A TWT POWER SUPPLY


by Marko Cebokli S56UUU
'96 EME conference

ABSTRACT: In this paper I describe the TWT power supply that I have designed and built for a TWT in EME service. The description is preceeded by a discussion of TWT power requrements, a review of existing solutions and some philosophy on which the design is based.

1. SOME HISTORICAL NOTES

I have finished this supply about two years ago. Many HAMs have expressed interest in it, so I have sent them my original manuscript diagrams, that describe the actually built version. But in the meantime, I have operated this supply quite a lot, mostly outdoors in adverse weather conditions (moisture), and gathered some experience and found some weak points. That's why I have decided to make some changes to the original circuit. The diagrams published here include these changes. Because of the pressing deadline, I couldn't yet try all of them in praxis, so some caution is in place with these diagrams. I hope to get everything tested before the conference. There are also some 'forgotten' components in the diagrams that (probably) aren't needed any more.
I also hope to get the line interface and timing&control PCB's ready till then.

Things that were changed are:
This is quite a long paper, becuse I believe that most HAMs, like me, dont have much prior experience as switch mode power supply designers.
It took me a year of study and thinking and six months of experimenting to arrive to this (not yet final) solution, and I try to put down as much of the knowledge and experience gained as possible, and to explain the trade offs that I made, and to describe possible alternatives....

2) REQUIREMENTS FOR A TWT POWER SUPPLY

The TWT requires the following voltages for operation:
  1. Heater supply. Mostly 6.3 V, up to 2 amperes. Manufacturers set very tight tolerances on this voltage, down to +-3%, to guarantee full life span. WA7CJO has been able to revive some old tubes by changing the heater voltage, so it is good to make this voltage stabilised and adjustable. When the wires between the tube and it's PSU are long, one must consider the voltage drop along them! One of the terminals is usually common with the cathode. In case of DC heater supply, manufacturers often specify the polarity, usually the positive going to the common f/k terminal. The heater must be switched on one to four minutes before electrons are drawn from the cathode (the beam is switched on). For tubes that haven't worked for some time, it's often possible to restore the cathode somehow by letting the heater on without cathode current for a few hours. However, some manufacturers, for some tubes, limit this standby time. For example SIEMENS sets this limit for most tubes to 1000 hours. (more than a month)
  2. Collector supply. This one takes up most of the beam current, and supplies most of the power. Typical values are abuot half the helix voltage, usually in the 800 to 3000 V range, 30 to 200 mA. The exact voltage is not very critical, the lower limit is set by the point where electrons return to the helix, the upper by the flashover absolute maximum. To get high efficiency, it should be as close to the lower limit as practical, since the collector dissipation is directly proportional to collector voltage. Hum and noise will do no harm, provided that the voltage stays above lower limit. The same goes for two-stage collector high efficiency types. Usually the collector voltage is not stabilised.
  3. Helix supply. This is the voltage to which the electrons in the beam are accelerated. Typical values are 1500 - 7000 V. Since the gain and phase of the tube are highly dependent on the beam velocity, any hum or noise on this voltage will cause parasitic modulation. It is mostly phase modulation, so you won't see it with a simple detector. The sidebands due to the switching frequency ripple of a SMPS can be detected by a SSB receiver, the hum modulation by a FM receiver. A top-notch spectrum analyser than can show 100 Hz sidebands at 10GHz can also be used. Another alternative is a SSB station and a software FFT analyzer. It is important to check at least the 20kHz sidebands, since they can be of equal power than the carrier! On 10GHz you probably won't cause QRM to anybody, but it's a big waste of power! If the tube is giving you 30W in form of three equal 20kHz spaced spectral lines, a thermal power meter (like HP 43x) will show 30W output power, but your corespondent with a 2.5kHz SSB receiver will hear you only as a 10W station. The parallel combination of TWTs for higher power requires still higher phase stability, so additional care must be paid to the helix voltage. It would be possible to run two tubes on one supply, provided that the helix protection and G2 circuits are separate. Ideally, the beam should go past the helix to the collector, and the current on this supply should be zero. Imperfect beam focusing causes some current in the helix. This current must be strictly limited to the maximum permitted value, typically 2 to 10 mA. In case of overcurrent, the protective circuit should instantly switch off the helix supply. However, the protection must be slow enough to 'ignore' the switch on transient. The helix current is a good indicator of the 'unhapiness' of the tube, be it wrong supply voltages, too much drive, poorly matched load etc. So a helix current indicator is a must.

    Since the phase velocity of the waves on the slow wave structure (the helix) is frequency dependent (dispersion), the optimum helix voltage will depend on frequency. Usually the waves get slower with increasing frequency, so a higher frequency will require a lower helix voltage. For example, for SIEMENS RW81, optimum helix voltage is 2.96 kV at 5.8 GHz and 2.84 kV at 8.5 GHz. Especially when using a TWT outside it's 'official' frequency range, this should be considered.

