Recently, I re-built a bench top power supply for +250Vdc to +500Vdc I had been
using since 1995.
It included a tubed regulator with 2 x 6080 and a 6BX6 and a
string of 50V zener diodes
to change output voltages in +/- 51Vdc steps.
But while testing at +250Vdc, the Vdc at 6080 anodes is +480Vdc at low
current when
the preceding PSU is switched for LC operation. I found that
after awhile, the 6080 began to arc
internally with Ea = 230Vdc.
A change to 6AS7 made no difference.
The reason was that 6080 and 6AS7
just DON'T LIKE excessive Vdc between anode and cathode.

So I figured an upgrade from tubes to solid state was needed and I removed
the tubes and added a
much better rotary switch to change the Vdc range,
and I added measures to prevent switching transients
and minor arcing to

I came up with this design...

Fig 1.

Fig 1 Schematic seems to work just fine and when using S1 to switch the
regulated B+ voltage
upwards or downwards between +126V and +586V.

Basic operation explained.

All the circuit items shown are surplus items left over after a career as
"amp worker" between 1994
and 2012 when I retired. The circuit is basic
"breadboard construction" with a 20mm thick layer of
marine plywood with a
metal front plate, back plate and top cover taken from some other large

dead electronic items which were cannibalized. There is a 150mm dia fan
in rear panel to fan
air onto the heatsink for bjts and to force air from rear
to front past heatsinks for resistors in series
from CLC filter elements and
the bjt collectors.

The unit would become very hot without a fan if used for say 30 minutes
testing tubes in an tube
output stage.

The PT1 largest transformer is rated for about 500VA and gave the useful
ac shown.

The rectifiers are 3 series 1N5408 on each 1/2 of HT sec, so adequate reverse
voltage rating exists.
While such diodes can handle expected currents very well,
and provide enough current to blow the HT fuse
in the HT winding CT to 0V path,
there is also a circuit with SCR C106D which works a relay to disconnect
PSU from devices under tests if the Idc exceeds about 0.6Adc. 

The B+ filter elements of capacitors and chokes may be arranged to form a
capacitor input CLCLC or
choke input LCLC filter to obtain two values of
B+ at output of L3, top of C15.

The switches S1a, b, and c are 3 wafers of a very heavy duty 4 wafer rotary
switch. Each wafer has one
position which is also the pole so 12 positions
are possible with 1 position where the pole connects to itself.

10 positions are used for 10 different switchable output voltages which change
in 51Vdc steps between
+126V and +586Vdc.

For S1 positions 1 to 5, S1a makes the B+ filter to be LCLC with 7H+850uF+1.8H
+180u .

Without any load at output the voltage at output of L3 = +481Vdc.
Looking back to diodes, Rout of PT1 + diodes + L2 + L3 = 150 ohms approximately.
Without any load at the regulated output after L4, the B+ after L3 is kept down
to a maximum of +481Vdc
by the Bleeder R plus R16 to 20 = 14.5k ohms.

S1b controls the amount of  resistance in series between output after L3 and
the collectors of BJTs Q2 to Q6.
The minimum R = 100 ohms, and maximum
= 500 ohms, thus preventing high peak currents in Q2 to Q6
should the output
ever be shorted to 0V. The series resistances are arranged so that during
continuous output
current production of waste heat produced in BJTs from
Current x Vc-e remains much lower than 20Watts,
well within the SOA for the
BJTs which could easily be fused by a short time of high current x high Vce.

For example, suppose position 1 on S1 is selected for 126Vdc and 0.4Adc.

0.4A flows through effective PT1 + filter Rout of 160r plus the 500r of R16 to
R20, ie, 660r.

The Vdc across 660r = 660 x 0.4 = 264Vdc.
The source voltage before loading = +481Vdc and Vdc at Q2-6 collectors
= 481V - 264V = 217Vdc.

Power dissipated in total of 660r = 105.6Watts. Collector to emitter voltage
= 217V - 125V = 92V and heat generated in BJTs = 92 x 0.4 = 36.8Watts

with 9.2Watts in each BJT, which is OK. BJTs are on a good heatsink with
14 fins at
150mm long x 50mm wide, and total area = 2,, with
125mm dia fan facing fins to ensure it
remains cool. Resistors R16-20
plus bleeder are siliconed onto a separate composite heat sink with

area about 1, made with 3mm thick scrap aluminium angles
and plates.

