Thyratron Power Supply

Introduction

Ever since watching CuriousMarc's video about a mercury-vapour thyratron teletype power supply I wanted to build my own mercury-vapour thyratron supply. I decided that if I built such a supply it would have to be useful to test vacuum tube circuits and have constant voltage (CV) and constant current (CC) regulation capability as per a standard benchtop power supply.

The Thyratron

The thyratron tube is the predecessor to silicon controlled rectifiers (SCRs) also known as 'thyristors', a combination of the words 'thyratron' and 'transistor'. Thyratron tubes are hot cathode triodes (or sometimes tetrodes, having two grids) which are filled with an inert gas such as argon or mercury vapour. For high current pulsed applications, hydrogen is used as a filling. When sufficient current is conducted, the inert gas ionises and an arc discharge occurs between electrodes, thus conducted current is no longer limited by space charge as per a regular vacuum tube. The voltage applied to the grid is therefore used to control the anode-cathode voltage at which gas ionisation occurs. Inert gas hyratrons have been made obsolete by semiconductors however hydrogen thyratrons are still used today in pulsed laser systems as well as other niche applications where solid state RF electronics are are not robust enough to handle the extreme instantaneous power dissipation that thyratrons can endure.

The General Electric 393A thyratrons used in this project are a directly heated type, with a oxide coated filament/cathode that glows a dim orange in operation. A small amount of metallic mercury typically condenses around the bottom of the glass envelope when not in use. Upon powering the filament, the heat causes the majority of the mercury to be vapourised into a colourless gas. A small amount of argon gas is also present in the tube to allow operation (albeit with a higher forward voltage) with a reduced heating time, before most of the mercury has vapourised. The grid electrode is a closed metal cylinder encompassing the cathode with a small slot on top to expose the cathode to the anode - a metal disc held at the top of the tube. Not all 393A tubes feature the perforations in the side of the grid cylinder. With gases deionised, the thyratron acts as a triode tube whereby the grid voltage controls the current between anode and cathode. When a critical current (a few milliamperes or so) is reached between anode and cathode, the gas ionises, providing an arc discharge path between anode and cathode which allows the tube to conduct amperes with a low voltage drop (approx 10-20V). Deionisation requires the anode-cathode voltage be reduced below 10-20V.

Phase Control

Two thyratrons can be arranged as a full-wave rectifier as shown in the following simplified schematic:

A full-bridge configuration using four thyratrons for single-phase operation was typically not used, as this would significantly complicate the drive circutry to drive each of the grids (3 different floating cathode voltages) and also introduces additional losses due to two tube forward voltage drops and twice the required heater power. A voltage doubler using two tubes is also possible, however suffers the same complications in driving the two gates which are floating with respect to each other. The two tube full-wave configuration as shown above has the advantage that both grids share a common cathode voltage which greatly simplifies the gate driving circuitry.

The simplest gate control regime is to apply a DC voltage to both gates with respect to the common-cathode node. Since the critical grid (-to-cathode) voltage which the tube ionises depends on the anode-to-cathode voltage (see datasheet excerpt below), a rudimentary amount of phase control can be achieved.

The limitation of using a DC grid control regime, however, is that the firing angle can only be varied between approximately 0degrees and 90degrees, giving a conduction angle between 90degrees and almost 180degrees. See the following animation depicting the relationship between a varying DC grid voltage (orange) and anode conduction (shaded light blue) for a 500Vpeak AC applied to the anode of a 393A tube.

While this is useful for regulated power supplies with an expected minimum load, it is less than ideal for a bench power supply where the output voltage and/or current may be varied all the way down to zero. When a conduction angle less than 90degrees is needed to maintain regulation the powersupply will start to 'pulse skip', alternating between 90degree conduction and 0degree conduction which will result in a large amount of output ripple with frequency components below 100Hz (for 50Hz AC input).

