power supply "pre-emphasis"

See your doctor for a brain scan, this may be serious...

Simply pass the output of the power supply through a 1:1 ratio transformer that is connected so that its two coils are wired together so that the AC fields oppose each other. DC will pass normally and the AC will be effectively choked out. The core has to be sufficient to avoid DC saturation.

In FPGAs most of the power is consumed in the switching transitions. Adding more transitions won't help that.

As for the power supply question: all you need is a "feed-back" device that looks forward in time to detect changes in the power source or the load, and makes the appropriate corrections in the PS control circuitry just in time.

Yes, I know the consumption will go up, but can we kill stubborn spikes? ECL gates are easy to decouple because of the differential nature of the circuit. What if I set up 1->0 transitions to balance out the 0->1 transitions? By setting up bogus complementary calculations?

I can't tell if you're making fun of me or agreeing. That is exactly what we want. We know what we'll be sending, can we do something to attenuate a known spike at 1GHz say, by injecting an out of phase 1GHz signal in the power supply?

It is almost impossible to get to the perfect cancellation. If you cancel a frequency at one place, it is likely to rise at some other place. It is better to stick with the conventional EMC measures.

It is difficult because the spikes due to the switching are unique to the particular circuit and depend on temperature, voltage and the logical state.

Why do you think the trash is getting through the power supply?

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

Basically yes. But you'll have to consider the paths the 1GHz signals go. And then you can choose for what (very small) space, temperature and moonphase your compensation works.

You may also consider providing a "compensation output", users can use to subtract the interference in all interfered devices, shaped with the appropriate impulse response (for each interfered device), of course.

Active compensation of EMI is increasingly difficult above 10MHz.

Power suppliy conversion components don't make spikes at 1GHz. They are simply too crude to do anything above about 50MHz, even accidentally.

Control, drive and interface circuitry might, however.

Track down the source and troubleshoot from there.

RL

Alas, the model of a single-degree-of-freedom is faulty. If you create a signal to null another, you are only nullifying the output on ONE axis, and instead of a dipole radiator you create, at best, a quadrupole radiator. The first rearrangement of wires will destroy the null, and you're back to a dipole.

It's a net win, but not a BIG net win. Damping near the source is a better strategy.

I once saw a factory for auto electronics where the devices were tested for EMI emissions, then the failing units were opened up, the wires re-stuffed into the case, and retested. Eventually, they all shipped. >>shudder

You should not have parasitic oscillations on a dc supply. Filtering them out is not the answer. You need a lag-lead circuit somewhere to provide phase advance to stop the oscillation. Sometimes an RC series network on the output of an amplifier (to ground) does the job. The resistance is normally 1 ohm!

You mean phase lead - just say it!

indeed. that being said, over the last year or so a number of papers have come out of places like VPEC, on active compensation. A typical approach is to whack a winding on the transformer, suitably phased and capacitively connected to earth, so that it generates spikes that tend to cancel out dV/dt spikes etc.

of course Vicor have a far, far better trick for dealing with that nasty bouncy drain tab.

Cheers Terry

Thet's a passive compensation technique, aimed at fundamental and major lower harmonics of the switching frequency - it is inneffective at the 'spike' frequencies due to unpredictability of phasing and general incoherence. The same thing is attempted using active nulling, at moderate signal power levels, with the same problems at the higher frequencies.

The OP's 'pre-emphasis' technique is more likely to be attempted within an active transmitter's pass-band to reduce distortion in the intelligence, rather than reducing effects of stray carrier harmonics.

Vicor products tend to produce unneccessary emi levels, no metter what topology they may choose to use, or the potential advantages offered. I've never really understood how they manage to do this, or why.

RL

I've gutted a few vicor modules. marvellous construction, and a lousy topology. Poor thermally too - their approach seems to be to try and reduce the thermal resistance, rather than reduce the losses. subsequently, its kinda hard to suck out full power with real heatsinks.

but the shielded magnetics are nice, and the ass-backwards FET connection (Drain to V+, Source to xfmr) deals nicely with the wobbly-drain issue.

actually uase them in a product? ROTFLMAO!

Cheers Terry

We use a lot of Vicor modules... often because it's what our customers want (they're... uhh... big government contractors, let's say). In general they seem perfectly reliable, which seems to make up for the dated design and a relatively "impressive" price tag.

Could you expound upon what, exactly, a "wobbly-drain issue" is?

Do you have a preferred brand? :-)

Thanks,

---Joel

conventional converters sit the FET(s) at the -Vdc rail, and the drain (the big Cu tab thingy) bounces fromm -Vdc to +Vdc (and maybe +Vdc + Vo*Np/ns) + leakage spike.

this is one of the major sources of dV/dt coupled noise - typically drain connected to (insulated) heatsink, which is usually earthed.

by sitting the drain at +Vdc and configuring the circuit "upside down" its the Source that wobbles a lot, but its only a little blob of metal, and usually isnt bolted to the heatsink. For a T0-xxx FET, this removes the entire noise source. for a DCB structure like the vicor bricks, the source also sits above the HS, but the area goes down a LOT, and thus so does CdV/dt current into DCB (which is usually earthed)

DIY. they do have some advantages (meeting EMI aint fun, the i/p & o/p currents arent always that nice, but C's & -26 Ls are cheap. I have used a couple, but only because they were there, and I didnt care at all about the cost.

extracting rated power tends to be somewhat tricky as well, as it tends to require a lot of extra heatsinking.

Syncor open-frame stuff is awesome. M. Schlect is a smart cookie.

Cheers Terry

I guess the major complaint is that EMI and even ripple controlling components tend not to be on-board. There's poor control of capacitive current into the plate - which carries source-connected control real estate, as much as drain real estate - particularly when there's no solderable connection to it in the footprint.

Close thermal coupling to the heatsink of ALL sections has it's drawbacks.

RL

yep, but without it the mnodule would fry. like I said, lousy design, great thermal management. result is a mediocre product.

Cheers Terry