.  Most variation of L occurs with non gapped cores for PP amps where
inductance at 0.05Watts, 18Vrms across 6k6 load, may be 1/6 of the 20Watt Lp level,
and so may be 50H. This L in in parallel with choke L and if the choke L = 50H then
the total L may drop to below 20H at very low levels. This WILL NOT cause any change
to the F response or bass performance because the tube Ra also shunts the L and the RLa,
and the 845 Ra = 2k2 plus 6k6 RLa makes the R shunting L = 1k7, so F1 pole with 20H
is at 13.5Hz. The 845 ( or any other tube ) will easily drive a low inductance load at very
low signal levels.
However, most people will wish to use NFB and the reduction of L at low levels causes the
F1 pole to rise at low levels despite the Ra and RLa shunting the L and the additional phase
may cause LF oscillations. Many old mass produced tube amps require a speaker to be
connected while the amp is turned on lest they begin to oscillate at some low F below
10Hz. I have often encountered such amps and the oscillations can be violent enough
to cause core saturation and heavy damaging tube currents, or else the amp oscillation
is limited to a low level because inductance rises with signal voltage so F1 moves down
and so does circuit gain so the amp in a state of equilibrium as a phase shift oscillator.
Connecting a speaker puts more R across L and reduces F1 and the amp stops
oscillating. Such amps have been designed by accountants. But now you see the need
for LF gain shelving R&C networks to prevent the oscillations at LF.

Above I said total L minimum = 52.5H. If the OPT Lp was in fact say 300H at 21Watts,
then we should prefer to ensure total XL = RLa-a at no higher than 20Hz.
Therefore the choke L should be 63.5H, so when in parallel with 300H the total = 52.5H.
Basically, we want all the choke L we can muster!

Design of a single choke could be....
Ia = 83mAdc, so chose wire size so current density = 2A/ for where Idc = 3 x Iadc.
For 249mAdc, wire size = 0.398mm Cu dia. Use high quality polyester-imide enameled
magnetic winding wire, 0.4mm Cu dia, overall size including enamel = 0.47mm dia.
Winding window available = 20mm x 72mm = 1,440 so random wound turns
possible = 1,440 / ( 0.47 x 0.47 ) = 6,518 turns. Expect to get 6,000 turns on.
Experience tells me that probably this will give enough L at 83mAdc, and with the correct
air gap. But YOU need to read my pages on choke design to verify that the choke design
is correct. The L is adjusted for a maximum by adjusting the air gap size with 83mAdc
present plus a 50Vrms Va signal at 50Hz, without any load connected at OPT.
Perhaps you may find choke L is higher than you wanted, which is great news, and bass
will be fabulous.

To make at least 2 chokes, one for each stereo channel, you would need about 2Kg of
new copper winding wire, 2mm fiber-board for making bobbins, and enough E&I
laminations. These may be taken from a large power transformer with a fused winding.
I have often used old laminations from defunct PTs for chokes.
To easily take the laminations apart, the transformer should be placed in a small wood
fire for just long enough to make the lams appear dull red. The varnish and all plastics
will be vaporized and burnt. The burnt out core is left to cool slowly and will easily
fall part when the bolts an wire are cut off. The firing is a messy process best done
late on winter nights if you have a fireplace. The heat may improve the magnetic
properties, but won't worsen magnetic properties.

There are other methods of avoiding the high +Vdc potential between an OPT primary
and the earthy secondary. The first uses a "floating" B+ Vdc supply 
so that say the earthy
rail of the supply connects to OPT primary at the normal B+ connection, while the other
end connects to 0V. The positive rail of B+ supply connects to output tube  anode/s.
This means the HT winding on PT and its diodes and filter caps and chokes all are
at an elevated B+ potential and all these items carry the anode signal. T
here is some
capacitance between all this hardware and 0V and chassis. There is capacitance
secondary HT winding and other PT windings such as mains windings which
may introduce noise
to the anode audio path.
Therefore the PT should have extremely good insulation between its HT winding and all
others especially if the B+ supply was 1,200Vdc for 845 or 211 tubes. If the maximum
possible peak signal voltage swing at any given instant was +1,100V, and the adjacent
mains voltage was -340Vpk, and the B+ was +1200V, then one could have 2,640Vpk
between windings. Hence the need for good insulation. To minimize HF diode switching
noise signal getting into the relatively high impedance anode circuit  the PT should have
an electrostatic shield. An ES is usually one layer of wire turns with one end connected
to 0V and other end left open. Many old radio sets had such a shield on their PT.
The amp heater windings can function as a shield if one whole layer of thick wire is
designed to be devoted to heaters which have a CT connected to 0V.
So thus the PT can become a custom wound item.

