Test & Measurement Designline Blogs
Spread-spectrum clocking to reduce EMI: clever or cheat?
Bill Schweber
4/13/2011 4:02 PM EDT
Most designers are familiar with the technique of using a spread-spectrum clock to reduce apparent EMI/RFI emissions. By deliberately dithering the system clock, the radiated energy is spread across the spectrum and thus its peaks are reduced, which allows the product to meet regulatory or industry specifications. The technique is now well-established, and vendors offer clock ICs with adjustable spread widths and rates, as well as advanced pseudorandom spread algorithms.
But when discussing the use of spread -spectrum clocking to meet EMI/RFI requirements, I find that engineers are divided on the approach. Some feel it is a legitimate, very-low-cost tool in the designer's kit that helps the design meets market requirements. Others feel it is short-cut cheat, too often used instead of proper EMI design methods. Still others feel it should be used only after all other conventional steps have been taken, such as shielding, grounding, layout changes, ferrite beads, to cite a few.
One argument against using spread-spectrum is that you are very likely actually making your design "problem" into someone else's. As you spread the energy, yes, you may meet a specification, but you also introduce the likelihood of unexpected problems when your spread energy mixes with as-yet unknown or undefined energy in other nearby or connected systems, each with their own frequencies and amplitudes. In short: "hey, I met the spec--after that, it's your problem, not mine!"
As with most engineering designs, there is no right or best answer. What makes sense depends for your project's priorities, budget, constraints, market forces, and the balance among all the tradeoffs which every design encompasses. Perhaps in the ideal world, the design would first be made as EMI-robust as possible, and then spread spectrum would be added for a little extra insurance, but only if needed.
[Ironically, using spread spectrum is absolutely contrary to another engineering imperative, central to test equipment and many data links: to have a clock which is as perfect and jitter-free as possible. That's one of the many contradictions and sources of discomfort engineers see with it.]
Have you ever used spread spectrum to reduce EMI and meet a spec? What's your view on spread spectrum as an EMI reduction technique?
•it's a great idea, go ahead and use it right away;
•use it only after everything else that should be done has been done, and you are still stuck;
•use it only after you have already met the spec, just to buy a little extra margin;
•or don’t use it at all, since the unforeseen consequences to the overall system and broader application are too risky?
What do you think? ♦
|
|
|
|
|
gmorita
4/13/2011 7:13 PM EDT
Personally I believe spread spectrum techniques are a cheap way to to meet EMI requirements. It takes advantage of the way EMI measurements are made by shifting the spectrum around while the EMI test system tunes across the spectrum. The noise is still there, it's just not where the test equipment is looking at the particular time.
Spread spectrum techniques do not reduce the amount of energy in the noise spectrum. If I have a DC-DC converter whose frequency varies from 1-1.1MHz over a 10 second period, the energy at any given operating frequency will be constant.
Spread spectrum techniques in switching converters can also cause low frequency ripple on their outputs which can cause issues with sensitive analog circuitry.
This whole issue reminds me of the philosophical question: "If a tree falls in the forest and there is no one there to hear it, does it make a sound?" The answer of course is YES, just because someone isn't there to hear it doesn't mean there is no sound.
Therefore just because the EMI test system doesn't see any noise at a particular frequency and point in time, it doesn't mean there isn't any noise at least some of the time.
Sign in to Reply
gmorita
4/13/2011 7:15 PM EDT
Sorry, I meant to say a way cheat, not cheap
gmorita
Sign in to Reply
zeeglen
4/13/2011 9:15 PM EDT
use it only after you have already met the spec, just to buy a little extra margin; or don’t use it at all if it puts the entire system at risk.
That's my vote. Have never used jitter to "reduce" EMI. Usually it works both ways, reduce emissions with good layout, clean grounds, chokes, capacitors, filters, and shielding and the equipment becomes more robust to external fields. A swept clock dies not protect from outside influences.
Sign in to Reply
gsdg90
4/15/2011 3:27 PM EDT
Two comments:
1) Spread Spectrum is equivalent to adding jitter-- this is quite undesirable for many applications.
2) It is not cheating and this is not a loophole. Yes, the total EMI energy will be the same, but it is no quirk that the FCC is more concerned with power spectral density than total radiated power. Spreading this energy so that it is less coherent makes it less burdensome. This fact was well understood when the FCC specified the measurement resolution bandwidths and detector characteristics.
