spread spectrum cheating (2023 Update)

You will not get the full 30dB improvement you need just by sweeping the clock, but you might get close if you go for an extreme sweep range. Assume that the maximum frequency deviation that you can use is +/-50% and that the measuring receiver has brick-wall filters:

Then the gain from sweeping is about 10*log(2*0.5*62/0.12) = 27dB

Check whether the results you have been given are for domestic (class B) or industrial (class A) environments. If the classification is wrong and you can use class A rather than class B you gain another 9.6dB, in which case the problem is solved!

John

The video bandwidth is also relevant. To get N dB improvement, your P-P sweep range needs to be > (10**N/10)*125 kHz, but your sweep rate can be slower by a factor (Video BW)/(Resolution BW).

It also matters what order the analyzer does things in. Most of us old timers are probably familiar with the problem of measuring noise with an average-reading AC voltmeter such as an HP 400EL whose dial is calibrated for sine wave input, where you have to add 1 dB to the reading to get the right answer.

Old-timey SAs and many scope FFTs take the log first and do the video BW filter afterwards. That makes the noise read low by 2.5 dB. (The HP/Agilent/Keysight app note AN150 is a super good read on this point, among many other things.)

Cheers

Phil Hobbs

Yeah, and it does work. But it won't be widely used until the patents run out. I bet that there are VFDs than can be programmed to vary the speed, if one knows the magic codes. I don't, and it wouldn't work with a big heavy chuck on an old manual lathe.

Joe Gwinn

30-33kHz modulation frequency seem to be standard. There are lots of ready-made SSC clock generators, and even 5*7mm-style crystal oscillators with SSC output.

For CPU clocks etc., I like the SI5351A clock generator (if you can get them since Skyworks bought the parts). You can either order them pre-programmed, or you can set the registers via I2C after each power-on (that even works on pre-programmed parts). PLLA can do SSC with configurable parameters, and

125MHz is in the range of supported outputs.

cu Michael

That's interesting. We could just drop one into our board. Cool.

We do have a PLL inside our FPGA that multiplies the 125 MHz clock up, to clock a 250 MHz ADC. A little jitter wouldn't bother me (the signal is grossly oversampled) but we need the PLL to still work.

Great. Thanks.

Why not convert the clock into a pseudo-random sequence and XOR that with the data. That way, both the clock and data will have a white noise spectrum (shaped with a sinc that has a zero at the clock frequency). Reverse the process at the other end. As you are using fpgas the extra overhead should be minimal. Alternatively, you could use Manchester coding which would achieve a similar result.

John

I've always wondered if frequency wobbling is a method to reduce the interference impact of the emissions, or if it is just a way to impact the measurement. The same amount of power is being emitted at a given bandwidth at the time the sweep passes that range, swept or not. I would image there are victim devices that would still be impacted in the same way, even with the frequency sweeping.

The customer has designed an elaborate controller (it took them a few years) and doesn't want to change it, but is failing EMI. We would prefer to not change our box much either, any time soon. It's an interesting political situation.

If we could change things, we could go 8b10b on the data lanes and have no clock pair in the cable. 8b10b is sort of spread-spectrum already, and tricks can make it better.

In that case, maybe you should revisit grounding the shield at both ends (with a series capacitor self resonant at 62MHz if they insist on the "ground loop" being broken). Also did you establish whether they were applying industrial (A) or light commercial/residential (B) emissions limits? It does seem as if they want you to jump over hurdles with both legs tied together.

John

Probably a dumb question, but those clock and data signals are source- terminated to drive ~100 ohms, right? If you have standing waves on the wire pairs, I can see them exciting the ungrounded shield at the current points, making a very effective antenna. Radiation from the cable could easily be 30 dB worse than expected.

-- john, KE5FX

torsdag den 3. november 2022 kl. 04.00.11 UTC+1 skrev John Larkin:

if it is balanced why do you need such a massive swing?

At least in mechanical systems, there aren't a lot of damping mechanisms at work, and resonances tend to be sharp. It's a real challenge to make a good shock absorber. Between ohmic and magnetic-hysteresis and dielectric losses, mainly a resonance in electric circuits is going to be lossy and broad (on the scale of the megahertz frequencies we're considering in test). I'm thinking that 'victim devices' won't run away if the noise source is broadband, OR... they're so sensitive at all frequencies that you just run into all the other noise sources first.

That presumes the victim box gets some shielding, and testing, before product release.

I've seen consumer devices that have to have good separation, or the WiFi box garbles the sound at the audio box; presumably that's magnetic coupling from power supplies.

The original design, about 10 years ago, used a big swing and a receive-end attenuator to give a lot of common-mode rejection against ground loops. This system is spread over floors of a big expensive building.

We were exempt from EMI standards then.

Both ends are terminated, so there should be no standing waves.

On a sunny day (Wed, 09 Nov 2022 12:41:42 -0800) it happened John Larkin <jlarkin@highland_atwork_technology.com> wrote in snipped-for-privacy@4ax.com:

Go optical?

...

Sure, standard trick in the arsenal. I can't give specific numbers, as those depend on how much above the limits you are, what kind of spreading you'll be using. One important thing to keep in mind is that many standards specify quasi peak detector with specific EMC bandwidths that are available in test receivers, but not that often in normal spectrum analyzers.

My most horrible work in the specman field was using slightly filtered PWM output to spread non-locked VCO used to downconvert a band limited signal to a log amp level controller. You should have seen the reaction of an RF engineer when he saw my LO implementation. But in EMC sense the thing was below detection limit in the lab!

-- mikko

It is a bit of both. If the radiated power at f0 is reduced by spreading it out a bit you might hit other resonances.

But a spectrum analyser is the worst case seeing all of it.

A 10 fold reduction at any single frequency might still be good enough to pass even if more frequencies are now affected by its influence.

Conservation of energy says that it has to go somewhere.

A 10-fold reduction isn't enough. JL needs to reduce the level by 30dB. The bandwidth of the EMC measurement receiver will almost certainly be

120kHz, so the frequency modulation would need to be extreme to achieve that by FM alone. The quasi-peak detector has an attack time of 1ms and a decay time of 550ms. With a uniform frequency sweep of +/-50% I estimate that the measurement will drop by about 27dB which is not enough.

John

I've seen some suggestions that we might pick up around 8 dB improvement in EMC qualification with a reasonable spread spectrum clock. Every 8 dB helps.