Building a PSU that is alternative to the normal arrangement of grounded B+ supply is
more difficult and expensive, and cannot improve sound quality.

Fig 11.
Fig 11 shows a KT88 operating in SEUL with a floating B+ supply, but also has
a choke feed. It is yet another way to avoid having a special expensive OPT with
air gapped core.
I bet nobody else has ever used the schematic I have here. Certainly I have not
ever seen a mass produced sample. Manufacturers always will try to avoid
excessive iron wound items and any extra components that could be avoided by
designing a good OPT with Idc flow and air gap. Often their attempts remind us
that an accountant designed the amp, not an engineer. Keen DIYers need not
heed any accountant's advice and they may be free to design according to basic
principles and be creative in the process.
The audio is amplified in the same way as for any standard SEUL amp.
But the choke feed is arranged differently to standard choke feed.
The B+ supply is floating, ie, not grounded anywhere, and I show the PT with its
HT secondary, diodes and 470uF caps and filter choke all operating as a +433Vdc battery
between a KT88 anode and a choke at 0Vdc potential. The Idc path is from positive rail
of B+ supply to anode then down to cathode, through R&C cathode biasing network, then
to 0V rail then through L1 "choke feed". The B+ PSU for +433Vdc has the anode signal
present at all parts, yet the floating PSU acts entirely without any interference from the
audio signals, and does not impart any noise into the audio signal path.
There is 0.85Vrms of 100Hz ripple across reservoir caps, and 1.1mV across the
pair of caps for +433Vdc.

The feature of noiseless floating B+ PSU is made possible with the use of an electrostatic
shield on PT between the B+ HT winding and other nearby windings with diode switching
Old amps and radios had a shield made with one layer of thin wire between Mains input
primary and HT winding. One end of the layer was taken to the amp or radio chassis,
with other end left open. But the tube heater winding of 6.3V can be made to occupy one
layer and have its CT taken to 0V rail and it acts as an ES between mains and floating
B+ HT winding. But an ES should be wound over this floating winding if additional B+
windings are which have diode rectifiers. An ES need not be a layer of wire, but can be
one turn of thin copper or brass foil of say 0.1mm thick and overlapped 10mm, but
prevented from being a shorted turn because insulation is between the overlapped
turn ends. A wire lead is brought out to 0V. Therefore the only energy transferred to
the floating HT winding is magnetic. I've shown a voltage doubler type of B+ supply
with CLC filtering and
1N5408 Si diodes, quite good enough.

The tube operates with Ea = 375Vdc, and with +418Vdc at its anode and with +43Vdc
normal cathode biasing and grid at 0Vdc potential. The anode signal is transferred to
the top of choke L1 55H which has one end connected to 0V. Notice that the Idc
through L1 produces a NEGATIVE Vdc across the choke. The higher the choke Rw,
the greater this -Vdc will become, and the higher the B+ supply voltage will have to
be to ensure the wanted Ea is produced.  So low resistance choke is wanted.

The L1 choke of 55H does not significantly load the KT88. 55H at 1kHz has
XL = 345k plus some shunt C of say 300pF = 530k. Now the proposed ideal RLa for
maximum Po = 4k2, so XL = RLa at 12.15Hz, well below 20Hz, so at full 12.6Watt Vo
level the KT88 does not put much signal current through L1 above 20Hz.
Fig 11 shows details of L1 and L3 chokes. L1 could possibly be 3 x Hammond 193C
20H chokes in series but then Rw = 540r total and the L1 Vdc = - 46Vdc, so B+
must be raised from +433Vdc to +464Vdc, which means +496Vdc must be produced
by HT winding so the highest voltage may not be high enough at 210Vrms.
The 193C have maximum Idc rating = 100mA, which is a bit low, and the DIYer can
make a MUCH better choke at home.

I hope everyone fully understands that if you change one thing among several
properties of one component, it can seriously affect more than one other thing and
perhaps ruin the
optimum operation of the circuit. SE amps with low max Po ability
need all the optimization possible so that a wide range of loads can be used.

The SEUL schematic has the anode signal appearing across L1, and to get the
wanted 12.6Watts of audio power there must be an R load across L1, and it is the
OPT which is cap coupled to the L choke with 1,000uF. The 1,000uF prevents any Idc
flow in the OPT primary which always remains at 0Vdc potential. Now  the top of choke
L1 is at -15Vdc and the 1,000uF cap may then be a low voltage type rated for say 35Vdc
and best are those with high ripple current rating normally used in PSU for SS amp circuits.
The Vdc available will adequately polarize the electrolytic to give linear operation. A 2uF
plastic film cap should be strapped across the el-cap to shunt its ESR, and make the cap
work as a pure cap to 10MHz.