Using the S1 positions of 6 to 10, the arrangement of B+ filter becomes

ie 90u+7H+850u+1.8H+90u. The Rout of the arrangement upstream of
L3 output = 330 ohms.
The same arrangement of switchable series R of
R16 to R20 is possible to limit peak currents
and to share waste heat
so BJTs never get too hot.

S1c selects from 10 different available fixed voltages from the string of
51V x 5Watt zener diodes. The selected fixed
voltage is applied to bottom
of R24 plus R23 which forma  voltage divider to the bottom of emitter

resistors of Q3 to Q6.
The voltage across R23 plus R24, each 4k7, always remains close to
50Vdc so the current flow
in these R and down through zener diode string
is virtually constant at 5.3mAdc. Each 5 watt rated 50V zener diode has Pd
= 0.265Watts, so they stay cool.

But stray higher peak currents may flow during switching transients so R21
330r is added to limit current to 151mA
when voltage is switched +/- 50Vdc.
C19 + C20 give 34uF to shunt the bottom of R24 to 0V and thus the change
output +Vdc is slowed down and hence peak currents everywhere are
well damped.

The output current from Q2-Q6 emitter resistors flows through R30 1r0 which
is a current sensing R, and if the voltage across
R30 is more than about 0.7V,
the SCR C106d latches on and QUICKLY disconnects the output load via a relay
causing excessively
high current.
To make sure the relay does not arc over, it is a DPDT type meant for switching
240V mains at 6A. Each relay section is in series.
if the Vo = 586V, then maximum
Vdc across opened contacts is 293V, less than the peak mains voltage rating of

C21 and C22 are 0.1uF which shunt the open contacts to prevent arcing.
R31 & R34 are actually 4 x 68k, to divide the voltage
equally across each pair
of open contacts.
To prevent excessive current surges when switching Vo
upwards, L4 is used so that HF switching transients see a high impedance

before the low impedance of C23-C25 & C26-C28 = 33uF which shunts Vo
to 0V. However, having an L4 with switched currents
means back EMFs so
100r shunts the L4 to keep its Z max < 100r. If ever the external load being
tested generates noise the
33uF shunts it and there is at least 100r between
load and BJTs.

The operation of the SCR C106d plus relay depend on a floating 12Vdc
supply which has no connection to 0V, but instead referenced to Vo.

PT4 for this Vdc supply is a small 6VA tranny with good insulation between
mains winding and sec. 

The BJTs are further protected to prevent the effects from some voltage higher
than the wanted Vo causing reverse flow input current
to BJTs. Hence a 1N5408
prevents the Vo ever rising above collector voltage by more than 0.7Vpk.

The Q3 to Q6 are in parallel and each with 2r2 current sharing R.
Their bases are paralleled, and driven by emitter of Q2, and the
Q2 - Q6 is an
effective Darlington pair. The BU508a has rather low Hfe because it is a high
voltage rated BJTs so the Darlington
connection is needed to get the base input
resistance of Q2 to be fairly high to make a load which Q1 can easily drive.

To ensure that Q2 base voltage never rises too far above emitters, there are 4
series 1N4007.
I am not sure of the final Hfe of the Darlington pair, perhaps about
100. This means that if 0.4A is output current, Q2 base
input current = 4mA.
Now to get Q1 to be able to control the Vo to be a nearly constant voltage,
I made a floating psu of +200V with 70V sec winding of PT3.
The 200Vdc is RC
filtered and shunt regulated by zener diodes to make a stable +150Vdc.
This acts as a battery like Vdc supply
to allow Q1 MJE340 to operate as a single
ended amp BJT where its emitter is kept at about +24V above the voltage at bottom
of R24. The Vc-e change is controlled by the small Vb-e change between zener
voltage and R23-24 divider voltage.
The actual V changes are quite small, but
what is really going on is the change of current in Q1, which changes current in Q2

base. With no output load, Q2 base input current is tiny, so Q1 is turned on more to
keep current in R22 10k0 at about 5mA.
When heavy output load current = 0.4A,
then Q1 turns off to allow R22 to deliver more current to Q2 base which raises its
so keeping Vo fairly constant. The action of the regulator is entirely similar
to so many described in text books which deal with very
basic issues of low Vdc
regulation. In my circuit, I have floated a typical low voltage regulator which uses
discrete BJTs which tend to
be more reliable than using a floating op-amp.
The operation of Q1 is best considered in terms of current change because the
actual voltage changes are so low between the base,

emitter and collector.

