spread-spectrum model

I'd check for cross-correlation as well, so no ganging up in systems using multiple channels in some signal path.

Depending on details, the problem could manifest itself as peaks or ripples in the time domain, your beloved homeland.

Joe Gwinn

When my engineers get too fussy about stuff like that, I remind them "it's just a power supply."

TI has a couple of intersting appnotes

Their little TPS54302 type parts have radical looking PWM, but the final DC is super clean. Nice trick.

In my world, we have multiple parallel components (like array sections) in the signal path powered by independent power supplies that are required to have independent noise, to prevent correlated gain when these parallel paths are summed, say in a radar beamformer.

Telling the power-supply folk that it's just a power supply is a good way to get buried in details.

TI stuff is widely used in radar, but in the most capable radars the dithering is provided by bespoke radar firmware, and not left to the converter chip. But those chips do work well.

Joe Gwinn

My intent was to keep it simple and get it done.

We've decided to use home-made half bridges in the 8-channel programmable power module. The TI and ADI switching regs are just too smart. We'll use the reg chips when we just want a fixed power supply.

I was thinking that we could use a DDS architecture to generate the PWM into the switching half-bridges. We could wobble the frequency setting to spread the spectrum.

Maybe replace some of the LSBs of the frequency-set register with a pseudorandom pattern, a different one for each power supply channel. Change those LSBs at some rate, 20 KHz or something, to fool an EMI-test spectrum analyzer.

A pseudorandom pattern will average to 0.5, which affects the average switcher frequency, but we can deal with that.

I suppose I could draw a diagram.

We'd like the fine-grain spectra to not overlap, across all 8 channels. Fun.

Given eight unipolar half-bridges, we'll allow users to use a pair as a full bridge to drive one bipolar load, or three to drive a 3-phase load like a torque motor. In those cases, I think we can still allow each phase to have its own independent spread-spectrum thing. The motors won't care.

The STM32F4 that I use in my inverter/charger has a SS clock option. I have not enabled that yet but intend to eventually for lower peaks from the PWM outputs driving all the power circuitry.

boB

That's cool. I'm designing a bunch of plugin modules that would all get a 50 MHz clock from the backplane. Since my PWM frequencies will be fairly low, we should be able to fuzz up the spectra in the FPGA on each board.

We could code a generic ss PWM block and use it everywhere.

The radar folk prefer TI over ADI for such things because TI does the digital parts better.

This is a common approach in radar.

Yep.

Another reason to dither the lsbs is to allow correlation processing to pull things up from below despite ADC quantization, where things get very granular and thus non-linear.

The max EMI regulations specify something like a one-second averaging window, so some alignments are probably OK. A sufficiently long pseudo random sequence will make coincidences rates small enough to not matter, even if the sequences are not actually orthogonal.

Small motors (and transformers) won't care. But one thing I learned after being buried by the power folk is that at the megawatt level, harmonics and overlaps must be handled, or the motor or transformer will fail prematurely due to corona induced within the windings, there being multiple ways to cause this.

.

Joe Gwinn

This dithers the frequency, which is a valid approach. But dithering the phase before going to the DAC core to generate the waveform is also widely used. And one can do both at the same time, particularly with different sequences, so the peaks are spread out in 2D.

Joe Gwinn

Noise at the local level is best correlated, as it is more predictable - you avoid low-frequency beat frequencies in the local regulators - which can and will show up in a detector's BW and in the regulators' outputs.

A master clock, phase shifted for various local users, can be dithered for the system (box), which is the actual, final radiator.

Your engineers can get REAL fussy, if the system's non-compliant way past the development's due date.

RL

But...but... it's just a power supply!

Presumably uncorrelated spread-spectrum will make wideband noise at an output, not a beat.

Our box has a 50 MHz clock that is bussed to all the plugin modules, and it can be locked to other boxes or to a 10 MHz reference, so we can't usefully dither that. I guess each module could have its own VCO, but that would mess up synchronizing modules, and complicate things. Spread-spectrum sounds easier.

Eventually, some giant customer may want CE stickers, so we'll do the easier things now, to improve our chances of passing an EMI test. A bit of VHDL in the FPGAs would be easy.

Unsynchronized power supplies on the same board can influence each other, unpredictably with load, to produce audible harmonics.

Ignore the effects at your peril.

RL

Given a common 50 MHz clock and a sensible architecture, we plan to synchronize modules. Like, for instance, paralleling 3-phase AC sources to get more power, or running two at phase offsets.

But I'm not concerned with acoustics, given that the modules are in a rackmount chassis with big fans. And a hiss would be better than a whine, which is the heart of the spead-spectrum concept. Ears are spectrum analyzers too.

Yes. Fix the shielding and grounding story until these effects no longer matter. This is done in radar, as discussed upthread.

Joe Gwinn

Pseudo-random frequency is probably best as long as the average frequency is the specified and wanted center frequency. So the PRN source may need to be massaged to do that ?

What I see most often in processors and an old SMPS chip I seem to remember all use triangle wave modulation to the clock frequency. This way, it averages to the center frequency. Those chips also (IIRC) have some other options where the modulation is biased to one side or the other of that spectrum. Top or bottom. I don't remember exactly.

I remember hearing about a 12V switchmode power supply for ham radio where you could adjust the switching frequency up or down slightly so that the EMI could be tuned away from the operator's radio frequency he was operating on so it would not interfere. I guess if you can't make the EMI go away by design, that might be the next best thing.

I have yet to find a 12V SMPS supply that I could not hear on my HF radios so I just use a linear PS for that at home.

boB

Yes, triangle wave modulation is common, but if center frequency is important, they modulate phase, and maybe amplitude.

It's far easier to move the frequency than to suppress the EMI everywhere. Ham operator budgets are not large.

That's the cheaper approach.

I don't know if this would be good enough, but a standard dodge is to have a switcher go from AC line in (120 or 240 Vac) to 48 Vdc, and then use linear regulators fed from that 48-V bus for all DC loads.

Joe Gwinn