An improvement can be made by summing an approximately 90-degree phase-lagged AC mains signal to the DC grid control voltage. The AC component on the grid compensates the non-linear anode vs. grid voltage characteristic of the thyratron as well as increasing the DC control voltage required to trigger the tube at firing angles <90deg and decreasing it at angles >90deg. This this gives a continuous positive relationship between a DC control voltage and conduction angle from near 0deg to near 180deg as shown in the following animation:

Download:  MATLAB scripts used to create the above animations [.zip 1.3kB]

This can be practically implemented by the addition of a small step-down transformer and R-C phase shifting networks thus adding the appropriate phase-shifted AC signal to each thyratron grid from a common 'DC' grid control signal.

The exact tuning of phase lag and AC amplitude produced by TR2 depends on the characteristics of the thyratron used, and, in the case of this powersupply the limitation of the grid control circuit to only produce negative DC control voltages (w.r.t. the output/cathode). Since a positive DC voltage would be required to counteract the negative swing of the AC signal, if the AC amplitude is too high the ability to trigger the thyratrons at a low firing angle and achieve a large conduction angle is lost.

Practical thyratron rectifier circuit

The practical implementation of the previous rectifier circuit is shown below.

Three Triad VPT230-700 mains transformers - TR1, TR2, TR3 - are connected with interleaved 115Vac secondary windings to create a single centre-tapped 690Vac secondary winding. The interleaved arrangement is used such that secondaries on a single transformer are placed either side of the centre tap, thus the transformer core/primary is subject to full-wave rectification instead of half-wave, preventing core saturation. This arrangement, however, places two other 115V secondary windings between the two 115V windings on a given transformer, thus we are potentially placing 3x the manufacturer's expected voltage between secondary windings. Secondary-to-secondary isolation voltage of multiple-secondary winding transformers is often unspecified. Many dual-secondary toroidal transformers (particularly ones with lower secondary voltages) may have nothing more than the enamel insulation of the magnet wire separating secondary windings as they only need to withstand the rated voltage of the secondary windings, since the transformer is designed to operate with windings connected in either series or parallel. I have found no issue with this arrangement with these transformers but if you replicate this arrangement you do so at your own risk. A 'safer' configuration would be to two or ideally one transformer which provide a sufficient centre-tapped output, however, 115/230V toroidal transformers were chosen as these are significantly cheaper than an equivalent higher voltage transformer(s).

Triad VPL2-10000 transformers power the thyratron filaments, with DC output taken from the centre-taps. R-C snubbers C5, C6, R9, R10 damp ringing rectifier circuit during thyratron deionisation. A Hammond Manufacturing 193U 0.2H choke allows for the negative resistance characteristic of the thyratron tubes, limits and lengthens rectified current pulses, reducing RF emission from the rectifier. A small Myrra 44130 2x15Vac transformer is used to provide AC compensation to the DC grid control. The AC amplitude can be adjusted by resistive divider network R5-R8 and the phase lag by R1, R2, C2, C3. The AC amplitude of the grid signal was lowered (in simulation) until a sufficient minimum firing angle (around 50degrees) was achieved and then phase lag was increased to restore the ability to operate down to a suitably low conduction angle (approx 15deg). The resulting AC amplitude at the grids is approximately 8Vpk-pk with 110degrees phase lag.

Shunt resistor Rshunt is used by the current control circuit to sense the load current by the voltage drop between ground and PGND (Power Ground). Switch S1 is the normally open contacts of a relay used to disable the high voltage supply when the output is disabled, as will be detailed in the following section.

'B-' control circuit supply

As the output is designed to be adjusted all the way down to 0V, a dedicated supply rail is needed to power the control circuitry which we will call 'B-'. A power transformer provides a 115Vac secondary which is voltage doubled (voltage across C2, C4) using two 6C4P-EV rectifier tubes and R-C filtered by R2, R9, C1 to produce an approximately -200V supply rail.

An output-enable circuit is built around a 0A4G cold-cathode thyratron tube. The tube controls a relay, the coil of which is shown as Lrelay here. This relay simultaneously enables/disables the high voltage supply (previous section) and voltage control (following section) to gracefully enable and disable the output of the supply on demand. R1, C3 implement a startup timer of approximately 1 minute, preventing the output being enabled and high voltage being applied to the thyratron rectifying tubes before they are sufficiently warmed up.

