High voltage polyphase buck converters?

[dmills]

Hi all,
I am working on a design for a slow(ish) drain supply modulator for an experimental ham radio system and find myself in need of a 30A supply at 12 - ~50V operating two quadrant with a modulation bandwidth of a few tens of KHz.

This is becoming an interesting challenge, particularly as the efficiency must be reasonably good....

My current thinking is a multiphase affair using those enhancement mode GaN things that EPC do, which will get me the efficiency at a sufficiently high switching speed to allow 500KHz or so per phase (Which is good if I want 10's of KHz of control bandwidth), but how to control the thing?

Nobody, it seems makes a high voltage buck controller that is easy to use in this way, all of them are either too smart for their own good or too low voltage, or don't do two quadrant well, and annoyingly there does not seem to be a generic PWM chip that will do what is wanted at the speed that is wanted and is syncronisable.

My current thinking is to build a master channel running as a voltage mode converter, plus a number of slaves that track the average inductor currents (with loop bandwidth somewhat smaller then the voltage mode loop so they don't interact too badly), but that quickly turns into a mess of discrete logic and opamps, it is enough to make a man consider measuring the inductor currents, feeding them to an ADC and hence an FPGA and doing all the heavy lifting there (But I hate that solution for all sorts of reasons)....

Ideas anyone?

Regards, Dan.

[T3sl4co1l]

I wouldn't think 99% efficiency would be an economical choice, when driving an RF amplifier at, whatever efficiency it is...

Multiphase is nice, though for 30A isn't really necessary.  Any other motivation?

You can run any number of single phase controllers (including the ancient TL494) in phase locked operation, with a suitable oscillator locking method of course.

Tim

[dmills]

Major motivation for multiphase is to get the voltage ripple at the drain of the RF amp down, much easier to filter a 2MHz switching output then a 100KHz one when you want a few tens of KHz of bandwidth.

Saturated drain efficiency in the RF power stage is close to 75%, and the whole point of the drain modulator is to keep it up there when the power is backed off 10dB or so (SSB spends most it its time at very much less then PEP), 99% I don't need, 95% @ 500KHz per phase would do, but notice that the range of output voltage is somewhat large which makes everything harder....

I want a hard switched CCM buck as these are to my way of thinking just simpler then the alternative approaches (PSFB with sync rectification for example, those tend to need very carefully designed transformers and such to work well).

Regards, Dan.

[Howardlong]

It's been a few years since I worked on modulation using exotic synthesis such as using LINC, Doherty, and other outphasing and pre-distortion techniques, but for a while it was my main job. You're right that going to digital correction brings its own complications, but you are lucky that computationally for the PA power you are talking about it is small potatoes: in my applications, the computational power became a significant concern of the design as we were constrained by power budget.

By far the biggest problem you face is maintaining linearity, minimising spectral growth, and modelling the PA and system sufficiently well to be able to correct sufficiently, especially bearing in mind that annoyances like temperature play a large part in that modelling. Depending on what you're trying to achieve, somethings can be relaxed more than others, but then that is what engineering is all about, a compromise.

Steve Cripps did a few books on the subject some years ago, as far as I am concerned he's the go to guru.

[dmills]

I already have the Cartesian loop working, two tone IMD3 ~-60dBc@1KW, which will do.
The nice thing about a Cartesian loop is that you can measure the phase error on the fly and correct for it dynamically so the thermal (And poor mosfet behaviour at low Vds) are less of an issue then in a predistortion based design.

FPGAs and high speed ADCs and DACs are **REALLY** nice for building HF radios, and a multiplier in a FPGA makes a mixer that has rather good linearity.

Doherty is tricky across a decade of tuning range, possible but the number of twisted wire couplers and relays would get old fast, so I am going for envelope tracking instead (Together with some dynamic bias tricks).

Regards, Dan.

[Howardlong]

Doherty is tricky across a decade of tuning range, possible but the number of twisted wire couplers and relays would get old fast, so I am going for envelope tracking instead (Together with some dynamic bias tricks).

