2 x SE35 cfb on

The SE35 mono blocs were built for a very discerning customer who wanted
high levels of undistorted sound with two large modern floor standing speakers,
the VAF I-66,  rated for about 90 dB/W/M.
He was amazed at the precision and detail these amps offered.

Unlike most 35 watt SE amps which one very large and expensive single triode
such as an 833, or at least 2 paralleled 845 triodes, these use four paralleled
EH6CA7 beam tetrodes or EL34 power pentodes.
The chassis is brass with aluminum top plate, the power supplies and OPT are
within a mild steel box behind the tubes. The top-plate and the bottom cover
plate is well drilled for ventilation to allow rising cool air around the output tubes.
No printed circuit boards are used, so there is no obstruction to easy air flow so
necessary to keep all tube amps cool.
All the transformers all use GOSS E&I lams, with very generously sized cores
to prevent high temperature rise in the power transformer and to get truly
blameless bass performance from the output transformer.
Without any NFB, the OPT response at full power to the rated load is from 20Hz
to over 70kHz. The response with NFB is shown in the graphs below.
The weight is approximately 25Kg per chassis.

The power supply has all solid state rectifiers for the B+ with a simple CLC filter
for the B+ anode voltage. There is shunt regulated fixed screen bias, and RC
filtered DC heater supply to the two input tubes. Individual cathode bias is used
for each output tube so unmatched tubes may be used.

The driver tubes for the 4 output tubes is an EL84/6BQ5 strapped as a triode,
input is 12AU7 with both sections paralleled. The right front tube is a 12AX7
in cathode follower mode to act as a buffer after the input signal to drive a
switchable frequency low pass input input filter to allow variable cut off
frequencies for bi-amping. This buffer stage is not needed where bi-amping
is not required, but it was found to be sonically transparent.

The owner wanted to try using an 8585 for bass and then have the midrange
and treble powered by the SE 35, and filtering out the bass signal voltages
from the midrange/treble amp effectively raises the power ceiling of the SE 35.
Because bass frequencies below 250Hz take up most of the voltage headroom
of any amplifier, the filtering out of bass frequencies drastically reduces
intermodulation distortions. The input cathode follower also allows the use of
a "passive" preamp, which in this case is a high grade resistance attenuator made
by Vishay, all without losses because input resistance of the follower is so high
with very little shunt capacitance.

The schematic shows some unusual techniques to reduce the normally very high
THD measurements with any SE amps to levels more normally seen with good
PP amps, especially within the first 10 Watts.
THD is less than 0.1% at any load between 3 and 12 ohms for up to 10 Watts.
Most other SE amps cannot achieve this. I am using low amounts of  local cathode
feedback in the output stage in what is called the "Acoustical connection" and
low amount of global negative feedback from the speaker secondary to the
input triode cathode. The total amount of local and global NFB is less than 20dB.

A more detailed description follows with schematics.

Sheet 1, Power amp schematic.
Sheet 2, Power supply.
Sheet 3, Protection.
Sheet 4, Harmonic distortion 1.
Sheet 5, Harmonic distortion plus response 2.
Sheet 6, Harmonic distortion comparison to SEUL 13E1

se 35 cfb power amp
This schematic was valid for 2004.
Input is fed into the high input resistance of V1 12AX7 cathode follower
with its output driving a switchable cut off low pass first order RC filter with
F settings for -3db at 7Hz, 50Hz, 140Hz, 510Hz, and 1020Hz.
R6, R7 form a divider to set the sensitivity of the amp to match a bass power amp,
so that equal voltages are sent to bass and treble speakers designed for use with a
normal single amplifier.

The tube line up in the power amp is V2 = 2AU7, V3 = 12BH7.
12BH7 did work well, but 6BQ5/EL84 works/sounds better.......

The power amp schematic was revised this year because of interest by DIYers
around the World. Most did not want to have any input buffer and low pass
filter so the tidier schematic here does not include it.

V2 is a paralleled 12AU7 with a transistor MJE350 which acts as dc supply
with extremely high ac impedance, ie, a CCS, or constant current dc supply.
This maximizes the linearity available from V2; important because any 2H
will *add* to the 2H of the output stage. V3 is EL84 in triode which gave
slightly more gain, but approximately the same THD, but lower Ra.

The EL84 has about 12mA of idle anode current, ie, 3 times the Ia of the
12BH7, and a choke added and R15 and R16 have been adjusted to suit.
The EL84 was found to give better sonic definition and dynamics.

V4 to V7 are 6CA7 but may be EL34 without any changes, or one could use
3 or 4 x 6550/KT88/KT90/KT120, or four 6L6 or 5881 for about the
same power outputs, providing the bias currents are adjusted by changing
the bias RC networks to each cathode circuit of each output tube so that
the total anode and screen power dissipation of the tubes used is no more
than 90 Watts.

