The power amplifier build part 4, power supplies

-

Part 4 – Power Supplies



At first glance, power supplies may seem straightforward: transformers, rectifiers, filter capacitors and you’re done. In this project, that’s true for some of the low-voltage rails, but the bias and especially the screen supply needed considerably more thought.

High Voltage Supply

The high-voltage (HV) supply is built around a 3 kVA UI-core transformer, followed by a full-wave rectifier, an oil-filled capacitor, and bleeder resistors.

Since Swedish homes typically have 3-phase service, I wired the two 230 V primaries in series across 400 V. This helps balance the load between phases and reduces light flicker during key-down. The secondary winding has multiple taps; I selected a tap that provides about 3.2 kV DC under load.




The two smaller transformers are 400-230 V transformers.
I run them from 230 V, with their secondaries connected
in series to get 800 VAC, rectified and smoothed to 1100V.

The board visible is the HV detect, HV multiplier and meter shunt board. A series of resistors drop the HV down to safe levels by a factor of 1000, eg. 3500 V becomes 3,5 V at the output.
This is used to operate the on board relay as HV detect, which in turn activates the screen grid transformer contactor. It also feeds the divided down HV to the Arduino Mega, which reads it and translates it to KiloVolts, and drives the analog meter.
Remember, we do not want screen grid voltage applied before we apply HV to the plates, thats why I included this function.


Screen Supply

This was the most time-consuming part of the project. Several designs were tried, and quite a few MOSFETs gave their lives before the regulator was stable. I eventually settled on a series-regulated design.

Why not a shunt supply?

A traditional approach is a shunt-regulated screen supply. Here, the voltage is dropped across a series resistor, and an active device (tube, transistor, or MOSFET) shunts excess current to ground to hold the screen at a stable potential. The great advantage is that if the tube ever drives current backwards into the screen supply, the regulator can safely sink it.

The alternative is a series regulator, where a MOSFET (or similar) sits in series with the screen and adjusts its effective resistance to stabilize the voltage. The drawback is that if current flows from the screen back into the supply, a simple series regulator cannot sink it. The screen voltage will rise until something gives.

Negative screen current explained

Normally, the screen grid current flows from the supply into the tube. But under certain conditions – when the plate (anode) voltage falls close to or below the screen voltage – the screen grid can act like the anode. Electrons can be collected by the screen instead of the plate. The result is negative screen current: current flowing from the tube into the supply. If that happens, the screen voltage can soar if the supply cannot sink the negative flowing current.

This behavior is especially associated with:

  • High-gm tetrodes and beam tetrodes (e.g. 4CX/4CX-series, 4-1000A, some TV sweep tubes)

  • Tubes operated with low anode voltage compared to screen voltage

  • Situations with large RF swing at the plate

In those cases, a shunt regulator is the safe option.

The GU-81M, however, is a large, relatively slow tube of 1930s design. Its screen structure and operating parameters make it very unlikely to exhibit negative screen resistance under normal HF amplifier conditions. Besides, all published designs all used series regulation. That convinced me to use a series regulator without the heat dissipation penalty of a shunt design, idling at over 420W dissipation for 400mA capacity.

My implementation

  • Early version: MOSFET series pass element, with a stack of 8×100 V zener diodes for reference. This worked, but temperature drift was severe: voltage crept from 800 V up toward 1 kV as the zeners warmed.

  • Later version: Added closed-loop regulation using an op-amp comparator and feedback divider. This stabilized the output, but the first attempt oscillated.

  • Final version: Based on HA5KJ’s regulator design, effectively a hybrid series regulator with shunt-like feedback behavior. By lowering the MOSFET gate resistor from 100 kΩ to 22 kΩ, stability improved and the supply ran quietly, though with a narrower adjustment range.

PCB layout turned out to matter just as much as the schematic. My first board oscillated and destroyed MOSFETs; the second board, closely following my home etched, experimental, single-sided layout, has been reliable.

Update September 2025.

Since I wrote this, I have tested the linear. During initial adjustments, the screen regulator exploded. Literally. R2 evaporized, together with the metering shunt and all the mosfets. It started with a small hiss, and before I got time to react, it took off. Due to the current meter shunt being taken out, it also took out the Arduino Mega by piercing the analog input with high voltage. I had to do some research and failure analysis, and I've come to the conclusion that C2, which spans the feedback divider network might be wrong. I had by accident installed a 1 nF capacitor instead of 100 nF. I had also installed 100 nF from potentiometer wiper to ground and this, combined with a no load situation might be the cause of oscillations. From my understanding, you should install one of these capacitors, not both, because you may disrupt the phase margin. Each part of the regulator introduces response lag due to capacitances. Q3 has gate capacitance which takes some time to charge. Q1 and Q2 has gate capacitance which also takes time to drain or charge depending on which way we need to regulate. Series resistors cause lag. If the output of the regulator is delayed enough, we might end up with positive feedback depending on how we set up the feedback circuit. Also, the load we are feeding also impacts the response. The failure occurred with the amplifier idling. No current was drawn by the plates, and we had cut-off bias on the control grids. It could have been that the regulator bottomed out due to low load. It tried to maintain 800 V with no current being drawn, and ended up "hitting bottom" and began to oscillate due to incorrect feedback circuit.

