Input, driver, and output amp stage details.
Fig 1. General basic layout of the amp and PSU chassis
Sheet 1A.  SE triode input stage plus LTP triode driver with CT choke.
Sheet 1B. ALTERNATIVE LTP input stage plus balanced triode driver with CT choke.
Sheet 2. Output stage with dynamic bias stabilization.

Sheet 1A.
Sheet 1A shows V1 SET 6CG7 input with anode CCS using MJE350.
It acts as a differential amp to amplify the difference between a fraction
of the output voltage and the input voltage. The SET voltage gain is close to
20, the of the 6CG7.

The paralleled 6CG7 behaves as a single triode with Ra = 5k, and = 20.
Its RLa is mainly the following bias R11 220k, which is 44 x Ra, so that
THD at 9.6Va < 0.1%, mainly 2H.

V2 and V3 are EL84 in triode and in a differential long tail pair voltage amp.
Each EL84 has gain of about 18 and the two commoned cathodes are taken to
a constant current sink, CCS, using MJE340. Ia in each EL84 = 15mAdc, and
the Ea = +304Vdc, and maximum possible Va swing is up to 150Vrms at
each anode.
This is made possible with use of L1 choke with CT which has L > 200H.
L1 acts like the primary of an OPT with a very high number of ohms inductive
reactance, XL. R21 & R22 6r8 are added so that the EL84 are loaded
by at least 6k8 below 10Hz where XL has reduced below 6k8 at LF, and above
10kHz where choke self capacitance XC has reduced to below 6k8.
So R21 and R22 prevent phase shift at extremes of frequency and thus
permit the easy use of global NFB.
But for the audio band the anode loading of L1 and R21,22 is a very high number
of ohms so that the resulting current change is negligible. The loading of the EL84
is then mainly the grid biasing resistors in the output stage. In my 300W amp,
there are 6 x 120k grid biasing R on each side. So on each EL84, load is 20k.
But the total of 20k bias R are bootstrapped to CFB windings. When output
stage is working with AB load of 3r0, and at clipping, the amount of bootstrapping
is minimum, so that the minimum effective value of bias R becomes 43k.
Where the EL84 anode signal is 87Vrms, the Ia change = 2mArms, and a very
small fraction of anode idle current. Therefore the EL84 are working with minimum
THD despite a large Va swing. The total RLa at 100Hz is about 20 x Ra of each
EL84 and THD at 87Vrms at one EL84 < 0.1%, mainly 3H.
pair of EL84 in triode set up as LTP with CCS for commoned cathodes and
a choke feed.

MJE340 and MJE350 transistors make excellent passive current sinks or
or sources and display effective collector input resistance above many megohms.

Transistors wired like this have no negative effect on the sound. They increase
the fidelity because they reduce THD and make tube operation better than can
be obtained by any other means.

The triodes do all the actual work on the signal amplitude.

The input tube distortion of input and driver stage is reduced about 10dB compared to
using only resistances to deliver dc to the tube anodes. Such resistances are usually
less ohms than the following cap coupled biasing R, so that RLa for 6CG7 would be
less than 6Ra and for EL84 would be 3Ra, which diminishes the gain and the amount
of internal NFB which exists in each and every triode. The higher anode load ohms
increases voltage gain to near and reduces THD while the Ra of the triode is
minimized because with CCS or choke the Ia can be increased and tube placed
within best operation region where gm is fairly high and linear.

Alternative input tubes could be used for V1 input such as 12AT7, 12AU7, 6DJ8.
Also 6BX6 or 6AU6 or other sharp cut off pentodes may be strapped as triodes.
But heaters and biasing must be re-arranged. Apart from 12AU7, most have
excessive gain. The ideal amp has adequate triode gain with low Ra and low THD
and high bandwidth and as I have indicated with 6CG7.

Where anyone might use an input tube with different gain, the global NFB
network R25, R26, C10 may have to be changed to maintain the same
amount of applied global NFB.

I show a Zobel network of R10 & C7 used to load V1 anode to reduce V1 gain
and phase shift above 20kHz. In my amps such a Zobel network was not needed.
But if you are going to build this amp, then you probably will have an OPT with
less HF bandwidth and you will need to carefully trim some values of R&C to give
unconditional LF and HF stability and give full Po bandwidth 20Hz to 65kHz
with rated load resistance and with 10dB GNFB.