Switches S1 and S2 are normally open contacts of a momentary toggle switch. Closing S1 ionises the 0A4G thyratron and enables the output while closing S2 deionises it and disables the output. C6, C7, R13, R14 provide arc suppression for the switches. The network R11, R12, C5 is carefully tuned to momentarily bring the anode-cathode voltage of the 0A4G low enough to deionise it, while not operating the relay if S2 is closed while the output is already disabled. Initially I had planned to just place a SPST toggle switch in series with the relay coil, however if the power supply experiences a temporary loss of mains power (either from operating the main power switch or line brownout), the voltage across C3 takes a long time to bleed down, thus the delay is not enforced when mains power is restored and the thyratrons may be operated when cold. With the circuit shown above, any loss of mains power longer than a couple of seconds will deionise the 0A4G tube and require it to be manually retriggered by the momentary switch, preventing damage to the thyratrons.

Constant voltage control circuit

Voltage control is achieved using a 6N2P-EV (Cyrillic: 6Н2П-ЕВ, similar to 12AX7/ECC83) dual triode tube, T1, utilised as a differential amplifier. An 0D3 voltage regulator tube provides a -150V reference voltage from which voltage control potentiometer VR1 is user adjusted to generate a variable negative reference voltage Vadj which corresponds to 0-360V at the power supply output (via feedback divider R3, R4, R12). SW1, the normally closed contacts of the relay in the previous section, pulls down on Vadj when the output is disabled. C1 slows the adustment speed preventing a sudden inrush current when the output is re-enabled.

Since the anode-cathode voltage of the differential pair can exceed 400V when the output is adjusted to 360V, a second 6N2P-EV tube, T2, is placed in quasi-cascode configuration with T1 to halve anode-cathode voltages of each tube, keeping them within maximum ratings. C4 slew-rate limits the differential pair while C3 filters ~50Hz ripple from the thyratron grid voltage which would otherwise cause unbalanced conduction between the two thyratron tubes.

Constant Current control circuit

Current control is implemented with a second quasi-cascode differential pair much the same as the voltage control. A second 0D3 voltage regulator tube provides a -150V reference (-150_B) with respect to PGND. The voltage drop across output shunt resistor Rshunt varies PGND up to 5V below ground @ 600mA output. VR2 provides approximately 5V of adjustment and therefore sets the current limit from 0-600mA when Cadj is compared to the -150V reference. C5 implements slew rate limiting and is tuned such that the voltage (CV) and current (CC) control circuits do not cause each other to oscillate.

CC/CV 'magic-eye' indicator

Having some spare room in the chassis and an unused 6.3V transformer winding, I thought I would implement an indicator to show if the power supply was operating in constant current or constant voltage. By comparing the CC and CV circuit nodes (see previous schematics), we can obtain a differential output to drive a 6AL7GT "Magic-Eye" tube which provides a visual indication of which circuit (CC or CV) has dominant control of the grid voltage. "Magic-Eye" tubes are a type of cathode ray tube. The 6AL7GT contains three active regions which can be independantly controlled.

The above circuit was devised. High impedance voltage dividers (R3-R6) level-shift the CC and CV signals to allow the differential amplifier and output to be referenced to the B- rail. Current draw must be kept low (below 10uA) from the CC and CV nodes to avoid affecting the performance of the control circuits. The result of needing this level-shifting is that the circuit has limited common-mode rejection and gain. This limitation, however, is not a problem as the differential outputs are displayed as two independant analogue bar graphs on the 6AL7GT tube. As such, the CV/CC indication is made by comparing the relative size of two bars rather than absolute size. P1 and P2 areas (see below) are used to represent CV and CC operation respectively, while the P3 area is not illuminated. THe limited gain of the differential pair is also used to advantage as neither triode can be driven into cutoff when the output voltage is high, thus maintaining substantial voltage drops across R2, R7 and keeping the anode-cathode voltages within the maximum allowable (300V) of the 6N2P-EV tube.

