[patrick-turner]

On 11 Nov, 11:32, "Alex Pogossov" wrote:
"patrick-turner" wrote in message

SNIP,

Floating supply sounds not too far fetched, given that similar approach has
been entertained in a "cyclotrone" topology.

But we NEVER see a single ended Circlotron! That's what I am sort of
proposing. Turnotron? maybe, I don't give a F what ppl call it, as
long as it works.

In your case, a good thing is that the floating supply (for plate only) can
be unregulated and unfiltered -- just a bridge and 220uF or so. Another
smaller power supply, well filtered though, will be powering the screen of
the KT88 and the rest of the circuit.


Yes, the filtering in the anode supply doesn't need to be perfect
because anode resistance is so high when pentode/tetrode is used, and
the local NFB works to oppose any PSU hum in load, ie, across the
whole OPT which has a tube across it acting as a far lower Ra than it
really has, and with RL shunting it, so if Ra'// RLa = say 500r, and
internal = 10k, then hum of 50mV appears as only 2.5mV which is
transformed to 0.135mV at OPT sec if OPT has 18:1 TR.
In the case of Jadis, hum at C2 470uF at OPT primary = 40mV. OPT
primary load Rla is 1k8 in series with Ra of 2 x 300B = 400r so hum
across OPT = 33mV.
OPT TR = 18 : 1, so 1.8mV appears at output if there is no GNFB. I
have not tested the noise performance yet of the Reformed Jadis, but I
know what to expect to find! I may yet be forced to remove the almost
useless Jadis filter chokes and replace with something with at least
1H, to reduce hum by say 20 dB.
In a good SET amp with no GNFB, hum 0.5mV is travelling well, with
input grounded. To see if hum is entering the signal path one may
ground the OP tube caps with say 2uF, and any hum in OP stage may
become visible on the CRO. Then you move te 2uF cap further towards
the input to the driver stage gris, and see if hum increases, and it
is does, voila, a source of hum is identified. If its F is 50Hz, Ah,
then you have maybe heater to cathode leakage in driver tube. At the
input tube, you need to ground input with shorting RCA plug, and also
short the 100r GNFB resistor between V1 cathode bias network and 0V.
Eventually, the source of the crap that some tube amps serve up to you
will be found, and then corrections can be made to 0V rails and
ground, ie, one does whatever it takes to get the noise low!!!!
My customer here has very sensitive speakers, and hum exceeding 1mV is
not acceptable.

Also I think that stray capacitance noise will not be so much of a
problem -- because it leaks only in 60% of the OPT, which has 36% of
impedance. Besides, GNFB will kill the residue of noise if any.

One may like to think like that, but all too often when we measure
stuff we find that what we hoped to expect was BS, and we have to
effect a cure with a dive into parts bin, a good think, and further
unforseen necessary a soldering.

The other bother is the drive voltage of 100Vrms. Be brave men, and
don't be pussy wussies, and use an EL84 in triode with say 4k7 plus
60H choke in series as the feed to +400Vdc with Ia at 14mA. That
should give you Ea at about 322Vdc, OK for trioded EL84, and swing
will exceed 120Vrms with 2H 1.5%, which should cancel with 2H of OP
stage - rather nicely, you shoud find.

There are plenty of brave men, but not many 60H chokes in junk boxes. What
if make a 500...600Vdc supply (doubler) and use an enormous glazed wire
wound resistor as EL84 plate load?-

I've done just this on a number of amps, easy to make HV at fairly low
current of say 14mAdc. If one has a 320VAc winding then you get 432Vdc
which can be R&C filtered for Eg2 with shunt reg, then also have a cap
and diode to make about +700Vdc. If Ea of EL84 is +330Vdc, then RLdc =
370V/14 = 26k0, and liberates 5.2W of heat, and if tube shorts, gives
10W, so the 26k could be say 4x4k7 + 1x6k8 = 25.6k, and each R is 5W
wire wound. The following grid bias R can be connected to end of
grounded OPT primary so its ac load ohms are thus raised by
bootstrapping. So if Rg were 2 x 47k = 23.5k, and boostrapping
increases this to say 90k.
Total AC load on EL84 becomes 26k//90k = 20k, and gain = 20 x 20 /
22.5 = 18.2, not to bad at all, and peak Ia swing at 140Vpk = 7mApk,
ie, 1/2 Ia, also doable, with low THD. But the choke plus say 4k7 to a
screen supply of +400Vdc is BETTER. Then the 25 dissapears, leaving
only 90k loading of Rg.
Gain approaches µ, distortion seems to vanish, and music is better.
The 4k7 series R has a gain shelving function so that gain slightly
reduces where high voltage swings are unlikely to appear at F below
20Hz and above 20kHz. Chokes of say 60H may have shunt C = 300pF, and
the 60H = Ra = 2k5 at around 7Hz. Therefore the use of some GNFB works
just fine even though it seems the choke has introduced another damn
pair of R&C and L&C time constants and phase shifts into the open loop
character. BUT, also JUST AS GOOD is to settle for a B+ rail of say 2
x Ea = 660Vdc, and the use a pir of 2SK1924 in series and arranged as
a as a constant current source. Its easy to get them to act just like
5MO shunted by about 70pF. The use of TWO mosfets in series means that
if Ea is shorted, you have only 1/2 the rain voltage across from drain
to source. MJE340, MJE350, KSE340, KSE350 are also usable bjts. They
can be arranged to sort of work like a choke by offering a lowish load
of say 20k at F below 2Hz, thus stabilising Ea a bit. There are
several ways of doing this stuff. Another is to use the 26k RLdc, but
then bootstrap the top 10k portion to the cathode with elcaps, leaving
the 16k maybe raised to effective 64k.
Patrick Turner.