power supply idea

Looks like you have invented the buck converter.

In thory, pulse-width contol of the output could give excellent stability under load -- but the filter is going to cause droop. Unless you are very careful about the design of the filter, the phase shifts it creates will make the feedback loop unstable. An integrator in the loop will stabilise this at the expense of a much slower response time.

Somewhere in the loop you need a dominant pole so that (to use audio amplifier terminology) your roll-off is 6dB per octave until the loop gain has dropped far enough for stability when all the other phase shifts kick in and the slope increases to 12dB per octave or more. Rather than integrating the feedback, transferring the dominant pole to the filter will result in less output noise and a faster response to a step increase in the load.

Is your "spread spectrum" dodad supposed to mitigate EMI?

Danke,

It smears it out over a range of frequencies, and makes it look better on the screen - no big frequency spikes, but many more smaller ones.

"Mitigate" depends on how the hash messes up your particular system.

Yes, but the only one that most designers care about is the EMC receiver at the compliance test lab.

John

I invented a control algorithm. All the buck chips that I know of are all feedback driven, and will slam into either rail if the feedback divider is broken. Blow things up.

An LC filter is at least 2-pole, usually more, and we have no idea what crazy stuff the customer might connect to our power supply. So the less we depend of feedback from the output, the safer things get.

A power supply without feedback is always stable.

It will help pass CE lab tests, by about 20 dB. It's easy, so why not?

It won't reduce ripple or fast switching spikes, which is what my users might care about. It would improve the peaks on a spectum analyzer, which is what regulators (government regulators, not voltage regulators) care about.

Actually, the DC power that one sees on an airplane or in a car is nastier than anything I can reasonably make, even on purpose. So my concern is stability and not blowing anything up.

If you made it three poles, with one of them significantly lower frequency than the other two, stability would be much easier to obtain.

An algorithm arguably eliminates a 555 triangle generator as a potential spread spectrum source. LOL. So, what's hidden in plain sight behind all of your left hand side, symbolic sleight of hand? In other words, how do you implement your control algorithm?

Danke,

It's all in plain sight. Well, the guts of the PWM converter isn't, but that's pretty obvious.

The PWM converter, and in fact everything, will be implemented in an FPGA, with an ADC to pick up the output voltage.

May as well go pseudo-random on the spread spectrum part. Any audible side effects would be hiss, not whine.

Capacitor ESR, native or added, helps.

I also want to kill the Q as seen from the load side, so it doesn't ring much if they switch an inductive load.

The load-from-hell is of course some box with a switching regulator power supply, that looks like a negative resistance load.

I have done what you propose, but I did not add the spread-spectrum part.

If you add a current sense on the output, you can characterize the non-linearity of the power stage, and do feedforward compensation. So your response will be snappy. You still have the settle time of the LC filter, that's harder to counteract with feedforward.

One concept I never had time to implement, was to do in circuit compensation. So in your function test, add a swept current load on the output at different output voltages, and feed the results to the feedforward lookup table. That will take care of variations on components, albeit wont reduce temperature affected errors.

One of our applications has a fixed, stiff 48 volt supply. So we could characterize the switcher output as an ohmic source, and use the sensed current to null out most or all of those ohms, so the integrator can have an even smaller influence range. Or even no integrator! We need a current sensor anyhow.

Another product will have an isolated dc/dc converter driving the half-bridge, and it will be fairly soft, nonlinear at that. We will digitize that 60 volt supply anyhow, so it and the current together could be compensated. That might require a divide in the FPGA. I'll ask my FPGA kids if they can divide.

Sounds like a fun project. You could do this with a 0.3 USD ARM CM0 processor, but I guess you have the FPGA on the board anyhow.

division is just like any other operation, it just takes more cycles since since it can't be done in parallel like a multiply

Have you seen the class-D audio amplifiers by Bruno Putzeys ?