Push Pull Output
stage configurations. April 2014.
This page is about unusual PP amps with seriesed output tubes in
triode class A,
and for use of normal UL PP OPTs to obtain local Cathode Feedback
greater linearity in the output stage.
Fig 1. Waveforms of
signal currents and 2H conventional PP Class A1 triode amp with
OPT with B+ at CT.
Fig 2. Waveforms of
signal currents and 2H in series connected class A1 triode amp
with capacitor coupled OPT.
Fig 3. Complete
series triode amplifier with bootstrapped concertina phase
Fig 4. Complete
series triode amplifier with IST used for phase inverter / driver.
Fig 5. Complete
unconventional UL amplifier which uses 40% screen taps to provide
local CFB from OPT.
Fig 6. Load line for
class A with 6550/KT88
Relevant notes and explanations about all.
There are many ways to build tube amplifiers; some are more
difficult than others.
The aim of the serious DIY hi-fi tube enthusiast or manufacturer
is to obtain the best sound quality
1. Safety of operation,
2. Low distortion and
noise. Should be no more than 0.2% at 1dB below clipping with
3. Low output resistance
so Damping Factor = Speaker Load / Amp Rout is greater than 8.
4. Wide bandwidth from
20Hz to 20kHz at least, at full clipping levels.
5. Adequate output power
for the intended speakers.
6. Good load matching to
best suit the speaker impedances.
7. Good tube longevity by
using sensible levels of bias so Pda at idle is not more than 75%
rated maximum Pda, and adequate tube cooling without fans.
8. Use minimum number of
parts without compromising technical operation, and have not too
9. Ensure the amp has
adequate initial class A1 power to cover 90% of listener needs,
before the amp
begins to work in class AB1.
10. Each and every stage
in the amp must work without any grid current with rated load
right up to clipping
onset and give full power bandwidth from 20Hz to 20kHz at 0.3% THD
at 1dB below clipping. No driver
stage should produce more than 0.5% THD at 1.5 x the signal
voltage needed for output tube grids at 1dB
11. Build the amp so it is
easy and safe to service, easy to bias without any special tools
other than a screw
driver, or needing to remove any metal covers or needing to read
voltages on meters.
12. Ensure the amp weight
does not exceed 25Kg, which means use of monoblocs not exceeding
rather than having two channels + PSU on one chassis weighing
Sometimes having both audio amp channels on one chassis is OK with
their PSU on another.
Everyone who studies tube amps will most likely be aware of the
most common PP amps where
at least 2 input / driver triodes/pentodes are used to drive 2
output tubes with the anodes connected to
each end of a primary OPT winding with a CT taken to the B+
supply. Classic examples are the Williamson,
Mullard 520, Quad-II, Leak2020. Manley Labs, ARC and many other
recent amps followed these old brands.
5050 is an example of a "normal" PP amp.
To keep the discussion simple until readers are ready to consider
the more complex amps, pure class A1
operation must be fully considered first, in a conventional normal
PP amp with OPT with CT.
With a class A1 amp, each of at least TWO output tubes have the
same signal Va-k and Ia waveforms
except that signal voltages and currents are oppositely phased and
2H current waves have the same phase.
Each tube of the pair works as a single ended ( SE ) tube.
The distortion currents in each output triode is mainly 2H, with
much lower levels of higher H.
Where the 2H distortion currents have the same phase in each
triode and have equal amplitude, there can
be no 2H current flow across the primary winding so the 2H
currents in each triode do not generate
2H distortion voltage in the load. But no two triodes are
perfectly matched, and there is always some net
amount of 2H but at 1dB below clipping it may be only 0.05% while
the 3H may be 1.0%. The use of
PP class A1 triodes can give extraordinarily low levels of THD,
and often 6 times lower than if the two
triodes were used in parallel.
Fig 1 shows a typical class A1 PP output stage with PP OPT with B+
fed to a primary winding CT.
The pure class A operation results in a slight increase of B+
current to each tube and in this case
the idle current level of 70mA is seen, and because of the 2H
production character of the triode
it operates as a rectifier and at the high signal level shown each
tube would require 75mAdc,
and in fact the B+ current will increase from 140mA to about
150mA. This slight rise in PSU
anode current is very low compared to what happens in class AB1
with a low value RLa-a
load and at clipping the B+ Iadc may double or triple depending on
the idle currents.
Series Push Pull tubes, aka Totem Pole Connection.
Fig 2 shows that in theory, there isn't anything anything
preventing output tubes being connected in series in
what is called a "totem pole" output tube configuration.