    The helix voltage must be switched on after the collector voltage, and switched off before it. With the helix voltage alone, the electrons would rush to the helix and destroy it. Therefore, a good TWT power supply must have an interlock that allows helix voltage only if collector voltage is also present.
  4. Control grid (sometimes called G2 or anode) supply. This voltage determines the beam current, and is usually derived from the helix voltage. It must be adjustable. Current consumption is in the micro ampere range. In most cases, this grid is also used for switching the beam on and off. Because this voltage has a role in beam focusing, it must be switched on and off very fast. While it is rising, the helix current can assume values several times the allowed absolute maximum, and this transient must be kept as short as possible. G2 voltage should not be present in absence of helix voltage, because excessive current would flow and destroy G2. Since in most cases G2 voltage is derived from the helix voltage, this is no problem.
  5. Other grids - I have seen data for tubes with up to four grids. Currents on those grids are mostly low enough, so that resistive dividers from the helix voltage can be used. Most tubes have only one additional grid, 'G1', connected directly to the cathode. Maybe it would be possible to use this grid to switch the beam on and off, to avoid high voltage switching on G2. I haven't tried that yet, because it requires another negative supply.
The usual switch-on sequence for these voltages is the following:


Some manufacturers also specify the rise times (Typically < 20 ms). Most of them allow simultaneous application of the last three voltages, provided that the rise times are met and that the ratios between the voltages are as required during the rise time. However, most important is that the heater is switched on a few minutes in advance and that the helix voltage is never applied before the collector voltage or removed after it. The collector voltage can be applied simultaneously with the heater, since the beam current won't flow till the helix and control voltages are applied.
An unusual solution is used in [3]. Two converters are used, one for heater and control grid, the other for collector and helix. The latter is switched on after delay. The helix overcurrent protection disconnects only the collector/helix converter.

3) PHILOSOPHY

Studying the literature, I came across various configurations of TWT power supplies:
  1. Single 50 Hz transformer with several secondary windings, rectifiers and some form of voltage stabilisation. [1] uses shunt zeners (LOTS of them), [2] uses a real voltage regulator, made with tubes. Advantages: -simplicity (at least [1]) -no noise generated Drawbacks: -energy stored in smoothing capacitors can destroy the tube -high voltage switching required -big and heavy
  2. Two separate switching converters for collector/helix with linear (dissipative) pre-regulators [3] Advantages: -relatively simple -no high voltage switching -not enough stored energy to kill the tube Drawbacks: -lots of heat generated, especially with a bigger tube
  3. Two separate switching converters with separate PWM pre-regulators [4]. Advantages: -could be maybe made without HV switching -not enough stored energy to kill the tube Drawbacks: -complicated

All of the above mentioned designs include some form of helix overcurrent protection and switch-on delay. As an option, [3] has an linear post-regulator for the helix, with a series pass transistor on the high voltage side.

I decided for two PWM regulated converters, one for helix and G2, the other for collector and heater. The heater has its own post-regulator, to allow independent collector voltage adjustment. I wanted to avoid high voltage switching, but for now I couldn't, because the helix current transient was too big. (The rise time of the helix supply was to slow.)

The helix converter has feedback regulation, to ensure accurate voltage. The collector converter is unregulated, because the collector voltage is not very critical, and because it is floating at high negative potential, so the feedback loop would have to go over an optoisolator or use a separate sense winding on the power transformer with a dedicated rectifier etc.

Maybe it would be better to use a (linear) pre-regulator and a 100% (well, 95%) fixed duty cycle converter for the helix as in [3], to reduce the switching frequency ripple on the helix voltage. It is really hard to smooth out a 10 kV square wave, as you can find out with almost any circuit simulation software on your PC. I really don't like much the 100nF capacitor on the helix, a 6 or more kV type is hard to get, and at 6500V it stores more than 4 Joules of energy (like a 400g hammer falling on your feet from 1m high.) Any ripple on this voltage causes severe parasitic modulation, so it's important to keep it "clean". A linear post-regulator as in [3] is probably the best solution, but I didn't find a way to combine it with a truly safe overcurrent protection.

I decided to design this supply for 220 v AC input for the following reasons: Even with lower power TWT's (10 - 20 W) the total consumption isn't likely to be under 100 W. A battery that can supply 100 W for a duration of a typical contest is bulkier and heavier than a small gasoline generator. A small backpacking hilltop expedition also probably won't include TWT's. On the other side, a bigger car-based one will almost certainly have an AC generator. For those who would still prefer DC input, I'll give some hints for modi- fication later on, but I haven't built such a version and can't tell you how it performs.

I tried to design the circuit using only components that a typical ham can easily find in his junk box or in aunt Lisa's deceased TV set. That's why I preferred "evergreen" IC's to fancy specialised types. Those are a little like woman's fashion: every spring there's a new collection and they are either not yet or no more available. There is also not much to be gained in terms of size, which is mainly determined by input smoothing capacitors, transformers and storage coils, and one would still need the additional IC's for the various control, timing, protection and interlock circuits. Also, this way the supply is very transparent: you know exactly what and why each part does, and if it doesn't work, there are plenty of points you can measure. On the other side, if you use a fancy chip, and 'he' chooses not to work....
Also, for the transformers and coils I used only such ferrite cores that can be salvaged from old TV's and PC's. The TV line transformer core has another advantage over pot cores and their derivatives: lots of winding space that can be used for efficient isolation. Anyhow, when building a high voltage device, you should resist the "Japanese syndrome" that makes people make things as small as possible (or even smaller). You can naturally use any core that is big enough and made of 'power ferrite' like SIEMENS N27 or similar. With TV line transformers you can be sure the material is right, and it is possible to use original HV windings from TV's for the helix supply. On the other side, using BIG potcores with a large magnetic cross section would mean less turns for a same voltage winding and therefore less parasitic capacitance.