The output transformer is a large core of GOSS E&I lams, tongue = 44mm,
stack = 62mm. There is a well adjusted air gap. And there is  lots of winding
The nominal anode load = 1k2, and nominal secondary is for 5 ohms which
is the load for maximum power. So in effect, each of the 4 output tubes sees
a load = 4k8.
Approximately 12.5% of the primary anode to cathode voltage is carried
in the cathode winding and used as local negative feed back to the output tubes.
This reduces OP tube THD from about 8% with no FB to 2% at 35 Watts.
The OPT CFB with 5 ohms loading is 8dB of local NFB, and the global
NFB = 7dB, so total NFB with 5 ohms = 15 dB.

Considerable cancellation of 2H generated by the EL84 driver stage and
output stage occurs because the 2H voltage generated by each stage has
opposite phase for much of the expected load range.
This is not an easy issue to fully understand.

SE pentode or beam tetrode tube output stages may be used
with a range of loads below and above the "centre value RL", ie, in this
case, 5 ohms, or RLa = 4k8 per tube where maximum possible power
is delivered. For loads below centre value, 2H produced has the same relative
phase as a triode. Usually the lowest load usable with an SE amp is 1/2
the centre load, ie, 2.5 ohms in this case. THD may be 8% max at 2.5 ohms
and as RL rises to 5 ohms, 2H falls to zero, and then begins to rise as RL
rises so that it will be perhaps 8% at 15 ohms, or 3 times the centre
value RL.

But the 2H generated for high RL values has opposite phase to loads
the centre value RL where 2H is zero.

The effect of the local OPT CFB much reduces the 2H produced on either
side of the centre value RL. In fact, there is more reduction on high RL
values because OP tube gain is highest and hence more FB is applied than
when RL is low, and OP tube gain is low. 

While such strange behavior occurs with 2H production relative to load value,
the 3H is maximal when RL is low, and declines gradually as RL increases
between the RL values of 2.5 ohms to 15 ohms. This 3H is unavoidable with
SE pentode and beam tubes, but is not any higher than what occurs in a
class A1 PP amp. The 3H is also much reduced by the CFB connection. 

Now with all SE amps using a triode OP tube and triode driver stage,
the 2H voltage produced by each stage cancels with all load values.
Let me say a typical OP triode produces 5% 2H at near clipping.
The triode driver stage may produce about 2% 2H when the OP tube clips.
The 2% 2H is amplified by the OP tube to produce an oppositely phased
2H signal so that overall 2H will be be 5% - 2% = 3% 2H. This is not a bad
phenomena and occurs in all SET amps. There are some low level IMD
harmonics products called "second order products" produced, but they
remain low enough to be ignored in this discussion.

If CFB is used in a pentode or beam tube OP stage, the same cancellation
occurs where the OP tube 2H is the same phase relative to fundamental test
tone, ie, at loads below the centre value. But the CFB reduces the OP tube
2H to say 2%, and the driver tube produces 2% 2H, and at this point the
overall 2H level becomes close to zero. It never ever reaches zero because
slight relative 2H phase differences between driver and OP stages exist.
But the amount of natural 2H cancellation is highly useful in bettering

Where OP tube loads are above the centre value, the 2H in the OP stage
is not cancelled by the driver stage because the OP 2H has phase which
is opposite to where RL is less than centre value.
But the CFB is more effective to reduce whatever THD exists at high RL,
and although the 2H of driver and OP tubes adds to increase overall 2H,
there is no more than in other typical SE amps.

Graph 1. 2H Cancellation.

This graphs of RL vs THD are difficult to understand because nobody else
bothers to ever draw such graphs.

The thickest lines are a set of 4 curves for CFB operation at power levels between
0.25 Watts and  16 Watts. 

There are thinnest lines of 4 curves for the same OP tubes used in SEUL,
at power levels between 0.25 Watts and 16 Watts.

At the 0.25Watt level the SEUL amp produces 0.15% THD into 2 ohms
which reduces smoothly to 0.04% at 16 ohms for the same power.

At the 0.25Watt level the CFB amp produces 0.1% THD into 2 ohms
which falls rapidly to 0.014% at 4.5 ohms, and then rises to equal the
SEUL level at 7.6 ohms, and rises further to 0.09% at 16 ohms.

Now for both SEUL and CFB amps, the THD for 0.25Watts is less than
0.15% for any load between 2 and 16 ohms. This is a reasonable result for
any good SE amp. Most of the HD will be 2H in general.

But the CFB amp has the advantage that 2H reduces considerably with loads
between 3 and 7 ohms, and when a nominal 4 ohm speaker is used, there will
be an average of about 1/4 of the THD generated by the SEUL amp.