Sign in to Reply
kendallcp
4/18/2011 7:34 PM EDT
Well, the comments so far seem to corroborate Bill's point about engineers being divided. It makes a big difference whether you're a producer or a consumer of the emitted signal. Spreading the spectrum will only make a big difference to the technical performance of the equipment if it's spread wide enough that some of it no longer falls inside your detection bandwidth. A signal passing through an IF filter with a 200kHz bandwidth will be just as messed up by a near-channel pure tone as by a near-channel spread spectrum with say 10kHz width. If you're unlucky (or rather just averagely lucky) the typical point at which you start to do good by spreading your spectrum is the point at which your own system doesn't work so well. Some systems that claim to be 'spread spectrum' are in fact systems that wouldn't operate at a constant frequency anyway. And the majority of them are designed not to reduce the impact they have on real systems but just to ensure that it's easier to pass measurement - just a fact of commercial life. The spectrum analyzer at the test house is the only piece of receiving equipment many interference producers ever explicitly encounter.
Sign in to Reply
pcsalex
4/20/2011 8:18 AM EDT
it is a cheating, which sooner or later it will "bite back", despite that it is one "nice inexpensive elegant solution", since you could cheat out the test system but not real life condition, but who cares about it after the testis passed and the gadget is sold, the money is in the bank....
Sign in to Reply
peralta_mike
4/20/2011 9:38 AM EDT
Spread-septic (I mean spectrum) is a cheat.
But as long as your product has full-disclosure
then your customer can make his own decision about it. Whether it will work in his system or not.
So my answer is go ahead and use spread-spectrum if you have to but make sure you have full disclosure.
Sign in to Reply
Bob Lacovara
4/20/2011 11:26 AM EDT
Aw, c'mon. It's just another tool in the box. If you can tolerate the jitter, and if it allows you to make spec, and if no one complains (no contact, no foul) then what's the problem. Let me ask that again: who is injured by the technique. If you make spec, but are pretty sure that 1 in 5 adjoining systems are going to take a hit, then you'd best think again. If you've pushed your spurious emissions around below everyone's pain threshold, then on with the show...
Sign in to Reply
greenmeanie
4/20/2011 7:30 PM EDT
As the author put it, design is all about trade-offs. Not every design is perfect and it needs a solution for EMI or any other problems. As Bob put it, if you introduce Spread Spectrum and if you do so in a way that it does not affect system functionality while passing regulatory compliance at the same time, I would gladly use it and have been using it. Infact, this has become a quite common solution these days that specifications like USB3.0 or PCIexpress have Spread Spectrum built-in. There are numerous other benefits: the spectral energy redistribution not only at the fundamental, but at the harmonics as well.
Let's look at the alternatives: Passive solutions have a low pass filter effect and you do not see EMI reduction after the 3dB roll-off point. Passive solutions also degrade signal integrity. So, if you can solve your EMI needs, save board space and be cost effective, with fast time to market when faced with EMI issues after the fact, maintain signal integrity, then you can see that the benefits tilt in favor of Spread Spectrum.
Sign in to Reply
ChipWeller
4/21/2011 9:51 AM EDT
Spreading the spectrum reduces the EMI only if you spread it outside the receiver's passband. The spectrum analyzer for EMI sweeps uses a fairly narrow passband filter width, 120 kHz in the 30 MHz and above frequency range. If the susceptible device is an intentional receiver with a bandwidth of 120 kHz or less, spreading will work great. If the device is an unintentional receiver it likely responses to a much broader bandwidth and spreading will not help to reduce interference.
Sign in to Reply
darrien
4/22/2011 2:11 AM EDT
In fact spreading the spectrum can be achieved by various ways: frequency modulation (FM),random pulse position or pulse width modulation,sigma-delta modulation, chaotic modulation etc. FM is the most successfull approach only if modulation frequency is higher than EMI receiver bandwidth (for CISPR22 RBW=9kHz). In fact it is proved in several papers that FM doesn't cause low-frequnecy output voltage ripples in dc-dc converters in CCM.
Sign in to Reply
BASH_#1
4/28/2011 3:20 AM EDT
My vote is for "•it's a great idea, go ahead and use it right away;"
Just be sure about your requirement at the start of design. Is your design can tolerate spread, means the jitter which will be result of spread and the frequency of spread modulation. You don't have to fully rely on spread spectrum for your design robustness, but it will save space, reduce inventory and so cost. It will be blessing for you when it is programmable and helps you to pass regulatory tests at critical point without spinning the board.