In the building of this supply there are only two critical areas: the power transformers and the high voltage isolation. With the transformers, I found the problem of parasitic capacitance to be quite severe - above all it was difficult to set the protection circuits for the power FETs so, that they didn't trip on capacitive spikes, but still protected the FET's. During this I sometimes tended to think that the art of switching supplies has more to do with pyrotechnics than electrotechnics, HI.

4) TRANSFORMER DESIGN

Manufacturers of ferrites often give you quite a lot of diagrams, curves and formulas for the design of SMPS power transformers. These are quite useful if you are forced to squeeze the last watt from a given core, for example in aerospace design, or if you are designing a million-pieces consumer item, where a one number smaller core can mean quite a lot of money saved. However, it's not a good idea to miniaturise a high voltage device, so I'm proposing a "KISS" method of transformer design here. We start with the following data: For the first three, I have chosen 0.3T, 10kHz and 3A/mm2. These values can be regarded as quasi independent of the particular supply design.

0.3T seems quite high, but I have found that even the oldest TV cores from the 50's don't heat more than 20 degrees above ambient. More recent cores stay almost totally cool, with only a slight non linearity noticeable in the magnetising current. Besides, in praxis the duty cycle is about 60-70%, reaching 95% only when line voltage is low, so that 0.3T is actually never reached.

10 kHz is low by SMPS standards, but with the long windings the stray capacitances, that have to be charged each cycle, are quite high. On the other side, a higher frequency means shorter windings and less capacitance. I don't know where the real optimum lies, so I followed [4] that suggests a low switching frequency.
Anyway, there is a trimpot in the circuit for the frequency, so you can turn it anytime to have 15kHz and 0.2T, or whatever you want. (BTW, the 10kHz whine is also a welcome warning signal that the circuit is operating and you should keep your fingers away, HI.)

3A/mm2 is quite conservative for transformers of this size. For frequencies in the range 10-20kHz, wires thicker than about 0.65mm should not be used, because of the skin effect. If there is more than 1A of current, several insulated wires in parallel should be used.

In a symmetric layout like this, the flux density changes from -0.3T to +0.3T in one half period, and back to -0.3T in the second half period. The flux den- sity time derivative (rate of change) is also 0.6T per half period, for 10kHz this is 0.6T / 50E-6 s = 12000 T/sec. The U57 core I have used has a magnetic cross section of 171 mm2 = 1.71E-4 m2. The total flux change is also: 12000 * 1.71E-4 = 2.05 V. This means, with this core and 10kHz we get two volts per turn.

Next, determine the wire sizes and numbers of turns for all of the windings. To leave some room for regulation and losses, the secondary windings are designed for 1.5 times the required voltage. (The heater winding for 250-300% to accomodate collector voltage adjustment.)

To minimise stray inductances, the transformer can be wound in a 'sandwich' configuration, that means the secondary between two series connected layers of the primary. This helps to keep the transformer 'stiff', but brings some problems with isolation and parasitic capacitances.

The current in the primary winding can be estimated from the output power. Add at least 20% to that. Due to spikes caused by parasitic capacitances, the effective primary current of the collector transformer in my prototipe is almost twice the value required by the output power. Fig 5. (8kB GIF)

The isolation is the hardest part here. DF7ZW suggests in [4] the use of plastic spray to impregnate the windings. I have found that this spray consists of 90% or more of solvent that evaporates away. So even if you spray the windings soaking wet, it is imposible to avoid bubbles inside. It seems that vacuum impregnation is the only 100% solution, but if you don't have acces to that, do use that spray or any fluid isolating stuff. I have repeatedly soaked the finished transformer, counting on the capillary effect to get the stuff in. I used epoxy to fill the big holes, and plumber's teflon tape to bandage the core. A special thing to cast high voltage devices in is the Wackersilicone SilGel 600A from Wacker chemie Muenchen, if they still produce it (I have a very old bottle). It is a two component thing that cures to a transparent, very elastic rubber. I don't know if the silicone used for window glass sealing would be useful, but they have warned me that two variants exist, one of them containing an acid, and therefore very unsuitable.
On the other side, too much of the stuff tends to increase the parasitic capacitances.
The circuit of Fig 6. (5kB GIF) can be used to estimate the parasitic capacitance of a single winding, or a complete transformer. Read the width of the capavitive spikes at 1/e (approx 1/3) of their height. This time is the RC product of the series resistor (1K, for example) and the parasitic capacitance. The inductance can also be estimated from the slope of the longer part of the curve, since dU/dT = R*dI/dT = R*U/L, U=12V.

5) CIRCUIT DESCRIPTION

The mains interface provides the 300, 150 and isolated 12V DC voltages with soft turn-on. The timing, control and protection block generates the proper switch on and off sequeces with required delays, monitors the critical currents and provides interlocks converters. The two converters are quite similar, incorporating PWM modulators and FET drivers. The high voltage part will differ somewhat for different tube types, and because of the high voltages, large isolating spaces are needed. That's why I don't think a PCB is the best solution for this part of the circuit, so I havent designed one.