In hindsight, powering this circuit from a negative supply rail that tracks the grid/output voltage would drastically improve the common-mode rejection and therefore improve the readability of the display, however this would also require at least one additional tube to achieve this.

Circuit simulation

The above circuits schematics were put together and simulated in LTspice.

Download:  LTSpice simulations [.zip 48kB]

Simulated Performance Characterisation

The regulation and ripple were characterised from simulations for both constant-voltage and constant-current modes of operation at a range of voltages (100, 200, 300V) and currents (75, 150, 300, 600mA)

Constant-voltage measurements were conducted by setting the value of VR1 to achieve the desired output voltage without a load, VR2 was maximised to eliminate the effect of the current regulation. Different resistive loads were connected to achieve the desired load currents (75, 150, 300, 600mA). The output voltage droop (in %) compared to the no-load condition was recorded for each load current.

For constant-current measurements, VR2 was set while driving a resistive load that would achieve a 100V drop at the desired set current. VR1 was maximised to eliminate the effect of voltage regulation. Different resistive loads were then substituted to achieve 200V and 300V respectively. The difference in load current (%) with respect to the initial set current was measured.

Output ripple (Volts peak-to-peak) was also measured for each load current.

Voltage regulation is improved at higher output voltages and ripple is proportional to load current. Current regulation is significantly poorer than voltage regulation, due to the relatively small signal range (5V swing) for setting current and the influence the output rail voltage has on control triode anode-cathode voltages and therefore grid voltages. Despite this, I've considered the current regulation performance acceptable since I intend to use it mostly as a type of overcurrent protection rather than a strictly regulated constant current source.

PCB Layout, Assembly

All subcircuits excepting the CC/CV indicator are laid out on a single PCB, mounted to the underside of the enclosure top panel. The PCB is a simple 2-layer design on 1.6mm FR-4 to reduce cost. Necessary PCB track spacing for creepage precludes being able to place a ground plane. The lack of ground plane was not a concern as bandwidth of the circuits is extremely low and EMI emitted from the thyratron tubes is inevitable since they are not shielded in any way. Thought was given to position high profile components (e.g. capacitors) such that they nest between the power transformers mounted to the bottom panel in order to achieve the lowest profile construction. I've chosen to use chassis mount tube sockets with short wires to the PCB as my personal opinion is that PCB mount tube sockets are prone to cracked solder joints due to heat cycling and physical stress on the PCB solder joints when inserting and removing tubes.

Download:  Altium project files [.zip 6.1MB]

Situated on the bottom panel of the chassis are the main toroidal power transformers, CC/CV indicator circuit and filtered air ventsto allow cool air to enter the chassis for natural convection cooling. Strategically placed ventilation holes around the tubes in both the PCB and chassis top panel expel air from inside the chassis, assisted by air convection around the tubes mounted on the top panel.

A note about balancing of thyratron tubes

An oversight in the design is that it assumes both thyratron tubes have identical control region characteristics, thus full-wave rectification occurs. In reality, a small difference in matching between the tubes may result in only one thyratron being ionised under certain conditions. Ideally we'd have a way to independently adjust the grid DC offset voltage for at least one of the tubes in order to match their control characteristics. As a compromise and as a simpler modification to the existing circuit I added a trimpot in parallel with one of the 10K resistors in the AC compensation network (see below). This has the effect of advancing the AC compensation phase to one of the tubes (the one which does not ionise as easily) which was able to achieve acceptable matching throughout all loading conditions. Tuning can be accomplished by adjusting the trimpot while monitoring the output voltage ripple on an oscilloscope (matching the height of 100Hz ripples). Tuning can also be done by ear since the audible hum/buzz from the power transformers becomes significantly quieter when the current through the transformers is devoid of a DC component.

Gallery

Characterisation

The assembled powersupply was subjected to the same tests as performed in simulation, to characterise the voltage and current regulation performance.