The output resistance measured at the Vo terminals = 6 ohms approximately.
This means that for a change of Iout = +400mA,
Vo changes by -2.4Vdc.
Now if I am testing a class AB output stage and B+ is a typical +483Vdc, and I out
varies between
60mA at idle to 200mA at max Po into a low amp RLa-a, then Vdc
change is an utterly negligible -0.84Vdc.
Therefore accurate measurements of
tube performance can be ascertained.

For example, it is said you can get 150Watts with 2 x KT90 in Class AB1.
But you SHOULD regulate the anode supply and you
MUST regulate the screen
supply to get correct operation. If the anode supply is not to be regulated in a real
amp, then at idle it must be allowed
to be higher than when tubes make 150Watts
where the Ea must not be less than about +520Vdc. At idle, the Ea should not be
able to rise to say
+600V with a danger of arcing at octal sockets.
But the screen supply must NOT be allowed to rise with anode supply and should
be SHUNT regulated to +500Vdc max above cathodes at 0Vdc.
The screen shunt
regulator should be a simple type with series resistance feed from anode supply to
some series HV BJTs.
If too much screen current flows,
then current creates a highish voltage in the screen
feed R thus causing the screen voltage to descend below the shunt regulator voltage
so the most
likely outcome is that the series R fuses open and tubes are turned off
by low screen voltage.
Having a series type regulator for screen supply
in any amp is a big mistake because
Eg2 is forced into being high, and screens can easily overheat and tubes fail as a

See my 8585 amp page for details of the screen shunt reg supply.

For testing beam tetrode and pentode power tubes, I may use a shunt regulator
mounted on the test circuit for the tubes, but all B+ power comes
from the
regulated supply of Fig 1.

Besides the B+ filtering with L and C and the BJT regulator element, I have put
in a variable bias supply after PT2 with -Vdc
controlled by wire wound pots.
PT5 provides power for the fan with 24Vdc rated DC motor.
But I have applied only +16vdc, and fan runs slower, yet fast enough, but
more reliably.
The ON red LED is also powered from PT5.

I did think of alternatives with a simpler schematic. It may have been possible
to make a big power transformer
with lots of low voltage windings for heaters
and bias, and one winding with 10 taps to give approximate
50Vdc steps in B+
from +100V to say +600Vdc. If the PT design is very good, and PT rated for 1kW,

then the natural regulation is good with an LC filter.
But the choke must be high L value, and hence for low
winding resistance it
must be huge. If the choke is say 50r like the 7H choke weighing 3.5Kg I have
in Fig1,
then with Idc change of 400mA the Vo changes -20Vdc. If the Vdc at
output is nominally 126Vdc with no load, then it will
drop to 106Vdc at 0.4A
which is a -16% Vdrop, and far too much. There is always some PT winding
resistance plus the choke
resistance and it is difficult to make a PT with
Rout < 3 ohms and have a choke with Rw < 3 ohms.
However, with switched
Vac input from switched taps on a HT winding then an SS regulator can be
used with a constant 100r
series R from B+ after LC to BJT collectors.
Instead of HV rated BJTs, HV rated mosfets may be better for the simplicity
that becomes possible. Instead of having a
gain BJT like Q1 in Fig 1, the
gates of mosfets may be switched to points along a resistance divider instead
of zener string.
The R divider might be 12 x 1k0 between 0V and a shunt
regulated +600Vdc. The mosfets have high gate input Z like the grid of a tube.

Typical power mosfets in source follower mode have Rout = 1r0, so with say 4
paralleled the Rout is very low.

I'm happy with my new gadget PSU because I didn't have to buy one single
new thing to upgrade what I used for so long.
The challenge was to better
understand the failure modes of BJT regulators. The first HV regulator I made
did use BU508, but only ONE,
and it failed twice when I wanted over 550Vdc.
I felt forced to use tubes. But they too had a weakness. But in two tube amps

I have BJT plate supply regulators which have been providing a fixed 450Vdc
for 16+ years and a couple of failing tubes and short
circuiting of outputs did
not worry what I had made, so I felt I should again try BJTs with my increased

One might be tempted to make a shunt regulated anode supply. DON'T.
It means you may need to have a gang of devices + resistances with normal

Ia at 0.3Adc and with 550Vdc wanted Vo, you waste 165Watts if device current
is low. Where test current is high, use series reg,
and shunt reg for where test
current is low, say for a screen supply.

Back to Education and DIY directory

Back to Index page