5.1. Mains interface.

Schematic is shown in Fig 1. (83kB PDF) The LF1 is a standard mains line filter, to reduce interference. Be sure it is rated at least 5A, because the current flows in hefty spikes. A cure for that would be a heavy iron choke between D1..4 and C4.
C3, R2..7, D5..7, Q1, SCR1 and RE1 are the soft switch on delay circuit. I have used a simpler solution with fixed 290V threshold, but it didn't work with low line voltages usually found at radio shacks. 750 ohm for R2 is probably too much, the charging takes a minute or two, so 75 ohm seems better to me. The energy dissipated on R2 per charging cycle is independent of it's value anyway.
R8,9 and C5,6 provide the 150V midpoint needed in a half-bridge converter design. The 12V for the control circuitry comes from T1 etc. A blower can be installed, but it is a mixed blessing - when operating outdoors in dense fog, it actually sucks the moisture in. So if you really dont have an heat problem, it's better to leave it avay.
For 110V power, the standard trick, used in most PC power supplies, is to use a voltage doubler. Use two series capacitors instead of C4 and two diodes. This automatically gives you the needed midpoint, so R8,9 and C5,6 are no longer necesary. The switch on delay circuit will need some changes in this case.

5.2. Timing, control and protection

Schematic is shown in Fig 2. (162kB PDF) After line power on, when the line interface delay switches on the 12V supply, and it reaches about 10..11V, the PWR OK output of the collector converter goes high. This enables the converter and after approx 2us (R2,C2) removes the reset condition from the OFF input. The same signal goes over a 0.5 ms delay (R1,C1) and clocks the collector converter ON.
Qnot goes low and removes reset from U2, which gives a high on pin 8 after approx 3 minutes, and enables the helix conveter. The TWT is now in standby, with heater and collector voltages on, but helix and G2 are off.

The TWT can now be switched to the 'RF ON' condition by S1. When it is switched on, reset condition is removed from the helix converter OFF input after 1ms (R8,C4), and the converter is clocked on after 5 ms (R7,C3). Q goes high and after 0.2 s (R12,C5) switches on RE2, the G2 relay, provided that the external TX GATE signal is present.
TWT is now under full power, and can amplify RF.

When S1 is switched off, it will first switch off RE2, after 1ms (R11, C5). Next, after 0.2 s (R9,C4), it will switch off the helix converter and return the TWT to the standby state.

The link over D10 is the collector interlock. If the collector converter goes off, it will instantly switch off the helix converter. It also prevents accidental switcing on of the helix converter, while collector is off. Helix voltage in absence of collector voltage is lethal for a TWT!
In absence of this circuit, helix secondary current protection would still do the same trick, but I opted for double safety, since it only cost me a resistor and a diode.

Another interlock goes over Q3 and D3, so that the collector converter is also switched off in case of helix secondary overcurrent (SEC FAULT). That is not strictly necesary, but I opted for the safe side. The price for increased safety is, that you must wait the tube heating delay each time the helix protection trips. But this also has the positive effect that one has time to THINK about why helix protection reacted, since nervous poking on the helix ON button is futile. If you dont like this, you can switch this interlock off by removing the jumper.

U2 sets the tube heating delay, after which the helix converter can be switched on by front panel S1. The amount of delay is approx one minute for each 10n of C3.

RE2 is a vacuum relay that switches on the electron beam by switching the voltage on G2. It is controlled by S1 on the front panel, and gated with an external signal (TXG). This signal should originate from a contact on the T/R coax or waveguide relay. It should be +12V when the relay is in TX position. Its purpose is not only to protect the tube against running into a reflection, but also to remove TWT noise during reception (This noise was the cause of failure of the first 10GHz EME attempts)
For RE2 I have used a russian vacuum relay, that I got at the flea market in Weinheim. They show up quite regularly.

5.3. The collector converter

Schematic is shown in Fig 3. (256kB PDF) The same general circuit is used for the collector and helix converter, with some minor modifications. The converters are also made on the same type PCB, with different components omitted in each one.

The circuit diagram is drawn for an 'universal' converter, and deviations from it are described in the text.

The diagrams are drawn for a full bridge converter, but for less than 500W of DC, half bridge is all that's needed. Just delete Q13,Q14,R45,R46,R47 and R48, and connect the other end of the power transformer's primary to MID at the line interface. You also need only two secondaries on the driving transformer (T1) in this case.

The secondary overcurrent protection is not used in the collector supply, so D18, R42, SCR1, R43, D16 and R44 are not used in it.

The circuit around Q2 and Q3 is a 'power OK' detector with histeresis for the 12V supply.

U1A, U1B and U2A provide the sawtooth, deadtime and phasing signals for the converters. P1 sets the frequency and R6 the DC pedestal.

The circuit around Q5 and Q4 softly starts the converter, to prevent core saturation and too big smoothing capacitor charging currents.

U1D is a comparator for the PWM regulation.

U1C is the regulation amplifier. Since the collector converter operates unregulated, voltage sense inputs aren't used, R14 is 1K, JP2 is ON and R11 is zero.

PWM is done on the falling slope of the sawtooth, the rising slope being the guaranted dead time, when both output transistors are OFF.

The driving transistors are enabled by the Q signal, that comes from U2B.

Maybe it would be possible to get a cleaner drive using BS170/BS250 pairs in a kind of CMOS configuration, but I havent tried that and dont know what kind of current spikes that might cause.