Real world measurements show a large discrepency between simulated and real world performance. This is best seen by plotting simultated and real world results to visually compare data for current regulation:

We can see that as the output voltage increases, output current reduces far more than the simulation predicts, the difference being particularly severe at lower current set points.

I suspected this was mostly due to the 393A thyratron tubes being used and no longer adhering to the nominal grid control region shown in the datasheet. Assuming the tubes are tired, we would expect somewhat higher critical grid voltages to ionise the tubes at a given anode-cathode voltage. Due to this, the 6N2P triode(s) driving the grid will be driven further towards cutoff (Ia-->0) to achieve the higher grid voltage, which in turn results in a more negative triode grid voltage and causes greater effective input offset voltage of the differential pair. This in turn causes the set current to fall. The same effect is present for the voltage control differential pair, however, the effect on voltage regulation is significantly less severe since a small input offset voltage causes a negligible difference (% change) in the set voltage when compared to the effect on set current. It is somewhat counter-intuitive to think that the thyratron grid voltage needs to be driven relatively high (closer to 0V) even for low output currents (e.g. 75mA), however, consider that at higher output voltages the thyratrons anode-cathode voltage becomes relatively small regardless of load current thus the grid voltage tends towards 0v. With this in mind, the effect of output voltage on regulation set points is present at both low and high current/power loads.

My plan of attack was first to update the simulation to more accurately model the real life behaviour, which would allow me to devise modifications to improve the real world performance. I first updated the 393A thyratron SPICE model to behave more like a 'tired' tube. The original SPICE model (blue line) approximates the grid control region in the datasheet by two linear segments. The updated model follows the highest grid control voltage characteristic shown in the datasheet (red line) instead of the nominal characteristic. Note that the thyratrons are never subjected to anode-cathode voltages above +500V in the design so the SPICE model needn't be accurate there.

While this change caused more droop of the set current in simulation, the droop was still significantly less than the manufactured supply. I suspected that the 6N2P SPICE model did not accurately model the real 6N2P behaviour near cutoff. Since most triode spice models are created by curve fitting the datasheet information, and many datasheets often do not plot all the way to complete cutoff, the model may only be accurate in the linear region and not near cutoff. I changed the 6N2P model to a different (12AX7) model, which made the simulation behave more similarly to the manufactured supply:

While still having a discrepency, I felt I was at least barking up the right tree and could move forward with devising modification(s) to improve performance since I was confident that the simultation was now strongly exhibiting the same effects that were hindering real world performance.

Performance Mod

If we could add a few volts DC offset to the thyratron grids such that the control triode still remains in its linear region when the grid voltage reaches 0V, we could improve regulation. In the initial design I considered adding a dedicated positive supply a handful of volts above the output rail which the control circuits pull-down from, however I dismissed the idea since it would add yet another power transformer to the design which I wanted to avoid.

In hindsight we can leverage the existing grid compensation transformer, rectifying the AC to create a smoothed supply of a few volts DC, positioned between the control and AC compensation circuits:

A 6AL5 dual diode tube (or soviet equiv. 6H2P/6Х2П) can be used as a full wave rectifier into a 10u smoothing cap, producing about 3.3V DC. A 100K resistor maintains DC grid control in the case that the diode tube fails.

Simulations predict a dramatic improvement to performance with this modification implemented:

The filament of the added 6H2P diode tube is powered from the 6.3V AC supply shared with the 6N2P quasi-cascode control tubes. Theoretically this slightly overloads the transformer secondary winding, so I hand selected tubes which had lower than nominal filament current draw. The modification is implemented on a small perf board mounted in the chassis:

Re-Characterisation

Constant current characterisation was re-run on the power supply after the modification was installed.

Regulation is improved dramatically from the initial performance and closely follows the simulation. What is not shown here is that the maximum power output has also substantially increased. The supply is now able to provide 600mA @ 340V (204W) whereas before the modification the maximum output was 600mA @ 250V (150W) before the supply dropped out of regulation.