The speaker load affects the voltage
gain of output tubes. The lower the load ohms, the lower the output tube gain where gain is open loop, ie Va-k / Vg-k.
The CFB within output stage above shows that Va-k = 280Vrms,
and Vg-k = 20Vrms, so open loop gain A = 14, but with the series voltage CFB the closed loop gain A' becomes = A / ( 1 + [ ß x A ] ) = 3.11.
Or as observed, A' = Va-k / Vg-0V = 280V / 90V = 3.11.
The amount of applied FB = 20 x log A / A' = about 10dB.
The output stage shown will produce higher output power with half the RLa-a shown but the lower load causes higher THD and the lower open loop gain A
is much reduced so applied becomes lower so THD is about doubled when the RLa-a is halved.
With 25% CFB as shown, and at 62 Watts just under clipping, and output load = 5r4, THD will be < 2% without GNFB. At average levels of 1Watt the THD
will be < 0.2%.
If GNFB is used, the 2% at clipping is reduced to < 0.5%, and at 1W the 0.2%
is reduced to 0.05%.
For nearly all pentodes or beam
tetrodes, the ideal load which produces the lowest
possible THD & IMD at clipping is just above the load
giving the maximum possible pure class A Po, in this case, when speakers = about 10r0.
of KT66, 6L6GC, 5881, 807, 6550, KT88, KT90, KT120.
The use of the larger octal tubes with higher Pda ratings and different biasing voltages requires a mention because some ppl might copy the schematic of Fig 1, then plug in KT88 without any thought about biasing consequences.
6L6GC, 5881, 807 may usually be used instead of EL34 / 6CA7 and the
only change needed is to grid bias Eg1, adjusted to give same Iadc at
Calculating the wanted grid
bias is a Royal PITA for most ppl not yet used to the
variety of what tube might be used or their different idle Pda and Pdg2 and hence
different Iadc and Ig2 at idle.
Calculating Grid Bias voltage, Eg1.
The grid Eg1 bias voltage
can be calculated within +/- 10% :-
[ ( Iadc x Triode Ra ) - Ea ] / Triode µ.
1. This is true for all negatively biased triodes and can be applied to power pentodes and beam tetrodes, and where :-
2. Triode Ra is is average value for Ra curve for Eg1 = 0.0V between 0.0mA and Iadc at idle.
3. Triode u is read along horizontal line for idle value of Iadc.
4. Ea is taken as the screen Eg2 where Ea is higher than Eg2 in cases where true pentode or beam is used including CFB applications.
5. For triode strapped pentodes and tetrodes and for UL, Ea = Eg2.
6550, KT90, KT120.
These larger tubes have Pda ratings from 42Watts to 60Watts.
For class AB1, idle Pda may be higher than for EL34, 6CA7, KT66, 6L6GC, 5881. A suitable value would be 28Watts which is 2/3 of the rated Pda max for KT88/6550. But in practice, there is no need for so much "waste" heat with a high idle Pda, so I suggest Pda = 20Watts. The PSU used for EL34 should not have to work harder to make higher idle current unless it is designed to do so.
Ia = Pda / Ea = 20W/ 450V = 44mAdc. Eg2 = 300V so expect Ig2 = 4mAdc. Total Pda+g2 at idle = 21.5Watts. This is very comfortable for KT88 etc, and the tubes will better withstand drift in bias conditions over time. So, for Eg1 calculations, Ia = 44mA, Ra = 1,100, µ = 6.7, then Ea for calculations = +300V.
Eg1 = [ ( 0.44 x 1,100 ) - 300 ] / 6.7 =
If you plug in KT88, 6550, KT90, KT120 to this amp with fixed bias meant for EL34 with bias set for -29Vdc, The Ia will be about 96mA, and tube Pda = 43Watts and tube will be too hot. The bias pots are set up to give a range of -21Vdc to -37Vdc, and there is not enough range Eg1 with KT88 etc.
a +/-8V Eg1 change centered around -37Vdc, R28 = 4k7 and R22 =
Fig 1 shows an input stage
using one 6DJ8 as a differential amp with one grid input for input
signal and the other grid for GNFB if it is used. The MJE340 acts as a
constant current sink for both commoned cathodes. This makes the two
anode outputs equal and dependent on the ohm equality of their loads.
The driver stage is a 6BL7
twin triode with each triode being about equal to 3 halves of a 12AU7
paralleled. The 6BL7 is set up as a balanced amp with common cathode
resistor 5k6 taken to -100Vdc rail. The anode load seen by each 6BL7
triode consists of 18k for Idc, and 2 x 120k grid bias resistors for
The 18k is bootstrapped so its effective loading on 6BL7 = 81k, and the two Rg are paralell and make 60k, so total anode load = 81k // 60k = 34.5k.
The peak Va swing with 90Vrms = 127Vpk, so max load Ia change = 127V / 34.5mA = +/- 3.69mA pk. The 6BL7 has Ia = 20mAdc and will very easily make the 127Vpk swings at less than 0.5% THD.
Picture of OPTs and chokes for Mr Zel's
These are the OPTs and chokes I
The OPTs have GOSS C-cores. Secondary has 9 sections of one wire layer each, and all
sections are paralleled. Primary has 8 sections with 2 layers of wire each, and each P
section is between the S sections for interleaving pattern
of 9S x 8P, ie 17 total sections using a total of 25 layers of wire. Insulation is 0.2mm polyester. 6 primary sections are used for the anode winding and it has taps for ultralinear use or bootstrapping. 2 primary sections in centre of winding build up are used for 25% local cathode feedback.
The amount of local CFB for the tubes is
nominally 13dB with the loadings shown in Fig 1.