To avoid confusion, and perhaps make it clear why the output with
series tubes is so good, Fig 2 gives
operation of series output triodes without showing the
complexity of a whole amp schematic
including biasing details of the KT88.
To obtain the above information an output stage was set up with
KT88 in triode, each with cathode R&C
biasing networks. This auto biasing for class A was found to give
good Idc control with equal Ea for each
KT88, and of course there MUST always be equal Idc in each KT88
because the pair are in series.
For triode operation the Ia mentioned always includes Ig2, because
g2 is tied to anode via 220r, not shown above.
A suitable IST with 2 : 1 + 1 ratio was used to drive the pair of
KT88. This IST had 2 isolated secondaries so
one was used for each KT88, but with the two windings producing
opposite phases. The use of the IST gave
very low DC resistance for biasing the grid at voltage at bottom
The source to drive the IST was from a voltage amp capable of
200Vrms 1Hz to 1MHz at Rout = 1k2.
To view the current wave forms in each KT88, two 10r were placed
as shown and a pair of spare AF transformers
normally used for driving a pair of PNP output transistors - as
they did in 1955, and in a similar circuit arrangement
as I have here.
The transformers allow us to view the current waves on a CRO of
each tube without the presence of the output
voltage appearing at the transformer secondaries. I used a dual
trace Hitachi CRO from about 1983.
The current waves for each tube show us immediately what is going
on with 2H in each tube and why there is
virtually no 2H in the Output Vo.
The test circuit could be driven well into class AB2 and give much
more AB2 Po than the 16 Watts one can get
with pure Class A1.
The Ek across each of the two R&C cathode bias networks begins
to increase with any increase in Vo
above zero signal level. This indicates that Idc is increasing in
both tubes due to slight rectifying behaviour
of each tube, due to its 2H generation non linearity. When the
input signal level is pushed above class A the
current waves become most distorted, with perhaps 20% THD with
multiple H present. The 3H at output
will then increase, but Vo will have MUCH less distortion voltage
present than the % THD in the current
waves. This is true for all class AB operation.
In Fig 1, the KT88 triodes need a +430Vdc B+ supply with 140mAdc
total Idc at idle.
In Fig 2 for the series connection, the B+ for each tube is in
series hence the B+ rail of +850Vdc with 70mAdc
in each KT88. The HT winding on a normal amp with center tapped
OPT may have a 325Vac winding
which has a diode bridge rectifier to generate about +430Vdc at
140mA for the 2 KT88 triodes.
But for the series connected tubes the SAME 325Vac HT winding is
used with only 2 diodes in a voltage
doubler with two 470uF / 450V in series to generate about +860Vdc
at 70mAdc, before being filtered
by CLC filter with two more 470uF caps.
The PSU supplies the same anode DC power for the two
There is only one Output
Voltage, and it is the same for both tubes. But the signal
currents in each KT88 have
opposite phase, and while one triode is increasing its current
flow the other is reducing its current flow.
When the V1 top KT88 is increasing its Ia, the V2 is reducing its
Ia, and the net change in current is applied to
the output load.
One may be forgiven for
thinking the top tube works like a cathode follower. There is no
NFB follower action in Fig 2,
and none in the following Fig 3 complete amp schematic. There
could possibly be local series voltage NFB in the
Fig 3 output stage as invented by Technics for use in OTL amps,
but little need for it with output triodes with a high RLa
relative to the low triode Ra. The Technics circuit is excellent
for mosfets and BJTs, but with tubes the drive voltage
supplied to Technics must exceed the Output Vo, which creates
complexity than is worth the trouble.
Fig 3 is a complete amplifier with series output class A1 triodes
with cathode biasing.
B+ = +850Vdc, and Iadc = 70mAdc. Both tubes work with the same
conditions as they would in Fig 1 or similar
"conventional" PP class A amp with an OPT with CT.
The power generated by the pair of series tubes is taken from
between 0V and the anode-to-cathode junction
of the two seriesed output tubes. The anode-to-cathode junction is
at a +425Vdc potential while one end of the OPT
primary is directly connected to 0V so there must be a coupling
cap to block Idc flow through OPT primary which has
low winding resistance. I have chosen 470uF. This capacitor should
be bypassed with a 1uF polypropylene
rated for +630V.
The V4 & V5 output triodes are driven by V3 EL34 which acts as a "Bootstrapped Concertina
Phase Inverter Driver."
It is so called because its like all concertina phase inverters
with one triode with equal anode and cathode
resistance loadings, in this case, both 2k2, 10W, rated to take
the 24mAdc in EL34 triode.
Now the top of anode 2k2 is connected to Vo via a "bootstrap" link formed by 100uF. To
get DC flow to V3
and its pair of 2k2, there is a 13k9 resistance to +850V rail.