The null of THD seen in CFB curves is all due to 2H being very much reduced
with the remaining HD being mainly 3H. Other odd number H are very low at
the 0.25Watt and 1Watt levels and these do not rise significantly until the
10Watt levels is reached. The CFB amp gives less THD even with a nominal
8 ohm speaker.

Most of the power in the audio band is between 100Hz and 500Hz where
often there is low speaker impedance and high current, and it is in this region
the CFB amp copes better than the SEUL amp.

The other benefit of the use of CFB is that most of the THD reduction occurs
in the OP stage, and any signal fed back to an earlier input stage has less THD
content, so less IMD products are produced by non-linearities in the input
or driver stages.

The curves above were recorded from the 2004 amp version with 12BH7 driver.
The amount of 2H in any chosen driver tube such as 12AU7, 6CG7, 12BH7
or EL84 will vary, and this changes the position of the nulls on the curves.
As the driver tube becomes more linear, the null position occurs at a higher
number of load ohms.

The lower THD and IMD with CFB does not seem to fully explain the
improved fidelity heard from the SE35. The use of the EL84 probably
contributes much to the sonic fidelity because it produces such a small
amount of THD/IMD compared to other triodes with lower Ia. I have
found both EL84 and EL34 in triode to be extremely fine driver tubes.
I've found the use of EL84 as PP driver stages in my PP amps to be
superior to other smaller twin triodes. It was for this reason that I changed
from 12BH7 to EL84.

In 2009, I made a headphone amp using SE EL84 in triode with OPT
and driven with 6CG7. The amp was integrated, and could be used to be
a very fine line level preamp.

When one examines the HD spectra in the 4.5 ohm THD, there is some 2H,
but there mostly 3H. There is also 3H, 4H, 5H, etc in diminishing quantities.
These are mostly hidden from view when viewing the THD on an oscilloscope
at low levels used for most listening below 2 Watts. When testing nearly all 
SE amps the 2H distortion is usually overwhelmingly dominant, often being
more than 15dB above the levels of any other harmonics, so that where 2H
is 0.1%, 3H will be 0.02%, with 4H, 5H being buried in the noise floor of the amp.

Distortion cancelling is frowned upon by some because they say the distortions
of a driver tube are themselves distorted by the output stage thus there is an
increasingly complex mixture of THD and IMD harmonics produced compared
to just trying to have a fairly linear driver and normal output tube with fairly high
My approach was build the driver stage so it is naturally linear, and at least no
less linear than anyone else might achieve with resistance dc load. One might
deliberately set up the driver tube to generate higher 2H to better cancel 2H
of the OP stage, but this is not the best approach. The 2H cancelling I do
achieve is really a by-product that is favorable, but not deliberately created.

2H distortion voltage cancellation occurs in every SET amp ever made. Very
little is ever said about it, but in SET amps where the output tube is a 300B
or 845 and which have no CFB, there is a substantial amount of 2H cancellation.
With a 300B there may be 180Vrms at its anode with THD = 6%, and 60Vrms
produced by the driver tube at 3% THD, so you get a resulting 3% THD overall.

If anyone wanted to use 4 x 6550 or 4 x EL34 all in triode mode, the possible max
PO one could get would be 32 Watts for 6550, and 25 watts for EL34, and there
would be similar levels of THD to the SEUL levels shown in the above graph.
In triode mode, there is no need for the CFB which becomes fairly ineffective
with the low triode gain, about equal to 4dB of FB.


The THD graphs were measured in 2004 for various values of RL from 3ohms
THD is lowest with loads between 4 and 6 ohms, and at 1 watt into 4ohms or
5 ohms THD < 0.032%, This is about 20dB lower than in most other SE amps
where it may be 0.32%.
THD with 3 ohms is higher, but would be *much* higher than what is shown if
there was no 2H cancellation between the driver and output stages.

The difference of THD between SE35CFB and SEUL 25W 13EI amps can be
seen here.....

The SEUL amp with one x 13EI tube is a very good sounding amplifier.
But like all SE amps, there is some THD, and it has a typical amount
shown above and with about 15dB of applied global NFB. The SEUL
graph would be also typical for many pure DH triode SE amps, or
those using triode strapped pentodes or beam tetrodes which might use
say 10dB global NFB.

The SE35CFB also sounds well, perhaps better, but its performance
with 5 ohms is so much better.
At low power levels where the amps are used, the SECFB amps have
6 times less THD, and THD which is about the same as a good PP
amp with the same amount of total NFB.

Frequency Response SE35 at full power, 5 ohms, and -6dB, -15dB.