The driving transformer T1 is wound on a SIEMENS RM-10 core of N41 material, with zero airgap. Any core of similar size and material could be used. It is possible to grind the eventual airgap away using carborundum powder and a piece of flat glass. It is wound in a 'sandwich' way, the secondaries between the two layers of the primary, to minimize stray inductances that cause ringing. The isolation between the secondaries must withstand at least 600V, and the one between the primary and the secondaries must satisfy the safety standards for mains isolation.
Because there is enough space for the windings, it is not necesary to drive the core to the maximum flux density. To minimize magnetising currents, in addition to zero airgap, it's better to use a low flux density. I have used 40 turns for each winding, with the primary divided into two layers (top and bottom) of 20 turns each. Wire is 0.3 mm Cul. A sensible alternative would be to use a complete drive transformer from a PC power supply. (WARNING - it could get saturated at 10 kHz!)

If one needs a full bridge converter for higher power, it is necesary to wind four secondaries.

If the power comes from low voltage batteries, a push-pull topology for the converter Fig 7. (5kB GIF) is best, and T1 can be omitted. Use a 4082 non invertig and gate instead of the 4012, then Q7..10 can directly drive the gates, the primary and secondary grounds connected in this case naturally. T2 in the current protection could alo be eliminated in this case, using just a series resistor connected directly to base and emitter of Q6, but that would mean 0.7V voltage drop at maximum current. At 12V power that is 6% of efficiency thrown away, so it is better to leave in T2 etc.

NOTE!!! if 12V power is used, a 12V/12V stabilised DC/DC converter is required for the supply of control electronic, since this voltage is used as a reference for the helix voltage regulation (or else, change the circuits around U1C).

If you use a push-pull topology with 110V mains power, you must keep T1 because of mains isolation safety!!!

Use a scope to check the waveforms on the gates of the output transistors, before applying voltage to their drains. It should be something like Fig 8. (5kB GIF) The FETs must be connected during this measurement, since their gate-source capacitance has a big influence on this waveform.
Check that the spikes at Tx are low enough not to open the FETs. If necesary, change values of R18..21 (and R45..48 if used).

Before connecting the power transformer, it is recomendable to test the switching behaviour of the FETs in the circuit of Fig 9. (5kB GIF) This test cannot be done with the transistors in the PCB, because the serial connection of Q11 and Q12 (and Q13/14 if used) must be broken. Its main purpose is to check that both transistors that are connected in series do not conduct simultaneously. Use a scope to monitor drain voltage: it should be a clean square wave with less than 0.5 microsecond rise and fall, and absolutely no dip at Tx. Note that there is 300V on C4 and the lamps are only 220V, so dont set the pulse width (P2) to more than 70%!

WARNING!!! WARNING!!! most osciloscopes are grounded thru the safety plug. R62 is connected to the mains through the D1..D4 bridge! Connecting the ground terminal of the scope to either side of C4 will cause a short! Inverting of the plugs doesn't help, because of rectification.
The only solution is an isolating transformer. If the isolation transformer has large stray inductance, C4 won't reach full voltage under load, because current flows in short pulses. So one has to put the scope on the transformer. In that case the whole case of the scope is at elevated potential!

Q11 and Q12 (and Q13 and Q14 if used) need some cooling, 5K/W should be enough. In my case, almost half of the dissipation is caused by the capacitive current spikes!

It is good practice to use stronger transistors as theoretically necesary. Even if 4A types, according to their manufacturer's data, would easily take the 2A working current plus 8A / 4 microsecond spikes in my prototype, I used 8A types. They also have a lower Rds on, and hence run cooler.

T2 is the current monitoring transformer. It is wound on a 20 X 8 mm ferrite toroid. The primary is just a wire passing thru the hole, and the secondary has 30 turns, equally spaced around the ring. A good source for the core or even the complete transformer is again an old PC power supply.
R41 sets the average current at which the protection will react. C4 sets a delay to come over the current spikes due to the parasitic capacitances. in the collector converter it has to be about 47n. Maximum safe value is about 150 nF, but that's really tickling the lion.
You can test the protection circuit by discharging a 10 microf. 16V capacitor through the primary of T2. (power transformer disconected, of course.) U2B shold go off (see LED D12 on the control circuit) regardles of the direction of the current.
It is necesarry to set the values of R41 and C4 to suit any particular case. Start with R41=22 ohm and C4=1nF. Leave the secondaries of the power transformer unconected. If protection trips, decrease R41 or/and increase C4. Don't go under 3.3 ohms with R41. (that's quite a dangerous value already.) After you get the converter running with unloaded secondary, connect the high voltage rectifier diodes (D1..D12 on the high voltage circuit) (Naturally, switch off the converter while connecting them, HI). If necesary, change R41 and C4 again. Repeat the procedure with the high voltage smoothing chokes, ETC.

I had to put R1..R4 on the high voltage circut in to get it running. If you can do it without them, just leave them out.

You have finished, when the converter runs with full load, for all pulse widths, and the protection safely switches it off when you make any kind of a short on the secondary side of the poer transformer, before or after rectification. Use a screwdriver and enjoy. This is a necesary test! There is a slight chance of a big bang, but you have to take it to be sure. Use the largest R41 and smallest C4 that gives you reliable operation.

To test this converter under full power, you need a 21K 300W resistor. You can use 30 680 ohm 10W resistors or something like that. I have used 11-15 220V 25W lightbulbs in series. Because lamps when cold have only a tenth of their nominal resistivity, the converter won't start at full voltage. The solution is to set zero output with P2 and then slowly increase it, or to substitute a big capacitor for C3 to have an extra slow start. A powerful triode would be a very nice adjustable load for testing, but you need extra supplies for the heater and grid.