The V2 & V3 LTP uses a 500H choke with CT in the anode circuit as well as the
resistor loads of 6k6 to each anode circuit, R21, 22, 23, 24. This technique with
a choke is a unique feature I try to use in all amps where a choke is used to
increase the AC impedance of the element through which Idc is supplied to the
triode anodes. The combination of choke plus resistance has a minimum combined
low impedance equal to the 6k6 plus low choke wire resistance at very low frequencies.
So a useful amount of dc current may be provided to each EL84 anode without
needing a very high B+ supply voltage.

The effect of the 6k6 also acts to allow useful gain at very low F below 10Hz
where there is a very low audio signal content in music, while also avoiding the
phase shift caused by L shunting Ra, so the L&R act as a gain shelving network
and stability with NFB is OK. At F above 10Hz the 500H choke reactance
increases at 6dB/octave and becomes a very high reactance load.
500H has Z = 31.5k at 10Hz, 315k at 100Hz, and much more for most of
the audio band until its self capacitance begins to reduce its reactance.
The resistance in series with the choke prevent the choke's capacitive
reactance causing phase shift at the anode because the 6k6 is several times
the Ra of each EL84 in triode.
Therefore the choke plus the series R provide a very high ohm loading at
each EL84 anode and therefore the voltage swing can be maximized without
the distortion which would occur if the choke was not used.
Therefore the resistance load value for each EL84 is mainly the capacitance
coupled bias resistors used for each output tube.

In my amp I have 6 x 120k ohm grid bias resistors on each side of the the PP
output stage.
All the bias supply ends of the 120k resistors are bootstrapped to the cathode
feedback winding through Cc, Sheet 2 below.
Thus the anode load experienced see by each EL84 of the the LTP is approximately
low enough to ensure good ac balance but high enough to ensure THD of the
driver LTP stage is about -10 dB lower than if the dc was brought to the
EL84 anodes via resistors which would have to be approximately 15k, which are
is really too low to get low THD.
The Ra of each EL84 as used = 2.2k approx, so the load of 40k = 18 x Ra which
ensures very low THD.

In other versions of this LTP with EL84 I have tried using R&C cathode biasing networks
for each EL84 and found this gives closer balance of dc current to each EL84 with no 
detriment to signal operation. Non bypassed Rk will cause the effective Ra to be much
higher, lower the stage gain, demand twice the drive voltage from V1, and reduce the
bandwidth thus reducing the sonic dynamics. In my circuit here with cathode feedback
in the output stage, the output tube grid signals are twice the levels of using the plain
ultralinear configured amps.

I have found the idle current balance in EL84 remains constant without any cathode
biasing over many years because the EL84 are set up with a very low amount of Ia
compared to when used in an output stage with Ia = 40mA.
But here they have Ia = 15mA, so the EL84 should have a long life without problems.

At normal levels and because of the global NFB the overall THD is quite negligible
because THD is about proportional to output voltage.

There some obviously acceptable other tubes that could be used for V2 and V3.
EL86 are pentodes which will work as well as EL84. EL86 gain is 1/2 that of EL84,
and Ea should be +250V, not +300V, so R between anode s and choke would be
increased for a given B+. EL86 triode Ra is only 1.4k, so there is some betterment
of open loop bandwidth. Using EL86 would mean the V1 gain would need to be doubled
to produce about 20Vrms to LTP as I show it in Sheet 1A above.

EL86 are not manufactured any longer. 6V6 could also be used and gain is very
similar to EL86. But Ra of the 6V6
triode is twice that of EL84. The best choice is EL84.

For better performance to get higher Va to drive tubes like 845 in PP, a pair of EL34
is hard to beat.

Since 2006, instead of V1 operating as a single ended stage as shown, I have tried
using a pair of triodes in an LTP with cathode CCS. This allows for balanced drive to
the same following LTP stage using EL84. The CCS used in the EL84 cathode circuit
may be replaced with a fixed resistance, say 5k6 to a negative supply rail, say -150Vdc.
The signal input is fed into one grid of the input LTP and global NFB fed into the other
grid of the V1 LTP. I have tried this on several re-engineered amplifiers, ARC VT100, and Dynaco
MkVI, and was rewarded with astounding music and excellent technical measurements.
One 2013 project was reforming a pair of RCA amps seen at Reformed RCA 30W amps

The page further explains the use for cascaded LTP input-driver amps, and the
reformed RCA schematic is fine for where output drive voltage is fairly low for
all OP tubes including KT88 in triode.

The dual LTP idea may be used for higher Vo to suit my 300Watt amp with local CFB
in using tertiary windings at the OPT, and here is what you might use......