This means the effective Ra of the tubes
and their THD/IMD is reduced to less than what would be achieved if the tubes were triode connected and without
The graph tells everyone what to expect
with a quad of EL34 in an output stage under following
Ea = 450V, Eg2 = fixed + 300Vdc, Fixed Eg1 bias, 25% CFB OPT CFB windings, or use 40% standard UL taps.
OPT gives 5k0 : 5r4. The graph assumes
total of primary plus secondary winding resistance = 5% when sec is loaded with 10r0.
This means that when sec load = 5r0, expect Rw total = 10%, and with sec = 2r5, expect Rw total = 20%. The total Rw percentage equals the amount of output power generated by tubes that is lost as heat in the OPT windings.
With 10r0 sec load, the
4 x EL34 produce 40Watts while the output at OPT sec = 38Watts.
But such losses are benign, and quite low, so there is nothing to worry about. Because the OPTs for Mr Z have C-cores with big winding windows relative to Afe, the total Rw losses are probably lower than I have assumed here.
The amp must not oscillate at
LF or at HF with no load connected, or with any load of pure C
0.05uF and 5uF and tested at any power level, including at very
low F < 20Hz when OPT is saturating at high Va-a levels. The amp
should be tested with 5kHz square wave, and overshoot with any C load
exceed +6dB and ringing F amplitude should decline to zero
within 50uS. A typical tube amp might produce some ringing F
of say 70kHz with say 0.22uF across OPT sec. This is OK because
the amp is effectively a second order bandpass filter. But when there is
pure R loading, there should be negligible ringing.
If the amp does not oscillate with any
pure C load, but has some ringing with square waves, it has
correct critical damping and it will never oscillate with any known speaker load.
The useful test for tolerance of ESL speakers is to use a dummy load with 16r bypassed with 1r5 in series with 2uF.
ESL present low impedance loading at HF. So if testing with this network, never use more than 1/10 full Po, which is about 1/3 of clipping voltage Vo levels.
full power response with GNFB and
pure R load should be between 14Hz and 60kHz at least. Loading
by the Zobel networks across OPT will reduce max possible Po
above 50kHz. At LF, effects of OPT core saturation
limit usable full Po. The saturation F is proportional to OPT
signal voltage. So at 1/2 Vo, -6dB, or when there is 1/4 full
Po, F response
should be from 7Hz to 65kHz with THD < 2%.
The Rout and THD without GNFB is low
enough to not have to use the GNFB. So indeed the sound is OK without the GNFB for many ppl.
Mr Z used his own design for PSU, and to
me its a bit "over the top" but it does give noise free Vdc
rails, so noise at the output without GNFB would be negligible.
My Fig 1 schematic begins with V2a and
v2b. So where is V1?
V1 would be included had Mr Z wanted an
inbuilt pair of line level preamps. Without GNFB the amp needs only 0.44Vrms input for 62Watts into 5r4 load. Mr
Z has 3 x RCA input per channel and uses a rotary source select switch and two gain pots - not shown at Fig
There is a schematic for preamp input
stage at bottom of this page at Fig 3.
V3 and V4 are halves of a 6BL7 twin
triode which works as a balanced amp with 5k6 from common
cathodes to a -100Vdc supply. The dc RLdc resistors to 6BL7 anodes are
connected to taps on anode winding of OPT. The taps supply +450Vdc, and also Vac which is same phase as
anode signal. The anodes have a "bootstrapped" anode RLdc load. This means the anode loading is
much higher than if RLdc was taken to a fixed B+. This means the maximum Va swing with low THD is more
than the 90Vrms shown for clipping levels.
did recommend Mr Z used paralleled
6SN7 or 6CG7 for the V3, V4 balanced amp. But a 6BL7 has
two triodes within and each 6BL7 triode can do about what 3
triode sections of 12AU7, and with less THD, so use of 6BL7 is very good.
Mr Z's PSU.
I have taken liberty with the original
drawing Mr Z tendered after he designed it. I have just
integrated the various B+ and B- supplies so that the 0V rail is more understandable.
MR Z explained......
there is a M.E.N. system which uses three wires. Active goes directly to the fuse for safety, thence with the
neutral to an RFI filter, which is contained in a die-cast box.
A DPDT switch with a HV cap across each set of contacts goes through a thermal cut-out bonded to the chassis. This supply goes to all mains transformer inputs.
When switched on, power is applied to all heaters whether via a regulated supply (for preamp & driver tubes) or direct from dedicated windings on the toroidal transformer (power tubes are referenced to ground via 39 ohm resistors.)
The B+ supply is
also energized upon switch-on from dual windings on the toroidal
transformer. A one ohm resistor limits surge currents going into the double Pi filter, which is
protected by a 500mA fuse. A dual RC network pre-filters the
screen supply before cascading into a split RC network that supplies
the anode loads of the 6DJ8 tubes.
The screen supply is
regulated by a 317 regulator protected by a MJE340 transistor
and 15V zener combination to give an output of ~300V. The regulator output is loaded by a
300V zener string until the screens are switched on.
Grid bias is applied also and cascaded into a 337 negative regulator to supply the current source and tail resistor of the 6DJ8s and 6BL7s respectively.
one-shot circuit based around the 555 timer is also energized
upon switch-on and a timer begins by using the combination of the 10M and
10uF capacitor to activate the relay coil after around 110
seconds. Once the relay is activated it switches the
actual screens to the 300V supply. Also it connects the 96V
supply to the driver and preamp tubes."""
Now for those willing and able to relate
what is said here to the PSU schematic, they may see that I
might do things slightly differently. The fundamental point to remember is that there is
always more than ONE way to build a PSU.
There is other information on PSU at my
relevant pages for PSUs and large power amps.
pics of work in progress.......
The B+ chokes are at each end of chassis, then a pair of OPTs and a large toroidal PT in center.
Work under chassis underway...
Point to point wiring.