If the protection keeps tripping, it is very unlikely that the protection circuit is defect! Don't get teased too soon to try bypassing it! I have done it twice, and each time the protection was OK, but a real short caused it to trip. Both times I was rewarded with fireworks!

In the case of the big bang (there is quite some energy in C4 on the line interface!), R18..21 and Q7..Q10 usualy get fried also. Pieces of Q11...Q14 tend to fly around, so be careful with your eyes.

One of the diodes D1..D12 on the high voltage circuit blown can be the cause for protection tripping. R28,29 on the high voltage circuit prevent static buildup on the core.

R49 is used only to monitor the current through the power transformer with a scope. It should be of low inductance, like 5 1 ohm 0.5W resistors in paralell. The waveform is drawn approximately in Fig 5. (8kB GIF) When the converter is operated without loading, then the amplitude of the capacitive spike is dependent on the pulse width. It is proportional to the voltage change on the primary of power transformer in the moment that the transistor is switched on. After the other transistor has switched off, there is a damped oscillation on the power transformer (the frequency of which is determined by the inductance and parasitic capacitances of the transformer), and the voltage change is dependent on where on this curve the switch on moment occurs. Under load, this variation is smaller. If you have a half bridge converter, it is better to put R49 on the other side of the primary, that is connected to 'MID', because there is less AC.

WARNING WARNING!!! You can cause a short with a grounded scope!!
Read the warning about isolation transformers above!!

The power transformer is the most critical component in the supply. It is wound on a U57 core from an old TV. The windings are on both legs. The primary consists of two paralell windings, one on each of the legs. Both have 72 turns of 0.65 mm wire, and are further divided in two layers, that comprise the top and bottom layers on each leg. The top and bottom winding on each leg are connected in series, then the series combinations of windings on each leg are connected in paralell.
The secondary windings are sandwiched between them. There is one secondary for the heater and three for the collector voltage. One of the high voltage windings is divided in half between the two legs. The heater winding has 10 turns of 0.65mm wire. The high voltage windings have 540 turns of 0.25 mm wire each. It would be better to have four windings of 405 turns, since 1100 V is a bit too much for the 1500V diodes used, and they tend to blow.

I found the trick with separate windings and rectifier bridges in [3]. The good side of it is that additional RC elements aren't needed to distribute the voltage evenly over the series connected diodes, and one can easily switch to a lower voltage by removing one winding. It is also handy for high efficiency tubes with multistage collectors. The drawback is that the capacitances between the separate windings must be charged to high voltages on each cycle, and this increases capacitive current spikes. The only way around this would be to wind the adjacent windings in oposite directions.

5.4. Helix converter

Helix converter is built on a identical PCB than the collector converter, with some minor changes.

The power OK circuit around Q1..Q3 is not needed and is left away.

The secondary current protection around SCR1 that wasn't present on the collector converter is needed here. SCR1 also acts as a fault memory, so that we can distinguish between primary and secondary protection faults.

Here a regulation feedback is used to set the pulse width. We are regulating a negative voltage, so R13 is omitted and JP2 is off. Information about the value of the helix voltage comes over a voltage divider to '- volt sense' and to the input of U1C that compares it with the value set by P2. I have set the value of R14 so, that the output voltage stays within <1% for 0 to 6mA of helix current. R14 sets the static gain of the loop, higher value means better regulation. The loop wasn't stable however, so I added C2. A resistor in series with it would probably improve the dynamic behaviour and 100 Hz ripple rejection. As said before, this part is not very good. One has to consider the high voltage smoothing L and C when designing around U1C. Actually, when I determined the above mentioned values for R14 and C2, I have only had 6n of smoothing C. Since it is now 100n, C2 could probably be significantly smaller.

Other circuitry is almost identical to the collector converter, so refer to 5.3. for the description.

The current protection (R41, C4) must be set separately, since the current values are different from the collector converter.

The primary of the power transformer is 72 turns of 0.65mm Cul wire. It is possible to use original HV windings from TV's for the secondary of T7, but you have to search for the right ones. Most of them are too short. I had to use very old ones, that were used without voltage multipliers. There is a very fast flux change during line flyback in a TV, so they get much more volts per turn as one would expect from the 15 kHz horizontal frequency. I had to use the longest two I could find in series. Modern ones used with voltage multipliers or physically integrated with the rest of the windings are not useful. If you find anything suitable, just wind the primary on the top. The inner end of a TV HV winding is usually not isolated enough, so use epoxy or something similar. The good property of these TV windings is their low parasitic capacitance.
If you don't find any suitable winding, calculate the required turns no. and wire size as explained above in part 3. For example, for 6500V about 5000 turns.
When you get it running, turn the lights off and look for corona glow and listen for the associated hissing. If you see it, use epoxy or something similar to fill the space where it occurs. DF7ZW suggests using an AM receiver to listen for discharges. I haven't tried that. A medical stetoscope with bare tube end is also useful for locating a sound source.
This converter must be loaded with the regulation and G2 voltage dividers (about 2mA) when running, otherwise the flyback spikes will charge C9 (on the high voltage circuit) well above 10kV and punch a hole in it. If you really want to run it unloaded (for some debugging purposes at the current protection for example), remove C27!!