This is (for me atleast) a complex area and I'm not quite sure how to tackle this. A defined load (bleeder) is a must. I will add a 30 K resistor stack to always have the regulator see a load, and shunt some of the non-linear load the screen represents. Regarding feedback capacitors I do not know. I have 3 choices. 1. Connecting a large-ish ,up to say 1 uF, capacitor across the potentiometer wiper and ground, which should filter out any quick changes and high frequencies. 2. HA5KJ placed 100 nF across Vout of the regulator and wiper so that's an option. 3. Omit them all. A lot of low voltage series regulators does not have any capacitors across the feedback voltage divider. This does however lets Q3, the error amplifier, to be open for everything that comes back down the feedback path.

What feels most "right" is to add a capacitor of say 100-500 nF across the wiper and ground. This should represent a very low impedance for frequencies above 90 Hz. Besides, it would only call for low voltage capacitor since the most voltage it´s going to see is around 27 Volts, ie, the zener reference voltage.

If all else fails. I´ll resort to eight OB2 gas regualtor tubes and a series mosfet. Just for kicks!

To be continued.


HA5KJ series regulator for 2xGU81M



My version based on HA5KJ. Note R1 is decreased to 22K to get smoother regulation.
Also, i added one more zener series resistor because I´m using higher voltages than HA5KJ










A simple series regulator. It failed due to temperature drift.
A different approach could be to use VR tubes, say 8 OB2 in series for 864 V regulated.
I got a call from a fellow SM amateur, who did just that, and claimed they dont drift the same as zeners do. Might be worth a try.




Experimental ADD-ON board with the HA5KJ regulator. Later made a new proffessional board


Bias Supply

The bias supply is much simpler: a relay-switched zener shunt regulator.

The control grid draws negligible current, but the supply must still be able to sink current in case of RF overdrive. In my design:

  • Standby bias is about –320 V

  • Operating bias is stabilized at –220 V

Low-Voltage Supplies

The remaining supplies are conventional:

  • Unregulated 24 V for stepper motors and vacuum relays

  • Regulated 12 V for logic and control circuits

A 300 VA toroidal transformer with added turns provides the necessary voltages.


Comments

Post a Comment

Popular posts from this blog

My HP8640 died!

-
Image
 My HP8640 died! This is me catching up on past projects. I did this back in 2022, figured someone would benefit from this information, since the 8640 is still around in great numbers at hams all over the globe. I bought a nice looking HP8640 at our local hamfest back in 2022. It was working when I got it home, but a day or two after I got it home, it started acting erratic, with RF ouput dropping out. eventually, there was no output whatsoever. Did some tests, and got to the conclusion that the cavity oscillator was not oscillating. Following various clues online, I tracked down the issue to the A3 unit (cavity), and a series resistor, which had drifted in value. By measuring the series resistance between the + feed and the A3Q1 socket and the - feed and the A3Q1 socket, I could see that the + lead seemed reasonable at 370ohms, whilst the - lead had drifted to 19Kohms, where I was expecting around 900ohms. The A3Q1 (HP21 pictured below) transistor sits under the small hex cap om t...
Read more

813 grounded grid linear amplifier

-
Image
2 x 813 grounded grid linear amplifier  This year marks 10 years since I built this amplifier, and it´s been in use on almost all QSO´s I´ve had since then.  When I got into amateur radio back in 2013/2014, I quite soon understood that I needed a vacuum tube amplifier. SB220´s and L4B´s were high on the wanted list, but there wasn´t anything for sale at that moment. I stumbled across a pair of 813´s and sockets, and that´s what set the wheels rolling. I begun by collecting the needed parts. High voltage transformer, variable caps, meters and such, and after that, it´s mostly a job of shaping metal and drilling holes. Most parts collected An aluminum chassis was bent using the sheet metal brake at work. Reinforcement extrusions was bolted to the underside of the chassis where the HV transformer would go. With a finished box a rough layout was determined, and the hole drilling excercise begun. Getting a feel of where to place things Lot of parts to fit. Note that pictured bandsw...
Read more

ARC-5 receiver power supply, an in depth look.

-
Image
 The power supply is essentially a vibrator power supply, but with an IC and mosfet instead of the mechanichal vibrator used in the old days. The schematic can be seen below: It´s actually quite straight forward:  The CD4047 is operated as an astable multivibrator, providing two , out of ohase, square waves, driving the gates of the FET´s. The FET´s ground one end of the transformer winding, causing current flow, and induces voltage to the secondary winding (which in this case in the old primary) Switching frequency is determined by capacitor C3 and resistor R4. These values in the schematic gives approx. 50Hz switching frequency.  Optional 28V operation: I included holes on the board to hold a zener diode at the input of the CD4047 (max vcc 18v) in case 28v operation would be used (non-modified filament strings). A different transformer is also needed. There is a transformer that fits the board, 14A-10R-36, by the number we can tell that it´s a centertapped 36v seco...
Read more