The output stage Vo is applied to the 13k9 and the 2k2 and there
is a negligible amount of wasted AC power
across the 13k9 resistance.
35Vrms must be generated to drive each KT88 between each g1 and k.
V3 anode resistor of 2k2 has
the same current flow as in cathode 2k2, so the SAME 35Vrms
exists across each 2k2, but they have opposite phase. The
bootstrapped anode 2k2 means the V3 anode
supplies an input grid voltage at V4 which is 35Vrms greater than
the Output voltage of V4+V5.
There is 35Vrms generated by
V3 at its cathode 2k2 from g1 to 0V, this drives V5 KT88 grid g1.
V3 has to produce a Va-k = 210V +
35V = 245Vac, a large amount of signal voltage for any driver
and to do that with low THD. EL34 in triode has been chosen.
Possibly an EL84 will do, or ECC99,
but I would NOT use a paralleled 6SN7, 6CG7 etc. The EL34 is a man
for a man's job, OK.
The V3 Vg-k = Va-k / open loop gain = 245 / 9 = 27Vac approx. With
35Vac at cathode, the V3 Vg-0V
input = 27 + 35 = 62Vac approx. This voltage produces the 245Vac
total output so gain with the local
current FB = 245 / 62 = 3.95. With low gain due to local current
FB, the V3 will be linear enough.
The signal current in 2k2 = 35 / 2.2 = 15.9mA rms, or 22.4mA peak.
Hence the need for the idle
current in EL34 to be at least 24mA, and it could in fact be
The bootstrapped anode 2k2 means that the anode Vac = 210Vac, and
that the anode load is made
virtually higher than it really is :- 210V / 15.9mA = 13.2k. The
cathode load = 2k2, so total load on V3
= 15k4, which is over 10 x Ra for EL34, so its open loop THD would
be less than 1.5% at voltages
shown, and the effect of the local current FB with 2k2 Rk means
THD of the stage will be about 0.5%
maximum. V3 bandwidth will be excellent, exceeding 150kHz.
Similar driver schematics have been used for OTL amps using
multiple seriesed 6AS7 or 6C33c.
But readers should realize I have many reasons why I don't like
OTL amps, unless of course you
replace the series output tubes with power mosfets. THEN the same
EL34 driver can be better
used and it will work fabulously well with low operating signal
voltages and negligible THD.
Are there any advantages for
using series output tubes?
The traditional PP amp with CT OPT is easier to build and will
give excellent operation with
class A1 triodes, even class AB2.
The series triodes requires biased heater supply to the top
triode, and a higher B+ which
some will complain that it is too dangerous. But +450Vdc can kill
just as well as +860Vdc.
The series tubes needs special understanding, plus a few more
parts, so mainstream
makers have mostly avoided the design.
A notable exception was Philips who used a pair of EL86 in series
class AB1 pentode, with 400Vdc
supply to drive a capacitor coupled 800 ohm load for 10Watts.
Special speakers were made with 800r impedance using extremely
fine wire in the voice coil.
Unfortunately, although Philips was extremely determined to get
rid of the OPT, they still needed
an OPT to get screen drive to top EL86. The 800r speakers could
not be used with any other amp,
which everyone is tempted to try to do. Th voice coils were very
fragile, and busted easily.
Phillips abandoned the
series tube idea at about
the time in 1962 when the awful solid state
devices began to become reliable enough to make 10Watts.
Possibly you may want to make use of the series connection where a
suitable OPT is available.
For series KT88 in triode and for Fig 2, each KT88 must see the
same RLa of about 3k3, and the
two tubes work in parallel on the output load
which must be about 1k7.
Now you could use a Hammond 1627SEA with an air gapped core to
suit a normal paralleled pair
of KT88 in SE triode, and with 140mAdc flow through the primary.
The 1627SEA has nominal OPT ZR = 2k5 : 4, 8 & 16. If the
wanted load for maximum pure
class A = 1k7, then loads at secondaries must be 2r8, 5r6, 11r2.
These loads are the minimum Z
for many speakers rated for nominal Z = 4r, 8r, 16r. The design
should be based on the speaker
minimum Z and then the amp always works in class A, even with the
lower than nominal ZL.
The 1627SEA can be used if tubes are in series, and output taken
from join of V5 anode
and V4 cathode, with Idc blocking cap of 470uF. We do not want Idc
flow in the primary.
In a normal SE OPT with Idc flow, the total maximum Bdc + Bac at
clipping may be 1.3Tesla.
If the total Bac + Bdc go higher the iron caused HD goes too high.