5.5. High voltage section.

Schematic is shown in Fig 4. (83kB PDF) The collector voltage comes from three secondary windings. It could be a single winding, but this way it is easy to adapt the supply for lower voltages. Each winding has a separate diode bridge. A single bridge with series diodes would require additional R's and C's to distribute the voltage between the diodes.
The BY228 diodes are 1500V 4A types normally used in TV horizontal deflection. The manufacturers data about their Trr is somewhat cryptic, but they are fast enough for this application, as long as you provide a few microseconds of dead time. To get maximum reliability, the rectified AC voltage should not be more than 800V, but they work quite well up to 1000V. As I said before, it would be better to use four secondaries and four bridges for a 2400V supply. The peak of the square wave voltage is simply DC output voltage divided by the duty cycle. Since the secondaries are wound about 50% longer to operate the converter at about 60-70% duty cycle, the peak voltage is about 3600V for 2400V output.

I had to put R1..4 in to keep the capacitive current spikes low enough so that the primary overcurrent protection could be set fast enough to still protect the output transistors in the case of a real short. This is a very bad and ugly solution, producing heat. It would be very desirable to improve T1 and smoothing choke L1 and get those resistors out. They burn quite some watts and make a undesirable hot spot. Try to get things working without them!!

The smoothing choke for the collector is also wound on a U57 core, 2500 turns of 0.25mm wire with 1.3mm airgap (0.65mm each leg). When calculating the requred airgap, it is usually ok to neglect the core and consider only the airgap length.
For example, for I=.12A, N=2500 and B=0.3T the required airgap is: l = 4*PI*1E-7*N*I/B = 4*3.14*1E-7*2500*0.12/0.3 = 0.0013m = 1.3mm where 4*PI*1E-7 is the magnetic permeability of air (u0). The U57 core itself is equal to about 0.1mm of airgap, so the actual airgap can be 0.1mm less than the result of the above formula. Its magnetic cross section (area) is 171 mm2, so the inductance in H can be approximately calculated as folows: L = 4*PI*1E-7*1.71E-4*N^2/(l+0.0001) For l=0.0013 and N=2500, L = 4*3.14*1E-7*1.71E-4*2500^2/0.0014 = 0.96 H.
It is possible to put more turns on this core (with larger airgap) to increase inductance, but the main problem is again the parasitic capacitance. Again, you can estimate this capacitance using the setup from Fig 6. (5kB GIF) I have tried to reduce the capacitive spikes by placing another winding of about 100 loosely wound turns on the other leg of the core and connecting them in series. Naturally, it must also have adequate high voltage isolation. It's two ends must be well isolated mutually and towards ground.
Under absence of loading, the voltage will be over 4kV, but that's no problem, since all the tubes I have seen data for, have the collector maximum voltage rating about twice its operating value (usually equal to the helix voltage rating). The increased voltage in free run comes from flyback spikes. Theoretically, in a bridge circuit, the flyback energy should go through the power FET's inverse diodes. But in the praxis, not all stray inductances are coupled to the primary, and those inverse diodes aren't infinitely fast, so some flyback energy will unavoidably end in the secondary smoothing capacitors.
The resistors R5..R8 divide the voltage in half for the series combination of C21..C24. You can decrease their values to reduce the free run voltage, but that means heat production.

Collector current metering is done over an optoisolator, to bring the meter down to room potential. The optoisolator brings some nonlinearity, si the scale for I1 must be calibrated separately. D22 reduces the dead zone at low currents.

The supply for the heater is a simple rectifier and LM317 regulator. It must be well isolated from its surroundings, because it floats at -6500V. I have salvaged L2 from a PC supply, and tripled the number of turns, because PC supplies run at higher frequencies. D13..16 are fast 3A types, also from a PC PSU.
D17..20 in the helix rectifier are BY187 or similar 10kV types. I have found mine in an old TV voltage multiplier, that was filled with soft silicone rubber. I don't know if the selenium 'sticks' also found in TV would work (probably not).
Put the diodes in plastic tubing. The separation between traces on a PCB should be at least 20mm, otherwise a vertical barrier made of vitronite or similar material must be glued between them, using epoxy.
When operating in high moisture, the high voltage tends to 'crawl' over. L3 is made on a U57 core too, and consists of two HV windings from TV's. It should have at least a few H, with low capacitance. The isolation is important too, of course. If using TV HV windings, pay attention to the isolation of the inner terminal.
If you have to wind it, 5000 turns of 0.08mm wire with 0.1mm airgap is OK. C9 must be a 7kV type. I use a 5kV type and pray. If you use a series combination of lower voltage types, you need a separate resistor ladder to divide the voltage. That is, you cannot use R15 or R20..22, because that would disturb other important functions. Here you can use higher values, say 10M each.
R15 is 5M 10W, and it must be very stable with temperature. 1% metal film resistors are the only choice. First I used 50 pieces of 100K 0.25 W, later HB9JAW sent me 3W types, to reduce the number. Similar goes for the R20 (26 * 100K) and R21 (56 * 47K). These resistors get quite hot. Their values are chosen quite low for two reasons: first, they serve as free run loading for the converter, to make it regulable under zero load, and second, high valued resistors are quite unreliable and unprecise.
These resistors determine the helix and G2 voltages, on which the TWT is very sensitive, so they must be very stable. In the beginning, I have used ordinary 5% carbon resistors, and have had big trouble. To always run the tube at the optimum working point, the stability of these resistors is crucial.
R18/19 determine the value of the current at which the helix protection will react. When the helix current is zero, there is still about 2mA current through R18, that comes thru R15 and R20..22. That's also why the other end of the helix current meter is connectet to about 1.6V on the top of LED D19 (HON).