Sign in to Reply
kcotton
4/28/2011 8:55 AM EDT
use it only after everything else that should be done has been done, and you are still stuck
I can't talk against this technique, it saved my life, particularly useful for LCD display, where I cannot modify the LCD panel design...
But before that case I was absolutly against that cheat. In many case it reduce the design margin and cause strange behavior.
Sign in to Reply
BeeGee
5/4/2011 5:02 PM EDT
The main reason for tight EMI emission standards is RF interference, and that only makes sense in the context of a given frequency. The fundamental issue isn't that your widget creates too much total emitted energy (unless your widget is very high power); but rather that the energy is concentrated at one frequency due to "flaws" in your product. These "flaws" are a side-effect of fast, synchronous designs and high-speed external connections. Spreading the energy just counteracts a side-effect of synchronous design.
Sign in to Reply
DU00000001
5/26/2011 3:35 PM EDT
Hej, from the shortwave listener's point-of-view, spread spectrum is cheating. It allows to pass EMI tests but - not reducing overall energy emitted - adds to the basic noise level.
It is what makes live hard. Like "light pollution" making the astronomers' lives hard.
Sign in to Reply
WKetel
5/28/2011 8:40 PM EDT
Spread spectrum is another "cheap trick" to get around a regulation. It does reduce some kinds of interference, but if a switching supply, or a microcontroller is radiating noise into my audio system, it is not going to make my problem any smaller, just a bit harder to locate the source. Noise at a constant frequency can be rejected by a tuned filter, but noise that is spread spectrum would need to be bypass filtered, or band-reject filtered, both of which are a bit more complex than a single frequency rejection filter system.
Sign in to Reply
Bob Lacovara
5/30/2011 1:55 PM EDT
WKetel, you have some right on your side: in some cases, spread spectrum is a means to dodge a regulation. But in an environment full of emitters, there has to be some level below which you, as the receiver, have to live with. Now if spreading your emissions widely enough puts potentially interfering emission far enough down, then that may be a case to accept the (admittedly) shady technique. At least until the requirements boys get a bit more clever and address SS techniques specifically.
Sign in to Reply
6J5
6/2/2011 4:26 AM EDT
In GPS & ESM circles this is called noise jamming. In the former spreading noise is tolerable due to distance to satellites but created the near-far problem when local augmentation pseudolites were proposed due to their relative proximity making precessing gain much less effective. If all and sundry went for SS ( cellular services, GPS/Galilleo (GNSS), EMI measures,TV broadcasting, etc) a rise antenna noise temperatures, and therefore noise floor, would begin to defeat the original attributes of SS. Too much of a good thing?
Sign in to Reply
RWatkins
4/10/2012 11:34 AM EDT
Although what you say is probably true, it is the case with many of the "newer" schemes. Take a look at what is discussed with ultra-wideband to see an even more interesting discussion on these lines.
Sign in to Reply
Rush.Hood_#4
11/14/2011 1:10 PM EST
Can't find the original citation, but my opinion is summed up by "This is like getting rid of a cow pie by stomping on it."
Sign in to Reply
HurinThalion
11/14/2011 1:34 PM EST
If you just want to pass a regluatory hurdle, I would make it my fix of last resort. However, the last time I saw someone use this was in a 90's automotive design. Everyone loved it until the platform folks did an FM radio listenting test. When you hit "seek the next higher station" button it would stop at an "empty" station and growl at you. I was the poor schmuck who was tasked with re-designing it.
It is a blatant method of "fooling" the quasi-peak detector and spectrum analyzer sampling response time. We all better hope that CISPR 14 & 16 don't address this one day.
Sign in to Reply
vintage_bits
4/9/2012 1:54 PM EDT
I think spread spectrum clocking is a terrible thing to do in a system designed to interoperate with other components. When integrating such a system, I have seen configurations that were unstable with spread spectrum clocking in use. So, for "open" platforms I think it is a terrible method. Even in a closed system, it feels like a cheat, but at least the full system will have been tested together and, presumably, at least function properly.
Sign in to Reply
RWatkins
4/10/2012 11:38 AM EDT
One kind of sneaky question: Why, if the receiver noise is in fact a problem for measurements taken in accordance with test directives, do test directives still specify an unrealistic measurement bandwidth? It seems that the EMI test suite should be set for a bandwidth comparable to the expected receivers that would be in use for the frequency band being tested. If this was properly implemented, then there would not be a question of sneaky interference from "certified" or "verified" devices.
Sign in to Reply