The Hammond 1627SEA is rated for 30W into 2k5, so rated maximum
Vac across primary can be 273Vrms.
Hammond OPTs seem to saturate at rated power at 30Hz, ie, Max Bac
is reached at 30Hz with Va = 273Vrms.
But without any Idc, there is no dc core magnetization, ie, no
Bdc, so Bac could double without saturation.
This means that the core will saturate at 15Hz with 273Vrms and no
But we only want to have Vo = 175Vrms and then Fsat < 10Hz.
The primary inductance of 1627SEA is rated as 20H, ( but without
any Idc it would be marginally higher.)
If the primary load = 1k7, then XLp = 1k7 at < 13.5Hz, ie, the
RL in parallel with Lp = 1k2 at 14Hz.
This means the maximum rated power at 1kHz can be maintained right
down to 20Hz without the Lp
wasting a lot of current flow in tubes, and causing high
distortion from overload.
So we get the 1627SEA to work
better at LF than Hammond ever intended.
Hammond do not have any other suitable PP OPTs with CT and which
are rated for about 30W
and with primary RLa-a loads less than 2k0.
One probably very suitable
OPT would be the Lundahl OPT with many available winding
In the past, series output tubes were not ever used because the
expense could never be justified,
although the OPT can be more easily wound with less
turns and thicker wire for a primary, and there
problem with balance of Idc in each 1/2 of a normal PP
So the conclusion I make
is that the series tubes does allow some OPTs to be used which
otherwise could not be used. In theory, the sound quality should
be excellent with series triodes.
practice, the Fig 3 amp begs us not to build it.
There are THREE input driver stages, and lots of R+C and C+L couplings which threaten LF stability
when NFB is used because of increased open loop phase shift at LF.
I suspect 12dB GNFB would be about as much NFB as possible, but
that is enough for any triode class A amp.
What could be done to make the series triode amp a little easier
to build, with maybe better technical and sonic results?
Fig 4 has an IST which may be a Lundahl such as the LL1635 IST.
The Lundahl specifications are MUCH better than IST from Hammond,
such as their 126B which has unwanted
bifilar windings which will give a high value of unwanted
capacitance between V3 anode and the output from the KT88.
The Lundahl prices are probably horrendous, but then, do you want
to make dreadful junk or make something
better than everyone else? Save your pennies for the
Most ppl would use an IST with primary between B+ and anode with
an Idc flow in the primary.
But the Lundahl specs say that with Iadc up to 35mA, Vo max =
90Vpk, or 63Vrms per winding.
Better operation is obtained with no Idc, and maximum Vac per
winding becomes 155Vrms, because there
is no dc magnetization of the core. Each of the two windings may
both be used to bias the KT88 grids with
a very low resistance of the winding Rw, eliminating the need for
R&C coupling, excepting what is needed
from EL34 anode to IST primary, which drives V5, with one end of
The secondary has one end to Vout, and biases V4. Each of the IST
windings will have oppositely phased
signals with equal amplitude. The whole IST secondary is
referenced to Vout, so V4 gets exactly the same
V drive as V5. Hence there is no need for a concertina phase
inverter stage and no need to produce a high
drive voltage, so one whole amp stage can be left out compared to
KT88 triodes in class A1 as I have them in Fig 3 will each have Ra
= 1k1. The two triodes are in parallel
so Rout at Vo without GNFB = 550r, and with RL = 1k8, the Damping
Factor without GNFB is a healthy 3.2.
Winding resistance may lessen this slightly. The GNFB will
increase the DF to at least 12 because the 550r
is reduced to at least 140r.
While much adulation is now given to KT120, KT150, it is still
better to use 4 x KT88 / 6550 than
2 x KT120 or KT150. The simple reason is that 2 parallel KT88 on
each side of a PP circuit have twice
the gm, half the Ra, and better manage a lower load than 1 x KT150
on each side of PP circuit. The
KT88 can run with less Iadc and total Pda so they will last
longer. The rated maximum Pda for KT150
is 70Watts, but only a fool would idle one at that level, and
maximum idle Pda is probably better
at 50Watts max, so two can have Pda = 100W and maximum triode
efficiency = about 33% so hence
max class A triode Po = 33Watts. The KT150 in pure beam tetrode
mode is far from linear, like
most other power tetrodes and power pentodes. But in a PP amp with
CFB, then they will be excellent
just like the others because of the way the local NFB is used.
Use of KT150 will give the most interesting results because they
look so spectacular, so much is expected
of them. But let us consider if they were used instead of just 2 x
KT88 with idle Ea = +380V.
KT150 could be safely biased
for Pda = 45Watts so Iadc = 45W / 380V = 118mAdc.
