Random PWM for SMPS? 5

[GonzaloEE]

Hi all,

I've read some papers on certain ramdom-PWM techniques for SMPS switching drivers (Google 'random PWM'). The authors claim reaching lower noise levels at the output, because - as they say, random switching spreads noise freq. components over a wider band than the standard clocked PWM.

I like the idea of having like 'white noise' at the output, easier to filter than those well-known voltage spikes of most SMPS.
Does anyone know where to get further info (or sample uC programs) on this subject?

Cheers,
Gonzalo

 

[itsmoked]

It is a powerful and commonly used tool for switching systems of all kinds. The end result is keeping any particular peak in a device's noise spectrum below a specific level to achieve legal certifications.

I know that some companies offer dithering oscillators chips for this reason.

Keith Cress
kcress -

http://www.flaminsystems.com

 

[Warpspeed]

The switching spikes will still be there, and are no easier to filter.

All you are really doing is adding random jitter which may be either more or less of a nuisance as an interfering noise source, depending on what is being interfered with.

Another problem is that if a circuit is designed to have a fixed constant operating frequency, it will also have a fixed known efficiency, fixed output power transfer, and known limits for maximum cyclic voltage and current within the circuit. Inother words everything can be optimised.

If the switching frequency jumps around randomly, so will the output power, that cannot be a good feature.

It may have some obscure application but I cannot think of any off hand.

Sounds more like a solution seeking a suitable problem.

 

[VEBill]

KC is correct - it is a trick used to pass EMI tests.

 

[sreid]

A star for VE1BLL. But it is a perfectly valid and accepted technique used to smear the noise spectrum.

 

[MacGyverS2000]

Also search on "spread spectrum", "dithering", and "clock jitter". I've implemented it in software in the past, but it's much easier to let specialized hardware take care of it. In particular, search on "spread spectrum clocks".

Dan - Owner

http://www.Hi-TecDesigns.com

 

[treez]

There are several Microcontrollers by Microchip which have an internal oscillator which can be dithered. And i think its the MCP360P (or something like those letters/numbers) which can be driven by one of those dithering microcontrollers in an smps. (Also dsPIC may have dithering ?)

As others have said, its just something you could do if the conducted EMI is bad. If youre ok on conducted, dont bother dithering...but it shows the versatility of uC oscillators in SMPS controllers in that you can implement this stuff if you want

The alternative being just to up the spec on the input EMI filter.
Of course, an active power factor corrector at the front end will make the emi filter able to be smaller too.

 

[GonzaloEE]

Thanks to everyone for your valuable contributions. Maybe this is just EMI make-up. Anyway I think it's very interesting having a chip taking out some harmonics at s/w level, like done in power inverters.

That's it.

 

[Warpspeed]

Gonzalo, it does not take out harmonics or reduce the REAL noise level at all. The peak or rms measured noise amplitude stays exactly the same.

All it does is play hide and seek with a spectrum analyzer.

 

[itsmoked]

I beg to differ. The peak certainly does not remain the same. Isn't that is the point? The RMS could well remain the same, heck, it could even get larger but the peak goes down as it is now distributed.



Keith Cress
kcress -

http://www.flaminsystems.com

 

[Warpspeed]

Keith, if a certain noise energy "packet" is created each time the supply switches, the total noise energy does not change if the frequency is dithered.

It could be radiated or conducted noise, but the effect is the same.

Spread spectrum is only an advantage if you are "looking" at a very narrow bandwidth window. If full spectrum noise power is taken into account the total energy remains exactly the same.

For instance, switching spikes on a dc rail will not be reduced in amplitude by dither. They just jitter around in time. If those noise spikes are troublesome, dither is not going to be of the least help in reducing the amplitude.

 

[itsmoked]

I see. Then it would need to be more specific as related to any particular FCC test and how it's administered (averaged).

Keith Cress
kcress -

http://www.flaminsystems.com

 

[Warpspeed]

Dither does absolutely nothing for reducing the actual real noise power of radiated or conducted noise.

All it does is flatten the observed peaks on a spectrum analyzer.

As the very narrow search bandwidth of the spectrum analyzer slowly sweeps across the spectrum, your horribly noisy switching supply dances around hiding.

It is a great way to gain easy FCC compliance, unfortunately the poorly designed, and probably VERY noisy switcher may cause untold grief when put into actual service.

This sort of thing is not to be encouraged.