Use a resistor load (NOT THE TUBE!!) to set and test the helix protection. C10 delays the helix protection to come over the switch-on transient. Use the lowest value that allows you to start reliably.

There is no other way to determine the right value of C10 than to experiment live with the tube. Start without C10. If the protection throws out when you switch on the beam (S1 via RE2), add 1nF for C10. Try again. If no-go, double the value of C10, and so on. If you have a storage scope, watch the transient at 'helix current sense'. My tube goes to about three times the allowed value for a few miliseconds (that is less than 0.5 Joule energy). How much switch on transient on the helix is allowable is hard to say. The only data for transient helix loading that I have seen was for SIEMENS RW 85, a 22W tube, and amounts to 45J (same value for G2). This tube has maximum 4mA of helix current at 3500V, that is 14W dissipation. I think a safe value would be one Joule for each Watt of output power.

R97-R101 set the G2 voltage. When the beam is off, G2 is connected to the cathode thru RE2. R25 supresses a transient that otherwise trips the protection.

6. MECHANICAL CONSIDERATIONS

I have built my prototype on a plexiglass plate as the chassis, to get good isolation. This assembly then comes in a 19" rack case. Keep distances of at least 20mm between the parts carying high voltage.

7. OPERATING

SAFETY: 2500V 120mA is lethal. Although when you touch it, there will probably momentarily run a higher current that will trip the protection in microseconds, don't count on that.
Helix voltage, albeit higher, is less dangerous, because it has its current limiters set to lower values. However, C27 can punch quite a hole in your finger. After switch off, it will discharge within seconds thru R15 and R20..22.
Some measurements are necesary on parts that are directly connected to the mains, so watch out. Read the warning about isolation transformers above.

TUNE UP: Before connecting the tube:
Get converters running and overcurrent protections set as described above in 5.3. Set the helix secondary protection to the value allowed by your tube. Set the voltages to the values specified in your tube's data. Reduce voltage on the control grid (R23) for about 100V. The collector voltage must be set under nominal load, that is your tube's rated operating beam current.

Check the timings and interlocks:

Switch on mains. After some time RE1 in the line interface goes on, LED D11 lights. Now the collector converter will start and LED D12 goes on. After about 3 min (time set by C3 in timing&control) the 'ready' led D13 comes on, and the helix converter can now be switched on by S1 (timing&control). 0.2 seconds after helix converter switch on, if TX GATE is present, RE2 (G2) will go on.
If the collector converter is switched off by current protection, the helix converter must also instantaneously go off. In the case of helix overcurrent (simulate with resistors), both converters must switch off, if the jumper is installed on timing&control.

Switch the PSU off, connect the tube. Connect loads to input and output of the tube. (the output load must be of adequate power rating, of course, since the tube might oscillate even without input signal).
Switch on power, collector, wait. Check that TX GATE is off (LED D16). Switch on S1 (RF ON). Helix converter should go on (LED D14), but here should be no current through the tube, either collector or helix.

Use a wire link to 12V to simulate the 'TX gate' signal. This should switch on the beam via RE2.

Be careful at this stage!!

If the protection throws out, proceed with increasing C10 on the HV circuit as described above in 5.5.

Too low collector voltage can also be the reason for helix overcurrent, so it is good to start with collector voltage maybe 15% over normal value, decreasing it later, when the beam is stable.

Wrong voltage on G2 may also make problems.

When you get the beam running in the tube, the collector current should be somewhat less than prescribed, because G2 is set 100V too low. Increase now the current with R98, until you get the right value.

When the DC regime of the tube is ok, connect RF input signal through a variable attenuator. Start with less than 0.1 mW (-10dBm) drive. Slowly increase power, and monitor output power. If the helix current starts to increase toward allowed maximum value before you have reached rated output power, try changing beam current and helix voltage.

Tune these for maximum power (staying inside tube ratings, of course, HI). Bad output match can or problems with focusing magnets can also cause excessive helix current.

If you are using a tube outside its 'official' frequency range, and if the helix overcurrent problem persists for all helix voltages and beam currents, while you are increasing the drive, it could be that the output matching structure is not matched at your frequency. You can check this by connecting the output terminal of the cold tube (no DC power applied) to a network analyzer (a directional coupler with a detector will do also). If the return loss at your desired frequency is less than 10 dB, then this is very probably the problem. Try using an external tuning circuit to make at least a narrow 'hole' at your desired frequency. Try at least for 15db, 20 or more is best. I have had this problem, but was lucky because the tube had waveguide output with tuning screws.

LITERATURE:

[1] LA6LCA: Simple and reliable TWT power supply, DUBUS 1/88
[2] HP 498A, HP 491C, HP 493A, HP 495A Microwave Aplifiers  Manuals
[3] HUGHES model 1177 instrumentation TWT amplifiers, instruction and
    maintenance manual. HUGHES aircraft company, electron dynamics division.
[4] DF7ZW: Schaltnetzteil fuer Wandelfeldroehren, UKW berichte 4/92
[5] SIEMENS Wanderfeldroehren, data book 82/83


Up to S57UUU EME Page
Up to S57UUU TWT Page