For max Class A1, RLa = ( 380V / 0.118A ) - 2,000 = 1,220 ohms.
But RLa should be at least twice
Ra for all class A triodes, and Ra is about 1k0. So, Ea would need
to be higher, so let us choose
Ea = +450Vdc. This means changes to PSU to use more rail caps
because the B+ with cathode bias
for series tubes will be maybe +1,020Vdc. With Ea at +450Vdc, and
Pda = 45W, then Iadc = 100mAdc
so RLa = ( 450V / 0.1A ) - 2,000 =2,500 ohms. The Series pair needs RL = 1.25k.
Therefore a Hammond 1640P could be used, nominal ZR = 1.25k : 4,
8, 16, and quite OK.
Maximum Po = 25Watts for pair. The outcome is remarkably similar
to using ONE SE 845 with Ea = 1,050Vdc
and OPT 12k0 : 4, 8, 16. Suitable
OPT for 845 are not easy to find.
PP amp with normal PP UL OPT, but using UL taps for
For the adventurous amplifier builder, I now submit a design which
allows a normal UL OPT with
typical 40% UL taps for screens to be used to make an amp with
local cathode feedback but
without a special OPT with has a dedicated tertiary winding for
Fig 5 shows what you can do with many existing UL amps if there is
enough chassis room available,
or if you wish to make a new amp, and you know what you are doing,
and you can wind very
The basic idea in Fig 5 has never ever been used by any brand name
manufacturers because of their universal
dumbing down of recommendations in text books such as Radiotron
Designer's Handbook 4th Ed, 1955.
But the educated DIYer without the profit motive to corrupt his
efforts can make fabulous amps by applying
the text book knowledge, and spending time to build it right,
after saving enough to purchase some truly splendid
components which are available if you search for them.
Firstly, you need to consider a conventional UL output stage such
as the 50W amp channel at my page at
This amp requires a well made UL OPT which almost nobody can find
at a reasonable cost.
However, there are some well made OPTs out there and something
from Lundahl will suit those who don't mind
saving up for a really fine OPT, better than Hammond, but costing
a lot more.
But allow us to consider first what Hammond offers. The Hammond
1650P is rated for 60W Class AB1,
with primary load of 6k6, sec loads 4r, 8r, 16r.
60W into 6k6 means 629Va-a, so there is 314Va to k. This is a peak
voltage swing of 444Vpk, so the Ea for KT88
would have to be about 500Vdc. But in my example above we have Ea
= 425V, and for 6k6 RLa-a the Ea swing
is 385pk, so signal Va-a = 544Va-a which generates 44.9Watts. But
the class A load for each tube for first few
watts is 3k3, and with Iadc in each KT88 = 60mAdc, pure class A Po
max = 11.8Watts, which covers 95% of
what any civilized person listens to. KT88 Pda at idle = 25.5Watts
which is very comfortable.
You could get maybe up to 60Watts class A with first 5Watts in
pure class A if the RLa-a was reduced to 3k3,
but then the outcome is not hi-fi. To get the high power at low
fidelity one could use a 4r0 speaker plugged to the
outlet labelled "8 ohms".
Now let us examine what happens with UL taps used for CFB as shown
in Fig 5, and with Hammond 1650P.
Fig 5 shows the voltages for class A loading at maximum Vo at
onset of clipping.
The UL taps are at 40% of primary. So the UL fraction = 0.4, or
But with the UL taps used for cross coupled CFB, the UL
fraction changes to 80V / 280V = 28.6%.
This places the KT88 closer to tetrode operation, and its gain =
Va-k / Vg-k = 14.
But 28.6% of the Va-k is feedback in series with the grid voltage,
and the Vg-0V must be 100V.
So then the gain is reduced from 14 to 280 / 100 = 2.8, so gain
with CFB is 1/5 of the KT88 with plain old UL
and its THD will be at least 1/4 of the the plain old UL case, and
Ra will be lower than triode, < 1,000r.
The class A load required for pure class A in each KT88 is 6k6,and
so with the voltages shown in Fig 5,
each KT88 has Va = 282Vrms, and this produces 12Watts of
pure class A, so 24Watts class A for both
KT88. 400Vrms is across whole winding. OPT ZR for 6k6 : 4ro
= 1,650:1 giving TR = 40.6:1, so there
is 9.8Vrms at the 4r0 outlet, and if a 4r0 load is used, then Po =
24Watts, and power is all pure class A.
This means that loads down to 2r0 will be tolerated well because
the cross coupling of primary has the effect of
changing the impedance ratio of OPT, and if 2r0 were used at 4r0
outlet, tubes will experience class AB loading
of 6k6, but they will handle it all the better because of the CFB.