Sheet 1B.

Sheet 1B has changes to Sheet 1A but I have kept changes simple as
possible. Experimenters may use either Sheet 1A or Sheet 1B while
making only minor changes to B+ and B- rail supplies shown on Sheet 4.
Sheet 1B does require a higher shunt regulated B+ rail for V1a&b, and
the B- fixed bias supply must be changed to get an additional -24Vdc
rail to provide for V1 current sink using MJE340. Sheet 4 has 2 diodes
used in a full wave rectifier
off 12.6V-0-12.6V Vac heater voltages producing
-17Vdc fixed bias for output tubes. The supply must be altered so a doubler
rectifier to produce an additional -34Vdc rail which is RC filtered down and
shunt regulated with 24V x 5W zener diode, so that 14mAdc is available to
MJE340 and divider resistors.

The 6CG7 V1 input tube is used as an LTP with input ports to both grids.
Input signal is to V1a grid and GNFB is to V1B grid. The differential gain is
16x approx. THD of this differential pair or "LTP" is much less than the Sheet 1A
version with SE V1a&b. Balance with 1% resistors will be excellent and
each of V1a and V1b need only generate 4.8Vrms to feed V3 and V4 grids.

V3 and V4 are a balanced amp, with balanced output produced by a balanced
input, giving least THD. The loading of anodes is the same as in Sheet 1A and
includes the same CT choke L1. V3 and V4 can each produce 150Vrms before

I could have used bootstrapped anode load resistors for V2 & V3 just like McIntosh
have done in their amps. But some positive FB is introduced and I preferred the purer
method for raising the anode load ohms by using the L1 choke.

The Sheet 2 output stage looks complex, but it is mostly repetition of a basic idea.
The R&C part identification seems strange but all resistors and capacitors in similar
functions for each tube are just labelled with the same number for R, and same letter
for C. I am sure any tech will get used to the idea which keeps my schematic

The two balanced outputs from the V2 & V3 LTP driver amp anodes on Sheet 1A
are fed to rails with 6 x 0.47u coupling caps Ca of 0.47 uF, on each side of PP circuit.
R1 2k2 grid stoppers are used on each tube to prevent RF oscillations.
Each output grid is biased with R2 120k, and all 6 on each side of the PP circuit
are taken to a -17.6Vdc fixed bias supply via R5 4k7. Sheet 4 shows the bias
supply. The -17.6 fixed bias is bootstrapped by Cc 470uF to the CFB windings on OPT.
This bootstrapping raises the effective ohm value of 6 parallel 120k R2 from 20k to
at least 43k. Thus the loading effect on V2&V3 of many bias R2 is kept low.

Each output tube has separate R&C cathode biasing with R3 = 500r formed
with 3 x 1k5 each 5W. With B+ at +500V, and screen supply at +387Vdc,
the Ek cathode bias voltage will be about +23Vdc, so idle Ea = +477Vdc, and
Eg2 = +364Vdc. The 500r for R2 is high enough for good bias regulation Ea
is kept usefully high and total idle grid bias -Eg1 = 23V + 17.6V = -40.6Vdc,
all without having a high amount of heat wasted in R2.

The anode supply is about +500V with mains at 240Vac. Mains voltages here change
between 235Vac at high mains load to 252Vac with low mains loads. This means
B+ will vary between +489V to +525V which will not trouble the 6550 with possible
high Pda which at idle with Ea at 477V should be 19.1Watts. The Eg2 is regulated
so Pdg2 is 4mA x 364V = 1.5W. Total Pda+Pdg2 = 20.6W which is less than 1/2
of the max rating for 6550 of 42W.

At no time did I find that the tubes wanted to oscillate at frequency well above
the audio band but it can happen when tubes become seriously overheated.

R4 330 ohm screen stoppers are rated for only 1/4 W and will burn open if Ig2
exceeds about 35mA.
But if any one or more 6550 tubes overheat the Ia and Ig2
flow in R3 500r R3 will increase thus raising Ek which triggers active protection
turn off the power supply in Sheet 3 well before tubes could seriously overheat.

On each side of PP circuit the earthy ends of six R3 are taken to one end of
the cathode FB winding on OPT.