If you want a decade of tuning range -and- very high efficiency, that's gonna be hard if you're looking at using a single HPA across that range. If you're prepared to look at several band specific HPAs the efficiency can be improved but of course there's a cost to that.

What are your specifications?

[dmills]

This is me playing, so pretty much whatever I can get.

IMD3 better then say -45dBc or so would be nice, 1KW PEP, CCS, HF+6M, RX IMD3 in the 40dBm region and at least two synchronous RX channels, TX efficiency north of 50% (And still north of 50% at 10dB backoff).

The RX is pretty much done (The same ADCs measure output voltage and current on transmit which both closes the cartesian loop and allows the complex load impedance to be computed, helps for setting the drain voltage correctly), it is this envelope tracking supply that is kicking my arse at the moment (A linear passbank proved the principle, but that is untenable in a real radio)...

PA stage is a conventional push pull broadband LDMOS affair, but the output filter bank plays games with harmonic open/short loads to optimise drain voltage envelope (Gets you about 8% efficiency improvement on some bands), I don't want to end up doing full on EER with class E or the like, too much of a pain.

Regards, Dan.

[MagicSmoker]

A current-mode control buck using a Type-II compensator should have a crossover frequency of at least 1/5th the switching frequency (better is possible), so you should be able to meet your bandwidth requirements with a conventional buck running at 100-200kHz.

In addition, a multiphase buck can't have maximally fast transient response AND minimally low ripple at the same time - to get faster transient response the inductors in a multiphase buck have to operate with a higher percentage of ripple current, negating some of the improvement in ripple from interleaving the switching periods.

Also, please clarify what you mean by "two quadrant operation"...

[dmills]

Two quadrant operation in this case refers to the buck output always being positive, but being able to source and SINK current, so as to be able to discharge the output caps rapidly back into the DC input caps on the falling slope of the envelope.

The issue with the switching frequency is not so much with the control loop as with the post filtering required to get the ripple down far enough while maintaining the required response.

Bear in mind that ripple on the final amplifier supply will end up as AM/PM sidebands in the amplified signal, so a LC filter going over above the modulation bandwidth but well below the switching frequency is pretty much going to be required, much easier at 20:1 or so frequency ratio then at 5:1.

Regards, Dan.

[vtp]

Since you are after two quandrant operation how about adopting a class-D amplifier to fit the purpose?

Use a half-bridge output stage and set the negative rail at 0 and positive higher than you maximum required output voltage to allow some (modulator) drop, bias the amp to midpoint of rails. If you take negative feedback after output LC-filter you can make it self oscillating by introducing some phase lead. High enough switching frequency could allow for reasonable modulation bandwidth if I understood the need correctly.

DC-connected it would follow your input voltage (with gain) and act as an 2-quadrant power supply.

Topology should be relatively easy to realise with discrete output stage, integrated high/low side drivers and an opamp at the input.

[MagicSmoker]

Two quadrant operation in this case refers to the buck output always being positive, but being able to source and SINK current, so as to be able to discharge the output caps rapidly back into the DC input caps on the falling slope of the envelope.

That's what I thought you meant, but wanted to make sure because it is an unusual requirement. Intuitively, it would seem that the rate at which the power supply voltage can be increased is much more important than the rate at which it can be decreased; too slow an "attack" component to the transient response will result in clipping, whereas too slow a "decay" component will merely result in slightly higher dissipation in the PA.

As you've already determined, making a bidirectional converter (ie - forward buck/reverse boost) will be difficult; compensating such a converter for good transient response and stability will be monumentally difficult and will almost certainly require a digital controller implementation (either with a MCU/DSP or a CPLD/FPGA).

You might want to try running your PA with a fast linear power supply and test hanging varying amounts of capacitance on its output.

The issue with the switching frequency is not so much with the control loop as with the post filtering required to get the ripple down far enough while maintaining the required response.

Aye, this is true. You might find OnSemi Application Note AND8143/D useful, as it discusses how to optimize the transient response of a buck converter (as compared to how to make a converter unconditionally stable).