One could take all this further by having cross coupled cathodes
taken beyond UL taps and to the anode
ends, and then the pure class A voltages would become Va-k =
280Vrms, with Va-0V = 140Vrms, Vk-0V = 140Vrms,
and Vg-0V = 160Vrms. But the choke at cathodes needs to be much
larger, and driver voltage has more THD,
so little is to be gained. But McIntosh do just what I have
mentioned but instead of a CT cathode choke, they
have TWO primary windings each with CT, and bifilar wound and the
outcome even in class B is acceptable.
But buying a McIntosh type of OPT with bifilar windings that will
have special wire enamel for 450Vdc
is probably not possible, and IMHO, certainly not needed.
The 28% of CFB is quite enough CFB to do all anyone wants with
CFB. Quad-II used 10%, and I have used
12.5% in my 8585 which gave about 0.7% THD at 50Watts with 4 x
6550 with NO GNFB.
Going from 12.5% to 50% like McIntosh does not improve matters
The other choice of 60Watt Hammond UL OPT is the 1650N which is
4k3 : 4, 8, 16.
The secondary loads for pure class A with conventional UL will be
determined by having RLa-a = 13k2,
so sec loads would need to be 12r2, 24r4 and 48r8.
The cross coupled cathodes to UL taps for CFB will double the
RLa-a so effective ZR = 8k6 : 4, 8, 16.
If the tubes see RLa-a = 13k2 for pure class A the sec loads could
be 6r1, 12r2, 24r4 for the 3 nominal
connections, and it shows that pure class A for 4r0 is not
Using KT90, KT120, KT150, the Pda of each can be increased from
25.5W for KT88 to 35W, 38W, and 45W
respectively. The increase in Iadc allows a lower RLa for each
tube in pure class A. Therefore using Hammond
1650N with 4k3 :4,8,16 would be OK for KT120 or KT150.
But do not expect a huge change to what you hear if you settled on
KT88 or 6550 etc.
If you do the load line analysis for any pair of octal tubes with
Ea at +425Vdc, Ia at 60mAdc, the load for each tube
may be plotted.....
In Fig 6 we see the Ra curves for a single 6550 or KT88 with UL
tap at 40% of primary turns on an SE OPT.
If this tube is one of a pair in a PP circuit then each would
operate with similar curves if the RLa-a anode load = 13k2.
Notice that the curves have a slope which are intermediate between
fairly flat Ra curves for pure tetrode
and the steeper slopes of triode connection. The pure tetrode Ra
will be about 30k at the idle Ea and Ia.
The triode Ra will be about 1k1 at the idle Ea and Ia. The Ra for
the above UL mode at idle is 2,750 ohms.
A straight line A - Q - B is shown for load value of 6k6 extending
from point A, at +20V and 20mA,
The maximum Ea swings are - 405Vpk, and + 400Vpk. At such
voltages, there is mainly 2H and 3H.
The possible load voltage = 805Vp-p = 285Vrms, yielding over
11Watts for 6k6.
At about 3/4 of the +/- swings for Ea, the THD is mainly 2H like
triode and < 5%. In PP, a pair of 6550 / KT88
in class A with RLa-a = 13k2, the THD is mainly 3H, and < 2% at
clipping onset at 23Watts.
Because of the crowding together of Ra curves at both extremes of
+/-Ea swings, the 3H increases at high levels,
and other odd numbered H suddenly appear as the signal clips.
To obtain equal Ea swings in both directions and with linearity
requires some external loop of negative feedback to
create an error correction signal at g1. The use of two output
tubes in push-pull create loading conditions for each
tube which differ slightly from what is shown above for an SE
In fact, the load line for each tube in PP becomes slightly
curved! Its OK, the PP outcome is more linear than the SE
example, and with with 20dB of GNFB the UL PP amp may have only
0.1% THD at 1dB below clipping while in pure
class A, which is what Williamson achieved with his famous 12W amp
of 1947 with two KT66 in triode mode.
To calculate the class A load ohms for any SE tube in Beam
tetrode, pentode, or UL or CFB mode, the formula is
RLa = 0.9 x Ea / Iadc, where Ea and Idc are the idle conditions.
In this case RLa can be calculated 0.9 x 425 / 0.06
= 6,375 ohms, which is close enough to the load line graph.
Using the graphical load of 6k6, the maximum class A Po for one
tube = 0.5 x RLa x Iadc squared = 0.5 x 6,600 x 0.06 x 0.06
= 11.88Watts, and because we have two tubes, expect 23.76Watts.