The OPT I recommend is Hammond
PP type 1615A, rated 5k0 : 4r, 8r, 16r.
The primary CT is used for the UL screen connection. But the screen must have the same
+Vdc voltage as the anode, so it is coupled to the OPT CT with 100uF el-cap, also
bypassed with say 1uF/630V plastic film. To get the screen to be fed with Vdc = anode Vdc,
I suggest the L2 choke, over 50H. It can be a Hammond 155C, giving 60H, but its Rw is
high at 2k7, and with screen "stopper" resistor 1k0, and Ig2 = 5mAdc, the screen Vdc is
18.5Vdc lower than anode Vdc. Best SEUL operation occurs with Eg2 = Ea when nearly
all power beam tetrodes and pentodes need Eg2 = Ea lest the maximum Po be restricted
similarly to triode operation. Ea to Eg2 difference should be limited to less than 5% of Ea.
There is no doubt that a keen DIYer could make a slightly larger choke than the Hammond
155C and with MUCH lower Rw using thicker winding wire and a bigger core.
Such a choke is less prone to fusing open if a screen becomes shorted to cathode,
or the coupling el-cap to OPT becomes shorted.
This L2 choke also has 1/2 the anode signal across it, so it must not saturate with
excessive total Bac + Bdc.

Fig 12.
Fig 12 gives the choke feed arrangement for floating B+ and triode operation.
It is very similar to Fig 10 for SEUL. The KT88 screen is connected to anode
with 220r stopper, so that the screen is being fed ALL the anode signal.
The screen g2 acts as an additional control grid in the same way an anode
in a "real" triode acts to partially control Ia in conjunction with the g1
control grid. Triode operation gives the maximum possible amount of screen
NFB introduced into the tube, so the voltage gain is lower than UL.
The SE triode operation is obviously simpler than SEUL, and the maximum
output power is not much less than SEUL. But notice that to get somewhere
near the SEUL Po, the triode Ea and Ek must be higher with Ia slightly lower
for the Pda to be 30Watts, same as for the SEUL use.

If the KT88 strapped as a triode was used with exactly the same Ea and Ek as
shown for SEUL in Fig 10, the Pda + Pg2 would still need to be kept at 30Watts.
The Ia+Ig2 = 85mAdc, so for maximum Po the RLa = ( 375 / 0.085 ) - ( 2 x 1,100 )
= 2,211 ohms. This RLa is only 2 x Ra of the triode. THD is higher than SEUL.
Max Po at anode = 8Watts.  Damping factor = RLa / Ra = 2k2 / 1k1 = 2 which and
no better than the SEUL with 4k2 load and 50% UL tap.
Then the OPT which was chosen for UL has nominal ZR = 5k0 : 4, 8, 16,
and for RLa = 2k2 triode anode loading, required secondary loads = 1r8, 3r5, 7r1.
But the winding losses will double to become 20%, so therefore the max Po with
triode at the secondary becomes about 6.4Watts only.
Therefore I hope everyone can see that to get the KT in triode to work nearly
as well as SEUL, you MUST use a higher Ea and lower Ia to suit the load which
is nearly 4 x Ra, and OK for good triode operation.

Now we might consider the KT88 in SEUL mode with Ea +440V and Ia + Ig2 69mA.
The Ia will be 64mA, and RLa for max Po = 0.9 x 440 / 0.064 = 6k2.
Max Po at anode = 12.7Watts. The OPT secondary loads needed to give RLa = 6k2
are 5r0, 10r0, 20r0. A speaker of "8r" could be used on the 8r outlet, but the primary
loads becomes 5k0, and Po max = 10Watts. But an 8r speaker may dip to 5r0 so it
is best used on the 4r outlet and the tube will produce highest power where the Z dip
occurs which is OK. But the use of a "4r" speaker with a dip to 3r will reduce RLa to
3k7, and Po max = 7.5Watts with high THD.
However, all this isn't as bad as using the KT88 in triode with Ea = 375V.

SE amps can only give maximum Po at ONE load value. Suppose the load for
12.7W max is 3r5, when a "4r" speaker is used on the 4r outlet. Then if speaker Z
dips to 2r0, Po max about 6.7Watts, and if Z rises to 8r0, Po max = 6.6Watts.
So although a "12Watt" amp can make 12Watts, the inevitable variation in speaker
Z across its frequency band and the low sensitivity of modern speakers will reduce
real capability of the amp to 6 Watts. Hence high sensitive speakers should be used.
However, with say 3 or 4 x KT88, the usable Po range becomes say 24Watts,
and then the load matching becomes far less critical and the amp will perform
well with any speaker.

Back to Education and DIY directory

Back to Index page