Dynamic Bias Stabilization is used.
The earthy ends of the six Cb caps, 1,000uF, are all taken to top of R56 which are
2 x 4r7 10W. The R6 is a negligible ohm resistance but enough to sense Ik flow
to CFB winding. During class A operation the Iac in R6 varies a maximum of +/- 0.24A.
This produces Vac across R6 = +/-0.56V.
But during class AB operation the +Iac rise could be up to 2.3Amp. We might expect
to see V6 Vac rising to 5.4Vpeak. The top of R6 is connected to Q1 base via
two R7 each 10r0. To prevent excessive base current input there are two d1 diodes
to limit rise of Vce. The Q1 ( and Q2 ) base is being driven by low resistance, needed
because Hfe of the Q1 is not high and base input resistance is lowish.
As soon as class AB action commences and R5 positive voltage peaks rise above
0.57V the bases turn on collector current which drains current from Cb 1,000uF to
prevent them charging up . The Q1 and Q2 act to keep Ek steady but only due to
signal effects. Meanwhile the Ek resulting from steady Ik is allowed to remain
free to move. So all 6550 will have slightly different Ek. The test points tp1 to tp12
are accessible through holes in side of chassis so that Ek for each 6550 can be
checked with a voltmeter every 3 months. Should any 6550 develop a fault so that
excessive Ia flows, then the amp automatically is turned off internally, and my other
schematic sheets on active protection shows how this is done. The Ikdc flow of the
6550 does not interfere with the action of Q1 or Q2 which only act due to class AB
signal currents. These amps will spend most of their life acting in pure class A so
the protection has little to do.
If Q1 and Q2 were not used, Ek rise with class AB could be from +23Vdc to +63Vdc.
Such a rise would cause very serious audible distortions well before clipping.

The active bias stabilization gives the benefits of fixed bias but also allows the
benefits of cathode "auto biasing" which eliminates the need of a bias adjust pot
for each 6550. With 2 channels there are 24 x 6550, and 24 adjustments are far
too many to worry about.

The Output Transformer has GOSS E&I laminations with maximum = 17,000.
Stack is 110mm, Tongue 51mm, window Height 25mm, window Length 76mm.

Wire is grade 2 polyester-imide enameled high temp magnetic winding wire.
Bobbins are hand made with 2mm fibreglass with fibre card centre former
with slight rounding at corners to prevent excessive wire bends during first
few layers.

Primary = 10 layers of 0.6mm Cu dia wire, 106turns per layer.
The primary has 5 sections of 2 layers each with 4 sections for anode circuit
and 1 section used for cathode circuit so there is 20% CFB.

Secondary is 6 layers of 0.9mm Cu dia with 72t per layer. Each sec layer
forms a secondary section between secondary sections.
Each secondary layer is sub-divided into 24t and 48t to allow a wide
range of useful load matches with equal current density in all sec wires, and
with constant winding losses less than 10% even with 1/2 the rated load RL.

The interleaving pattern is 6S x 5P in section sequence SPSPSPSPSPS.
The central P section is the CFB winding.

Bandwidth at 250 Watts with rated resistance loading is 15Hz to 270kHz without
any NFB. 

Not all the possible HF bandwidth is used when global NFB is added, and bandwidth
is reduced with GNFB to 65kHz for stability reasons.

Load match strappings of OPT are ideally for 2r5 or 5r6.
The 2r5 strapping suits speakers between 1r3 to 5r0.
The 5r6 strapping suits anything over 3r5.

The nominal anode load is 1k2 for 12 x 6550, so the loading is easy and each pair
of 6550 in the output stage has a load of 7k2 for a high amount of initial class A.
The Class A power is 34Watts before AB power begins. An absolute maximum
power of 450Watts with sine wave drive is possible.

It is a somewhat complicated task for a non technical person to change output
transformer matching, so the default setting is the 5.6 ohm load match.

Some speakers have appalling impedance curves with dips in Z well below their
nominal claimed Z but this amp design will handle them all with ease.

The only stabilizing zobel network needed is the 4.7 ohms + 0.22uF across the output
terminals. Thus at 154kHz, the reactance Ce 0.22uF = R7 4r7 so as frequency rises
above 160kHz there is an increasingly resistive load across the output terminals.

Any value of capacitance across the output terminals and without any parallel or
series resistance load does not provoke any HF oscillations.

Picture of 300W amp under-chassis with nearly completed work.

You are at 300W amp input/driver and output stages.

Other pages on 300W amps.....

300W amp power supply.
300W amp active protection.
300W amp dynamic bias stabilization.
300W amp power vs load graphs.
300W amp images, tubes with blue glow, and more views of amps.

Back to Power amplifiers.
Back to Index Page