But 5% OPT winding losses may reduce this to 22.5Watts.
The Pda of two tubes = 425 x 0.06 x 2 = 51Watts, and anode
efficiency = 44% including the winding losses.
Another way to calculate class A Po = 0.50 x Iadc squared x RLa.
In this case we would get
Po = 0.5 x 0.06 x 0.06 x 6,600 = 11.88Watts.
Using formulas for class A Po isn't always best idea unless you
know what you are doing.
For RLa more ohms than
that for maximum pure class A, the peak to peak Ea swing is
twice the maximum possible negative going Ea swing.
For RLa less ohms than
that than for maximum pure class A, the peak to peak Ea swing is
twice the maximum possible positive going Ea swing.
The formula Po = 0.5 x Iadc squared x RLa only works with the load for
maximum class A Po.
Class AB operation involves more complicated formulas which may
confuse everyone even more than they are already,
so I suggest load line analysis be studied further in my pages
about load matching.
See the Education
and DIY directory.
All of what I am saying about loading applies to the circuit of
Fig 5 and where CFB is used.
The use of KT150 Pda rated for max = 70Watts, Pda at idle could
safely be 45Watts, so with Ea at +425V,
Ia can be 106mAdc.
The class A RLa = 0.9 x 425 / 0.106 = 3k6. Po from one SE UL KT150
will be = 20Watts, and winding losses
of 5% would reduce this to say 19Watts. Two KT150 would be capable
of 36Watts of pure class A,
and would need an normal UL OPT with primary load of 7k2 for
normal PP, with no CFB.
The use of a PP UL OPT with 40% UL taps to give CFB as I have it
in Fig 5 will always mean the RLa-a seen
by the two output tubes will double for class A. So where one uses
2 x KT150, the OPT can have nominal
primary load = 3k6 : 3, 8, 16 and when the CFB connection is used
the load on OPT can be 4, 8, 16 for
pure class A, or 2, 4, 16 for the higher class AB1 level
The 3 main PP OPT worthy of consideration are the 8lb 1650P, 6k6
primary RLa-a, 1650N, 4k3 primary RLa-a,
12lb 1650R 5k0 primary RLa-a.
Fig 5 circuit needs some further explanations.
L1 choke is a challenge to make, but it is a very easy task
compared winding a PT or OPT.
The core size is a 20mm stack of 25mm tongue E&I lams. These
may be recycled from any old fused PT. which you
can pull apart after cutting off the existing windings with a
hacksaw and pliers. The laminations can be levered off
a varnished stack using a sharp knife or paint scraper blade.
The best bobbin is a moulded preformed type with vertical division
in center as used for small 5VA mains transformers.
0.25Cu dia enameled grade 2 magnetic winding wire used for winding
motor armatures is needed.
250 grams would be enough. The wire is wound on one side of bobbin
with slow traversing speed, with random turns
which don't need to be neat layers. The slow traverse across the
bobbin prevents wires crossing over each other at
angles > 5 degrees so that wires will not short over time.
Spray can varnish is sprayed on to soak the turns
as the height increases.
The turns must be counted accurately.
You should get about 1,600 turns onto 1/2 a bobbin, with winding
kept fairly level and allowing Es to be inserted
without touching wires. Then 1,600 turns are wound on the other
side of bobbin, The wire end of first winding is
connected to wire end start of second winding to make the CT.
sleeving should cover the exit wires to windings.
The varnish will harden over time. The E&I lams can be
inserted with maximum interleaving in each direction,
and clamp yokes bolted on, and a terminal board for winding ends
and the CT. All must be made rugged so wire
ends cannot be pulled out of winding easily. angle brackets may
need to be made for a nut&bolt to secure it at
Usually there is enough chassis space for such a small sized
choke. It needs to be away from a PT or OPT to
prevent stray magnetic coupling, and perhaps oriented for least
The LI inductance should act as a high impedance element for
sinking the cathode currents to 0V without causing
phase shift or loading at audio F.
Each cathode is cross coupled to the UL taps on OPT using 220uF
caps. The idea is to stop Idc flow from B+
supply through the choke, but allow unimpeded signal current flow
between OPT primary winding taps and cathodes.
The resulting Vac across the CT choke will be up to about 160Vrms
cathode to cathode. The choke should end up
having L = 100H at least, and its impedance end to end = 6k3 at
10Hz, and 630k at 1kHz, so the loading effect
of the choke is completely negligible at audio frequencies. The
choke allows the cathode current to flow between
cathodes to 0V, and as it normally does where cathodes are
directly taken to 0V, or there is a bypassed Rk.