The top graphs show the response of the amp with 5 ohms at clipping, and
notice that some bandwidth limiting occurs at extreme LF and HF, but -3dB
points are 14Hz and 32kHz. The responses are with full amounts of NFB.
The top graph has the input filter switch set for zero LF attenuation.

More bandwidth becomes available away from near clipping as shown in the
-6dB line, or 9 Watts.

The bottom graph at -15dB low levels show the frequency peaking effects
of purely capacitive loads between 4uF and 0.33uF, along the right side of
the -15dB line, and there is virtually no peaking below 20kHz, showing that
the amp can drive any
ESL load, and that without any resistance, any pure C load will not
cause HF oscillations. The line without peaking under C peaks is for a pure
5 ohms. The peaking with pure C loads is considerably reduced if there is also
a parallel resistance with capacitance load. 
The response of the low pass filter is also shown for different switch positions.


The above PSU schematic gives close to what is needed for the 2004
and 2011 amplifier schematics.
The PSU does not use tube rectifiers because they do NOT improve the
The CLC B+ filter with C1, L1, C3, C4 will give excellent low noise
at the OPT connection = 3.6mVrms, 100Hz even without C2&R1.
This depends on L1 = 1H minimum and it should have Rw < 40 ohms
so that heat in L is less than 3Watts for a core size of T = 25mm, H = 25mm.
The choke may be larger with more L but Rw should be kept < 40 ohms.
The C2 and R1 combine with L1 to make a 100Hz damped parallel resonant
network which increases 100Hz attenuation about +12dB so that with only
1H, Vripple < 1.5mVrms.
L1 with C3, C4 have resonance at 5.2Hz, low enough to not cause too
much emphasis of LF mains noise, caused by so many other users
switching electrical gear on and off in your street or apartment block.

Since the original 2004 PSU schematic was drawn, dc has been applied to
the 12AX7 if used and 12AU7 input tube heaters to ensure a super low noise
floor. There is no need to use DC to the driver stage but anyone is always free
to do this if they like to trim values of series resistors in the schematic.

Active protection and delayed B+ turn on for the SE35.

I repeat text on the schematic :-
DELAY. After turn on there is a slow rise in
voltage at C2 fed by current through R2 so that after 25 seconds current will
flow through the 8.2V zener diode and turn the darlington pair of Q1 and Q2
to quickly close a relay in series with the HT of the power transformer.

PROTECTION. The four cathode dc voltages are reduced by dividers and
each fed through 1N4007 to the base of emitter follower Q3.
If one or more of the Ek cathode bias voltages rises due to increased cathode
current, a sample fraction of Ek is applied through diodes to the base of Q3,
which is an emitter follower buffer. The emitter voltage at Q3 will also rise
and if that exceeds the threshold for forward current flow in 3 series red LED,
d6, d7, d8, then the SCR gate voltage will rise enough to latch it on.
Once turned on, the SCR stays turned on regardless of the Ek. And when turned
on, the SCR anode has low resistance to 0V thus draining voltage charge from C2
so that Q1 and Q2 are turned off and relay coil is denied current. Relay points
open, thus HT winding is opened and B+ reduces to near 0V. The d5 red LED
is turned on to indicate there is no B+ present.
The EK voltage required to trip the SCR is +27Vdc. Vac at cathodes is filtered
away by networks 47k, 15k and 220uF so that AC signals do not trip the SCR.
If the amp is turned off at the mains switch, the +18Vdc rail in the PSU will
rapidly reduce to 0V because it is loaded with the low resistance filaments of
input tubes. So then the SCR quickly becomes "unlatched" and the amp
may be turned on again to "reset" it. If the SCR is soon tripped again after
the 25second delay then something is wrong with an output tube.

If any output tube misbehaves, there will not be any smoke in your lounge room,
or any expensive repairs to output transformers.
The reason for active protection rather than reliance on fuses alone is because if
only one tube were to become saturated with say 300mA of idle current then
the mains fuse will not blow.
So the active protection works well before an errant tube can become saturated.
The  circuit will work when the Ik current in a single tube rises from the normal
62mA to 100mA.

Since 2006, the protection circuit fitted to SE35 has earned its keep. Owners
cannot resist the urge to buy expensive NOS EL34 which they think sound
will better. While this may be the case, some NOS tubes that have spent
perhaps 40 years sitting on a storage shelf may have developed tiny
defects in the glass and after being used for a week, a month, or 6 months,
the defect slowly allows air into the tube and the tube loses bias control and
tries to conduct excessive Ia before it finally self destructs, To avoid the pyro-
technical display and collateral amp damage, the amp is turned off well before
faults become excessive. Protection also guards against grid connections to
tube sockets going open, owners turning up volume with a shorted speaker lead,
or coupling cap failure. 

Back to index page