I have personally designed commercial switching supplies used in the telecoms industry. If I had tried this dither technique to get through EMC testing, they would probably have shot me at dawn.

 

[sreid]

Unless traditional noise reduction techniques are used, random frequency hopping will not usually allow you to pass EMI tests. However, when it allows your device to blend with background noise, dithering appears to me to be "Fair."

 

[Warpspeed]

But that is the whole point, it does NOT blend the noise output with background noise.

There are still massive bursts of energy appearing first here and then there randomly hopping up and down the whole frequency spectrum.

If the victim circuit that is being interfered with is broad band, these noise bursts simply cannot disappear or average out. Each noise burst appears with full ferocity undiminished in the susceptible circuit.

That is the whole problem. It only APPEARS to disappear when you are only looking at an extremely narrow bandwidth window in the whole spectrum, but very few real world circuits are like that.

Frequency dither is a party trick only, it has no practical application in switching supply design, except as a rather dubious means to circumvent the "spirit" of noise compliance measurement where a spectrum analyzer is used in the testing.

 

[GonzaloEE]

I appreciate your corrections on dithering, my apologies for not being clear enough in my last post about 'removing' harmonics.
I understand that most of these chips are actually 'spreading around the dust' over a wider bandwidth than the natural freq. pattern of clocked PWM, and the total amount of noise energy remains the same.

I'm just trying to go a little further with it, hopefully doing some signal synthesis at the switching stage, for clearing or at least snubbing some selected (lower) harmonics, leaving the rest for the L-C output filter. Thanks to your useful suggestions, now I know dithering is not a good option.
We agree that any switch chopping sharp a current will produce high freq noise anyway. I'm actually playing a little with Fourier and switch timing, for cutting down some lower freq. harmonics at s/w level, keeping the same PWM frequency to meet the SMPS filter/Xformer freq. ratings.

I'll post some links when I got the time, for clearing this a little more. Thanks to all!

Regards,
Gonzalo

 

[Warpspeed]

The faster you can make it switch, the more efficient will be the switching action. Raw switching speed (as opposed to operating frequency) is highly desirable, if you can actually get away with it.

The first aim should be to prevent these violent switching changes from radiating directly. That is more a function of physical layout and construction method than schematic design or topology.

The second aim is to reduce conducted noise transmission at both the input, and the output. That is more a function of both topology and filtering. Many switching topologies produce a violently pulsing current at either the input, the output or both. These can be much more difficult to filter than a more benign "constant current" topology.

If the whole thing is carefully thought through, and carefully planned, the results can be pretty good, even without resorting to tricks or band aid solutions to sneak it through final compliance testing.

There still seems to be a place for dithering switchers, as the semiconductor manufacturers make the chips. But my own personal feelings are very strongly against this technique.

In the end, it is your customers that are going to be either happy, or mad as hell. Short cuts are for short term employees <grin>.

 

[eeganmtl]

Spread spectrum clocks are definitely helpful for passing EM emissions tests - without, as has been pointed out, actually reducing the total noise energy.

Whether the "smearing out" of the noise spectrum is actually advantageous (other than in passing the test) depends on the nature of the noise "victim".

The fact that the testing is done with a narrow band spectrum analyzer reflects two factors:

1. The testing primarily exists to protect RF communications, some of which are still conducted in narrow frequency bands.

2. The test was devised before "spread spectrum clocking" was a widespread technique.

The bottom line is that in some cases, if the real-world communication system is narrow-band, spreading the spectrum will reduce the DUT's worst-case interference with it.

 

[GonzaloEE]

Thanks Warpspeed for that lead on the design goals. Yes, I'm going in that direction, solving the EM radiation and then the conducted noise. My objective is not robbing a FCC stamp, but researching some methods for making the LC filter job easier, through some kind of 'smart' switching timing. Dithering may well be a subset of these solutions, doing fine in preventing the DUT from jamming communications -as eeganmtl say. Maybe the problem is not using dithering, but using ONLY dithering for handling noise issues.

I'm looking for some way to modify amplitude and phase of harmonics by means of switching timing, then reducing the lower(bigger) ones. I'm OK with starting with some uC or DSP, though finding some off-the-shelf IC to do the job would be just fine.

OK, I'll talk more of this when I got more time. Thanks for your help!

Gonzalo

 

[benta]

If you do not need input-output isolation, try checking some lesser known topologies: SEPIC and Cuk.
These have the good feature of a series input coil, giving a nice smooth current draw and few emissions back on the supply line.

Benta.