But the signal currents in each tube flow round a circuit from B+
at CT through 1/2 of OPT primary turns, then to
anode and through tube, and then to the 40% UL tap on the OTHER
1/2 of primary turns and then to B+ CT,
completing the circuit.
So each tube operating in class A uses 70% of the total primary
turns on OPT. Between each UL tap there are
80% of the total turns which are shared by each tube.
The result is that MORE of the primary turns are used by each tube
to give better magnetic coupling to the secondary
turns and the secondary load. Because each tube has 140% more
turns for a given Va-k signal voltage, the Fast
will be lower, or, the transformer can support a higher Vac per
turn without raising the Fsat.
There is the overall effect of changing the impedance match of the
OPT without physically changing the turn ratios.
So what is the big deal? Are not ordinary UL amps just great
without the choke and extra caps?
Well sure they are, but there's so often better lurking around the
corner of conventional thinking.
Now why are the screens taken to a fixed B+ ?
The signal Vg2-k is 28% of the total Va-k, so there is substantial
screen NFB just like in a normal UL amp with
screens taken to UL taps on primary. The KT88 screens could indeed
be taken to their normal UL taps but it will
not do very much because the majority of tube linearization is
performed by the CFB in g1 loop.
The fixed Eg2 rail can be well filtered, and the Eg2 can be lower
than anode B+ because the circuit will not be
working to give extremely high Po with very low RLa-a, where a
high Eg2 is necessary. The lower Eg2 allows
a lower voltage of fixed bias for g1.
The output stage needs 200Vrms grid to grid at KT88.
Instead of using two halves of a 6SN7 or 6CG7 to drive the output
tubes with say 35Vrms to each output grid,
two EL84 in triode should be used for the balanced driver amp. The
best way to supply anode current to EL84
is with a CT anode choke and series anode resistors. The pair of
EL84 set up as I have them will make
THD < 0.1% at 200Vrms anode to anode, and less than using just
a single 6SN7 with its two triodes making 70Va-a.
I have used EL84 in triode in numerous amps as drivers, in my 300W
monoblocs, 8585 amps and in a pair
of re-engineered Dynaco Mk6 monoblocs with 4 x KT88 in output.
Trust me, a good driver stage is essential for any output stage
tube. The sound you hear is the sum of the parts used.
Input and driver tubes can affect the sound as much as the output
To minimize the driver amp THD, the input stage must also be
linear, best done with an LTP which produces a
balanced pair of opposite phases with very low THD. The twin
triode for input is ideally 6CG7 / 6SN7,
and with a CCS for cathode current.
Because the amount of NFB within the output stage totals about
10dB, and because the input and driver amp
are so linear, the amount of GNFB can be quite low at 10dB, so
that total amount of NFB = 20dB.
The amp will not be impossible to stabilize unconditionally. There
are two balanced networks needed to gain the
essential unconditional stability, ie, the amp will not oscillate
under any circumstance, no exceptions.
For LF stability, see C8+R14+R17, and C9+R15+R18. I doubt you will
need to use different values to those I show.
For HF stability, there are 3 networks, C6+R13 and C4+R12 and
These values mainly depend on the LCR characteristics of the OPT
as I have used it, and those characteristics
are always never fully known. It is impossible to quantify the
exact nature of the leakage inductances between
connection points on the OP primary. But you can assume that the
OPT behaves at least like an L+C or C+L
second order filter at F above 20kHz, and often there are a number
of undamped series or parallel resonances
above 20kHz up to maybe 500kHz. These become most apparent when
the Ra source driving the OPT is low.
To damp such L+C resonances, Zobel networks are used such as C6+R13 and C18+R37. But additional R+C
may be needed across OPT 1/2 primaries, say 4k7/5W + 1nF between
OPT anode connection and B+,
or from anode connection and UL tap used for cathode connection,
so that the KT88 begin to see a resistance
loading at above 32kHz. By 100kHz, the reactance of 1nF = 1k6, and
with 4k7 in series, the total Z of the R&C
network is mainly resistive and about 5k0. But at 20kHz, the Z =
16k for C in series with 4k7 so loading
effect is negligible.
Stabilization of amplifiers like this one require deep
understanding of basic LCR behaviour. You don't need to
know the extremely complex maths, but you need to realize
practical application of basic concepts is imperative.
You must chose the R&C values by educated guess and trial and
error and that process is the secret business
requiring logic that is explained elsewhere at this site, and in
some old books about tube amps.
Most DIYer have no idea about how to consider the amp as an active
bandpass filter with loops of NFB.
Most don't understand the principles of Nyquist theory so they
make amps which become fine oscillators,
and that is to be avoided like the plague!
Education and DIY directory
Back to Index Page