Scope vs Test Receiver - Conducted Emissions

[Glenn0010]

Hi all,

I'm working from two sites far apart on a University Project. A company's premises and a University.

I need to work from the university mainly and they don't have a test receiver available there. I'm trying figure out whether I can replicate the CE setup from the company at the Uni.

I am trying to validate weather I can take the time domain measurement off the LISN using and scope and then doing an FFT and converting that to dBuV to get the same results as the receiver.

Here are my results currently. In blue is the test receiver Average measurement (QP measurement is higher across the spectra obviously), in orange is the scope measurement with an FFT taken over a 20 ms window at 80 mV/div using 50R termination.



I have a few questions as I'm not sure how a test receiver does AVG and QP measurements. I'm guessing that since I am using a 20 ms window (one fundamental cycle) for the FFT on the scope, this is kind of an AVG hence why I am comparing it to the AVG on the receiver. Is this a valid assumption?

One thing which I am not sure about is also why I'm tending to be getting lower measurements at the HF end of the spectra using the scope. I'm wondering whether because I'm taking the fft from a large period (20 ms), my current scope gets limited to 10 ns/pt which means I could potentially miss the HF peaks leading to a lower measurement?. Or whether I'm operating close to the noise floor on the scope (some points <0 dB) due to using a (10 dB attenuator/pulse limiter) to protect the front end of the scope. However, the same thing is also used on the receiver.

If anyone could critique my setups and help me understand where the variations might come from regarding the two measurements it would be great.

Cheers

Glenn

[wilhe_jo]

Just looked up the uncertainty numbers from CISPR 16-4-2... for measurements with the LISN, they suggest Ucispr=2.9dB.
I'd say you're already pretty spot-on.

People from Gauss Instruments did quite a lot of scientific papers on this topic... maybe you could start with this one: https://mediatum.ub.tum.de/doc/1163324/573573.pdf

Regards

[gf]

1) What is the bandwidth and frequency response of the scope? Naturally, the frequency response affects your measurements.

2) Did you use a flattop window function? Other window functions may suffer from more or less scalloping loss.

3) The power at 30 MHz is still quite high. What is the signal you are trying to measure? Is the signal bandwidth-limited? If this is a fast comb generator, then frequencies may extend into the GHz range. But at 100 MSa/s, Nyquist is only 50 MHz. So likely you do suffer from aliasing. Is your scope really limited to 100 MSa/s? Which scope model?

[Glenn0010]

Hi, Thanks for the pointer, I read your reference and a few others, looks like it can also be the detector being used which can bring a change in the overall results. I've run 10 different repetitions and the data is very repeatable so am hoping at the very least they will be more than good enough for relative tests.

[Glenn0010]

Hi gf,

1) the BW of the scope is 3.5 GHz, I have a 6 GHz one available at the university which will hopefully improve things. Haven't look at the frequency response of the scope yet.
2) I've used a gaussion window function, to be honest I didn't look much into this before. But because of your comment, I've now looked into it. As I can understand the flat top window should give a more accurate, representation of the  amplitude, which is what I want.
3) In the images above, I'm measuring the emissions from a wide band gap VSD without any filters as a base line (measurement across a LISN). When I try to set the window to 20 ms, the scope limits my sampling rate to 100 Ms/s, in reality the scope can do much up in the Gs/s The scope at work is a Tektronix DPO 7254. At Uni I'll have access to one of the new MSO 6 series which should have better specs.

Any feedback would be great thanks.

Best regards

Glenn

edit : typo, dpo 7254 not dpr

[wilhe_jo]

Your limiting factor will be the resolution... in hi-res mode, you'll only get 11bit (just looked at the specs of your scope). That'll be just about adequate....

Quasi-Peak and AVG will be much more stable than PK.

When I emulated AVG the last time, I just took about 1s of data, did a FFT. That'll give you an average over this time.
To get the desired frequency resolution, I just added up the power in the bins...

For PK, I usually just take enough data to get the desired bandwidth considering the sampling rate.
I then step through that same 1s of data (with 50% overlap) and do a "max-hold" on the bins.

That's a little crude, but was pretty much ok for my purposes.


QP isn't normally what I'm after for these pre-pre-pre-compliance measurements... If QP and PK is very different, you'll better have a look on your control-loops,...
For the "real" measurements I have several fully compliant receivers.

OTOH, the math in the references is not too complicated...
If you have some time and some reasonably decent CPU, it should run pretty much real-time for the amount of data you can stream from the scope

73

[Glenn0010]

At the uni, when I will be using their scope and MSO 6 series, I have 12 bit resolution so that should help. Would you recommend using high res mode instead of sample?

I have all the time domain data along with the FFT from the scope so when I will have some time I can run those equations in the paper and see if I can get something that more approximates the receiver.

Cheers

[gf]

The first question is IMO: What's the actual occupied bandwidth of the given signal. This determines what minimum sample rate is required in order to capture the signal without aliasing. So I would start with a FFT at maximum sample rate in order to find out.

[Glenn0010]

Hi,

Today is my last day at the company site so will have to continue work at the university on the better university scope. I will try some of your more recent suggestions.

I have tried testing the flattop (called "flattop2" on DPO scope) window instead of gaussian. Seems to give better results.

Here you can see the difference. Note that the receiver is not a full scan but the avg pre scan. there is a gap of around 5 dB roughly.



I also analyzed 10 receiver pre scan runs, found that there is a variance of around 4-6dB across the whole spectrum on average between the maximum and minimum points at any given frequency.
   

[gf]

Here you can see the difference. Note that the receiver is not a full scan but the avg pre scan. there is a gap of around 5 dB roughly.

Did you honor the power of the (discarded) negative freqencies?
If not, this explains 3dB of the gap.

[Glenn0010]

Here you can see the difference. Note that the receiver is not a full scan but the avg pre scan. there is a gap of around 5 dB roughly.

Did you honor the power of the (discarded) negative freqencies?
If not, this explains 3dB of the gap.

Hi Gf, to be honest I don't really understand what you mean by this, maybe I did not pay enough attention in FFT lecture in school. Could you explain this a bit futher/point me to a resource?

[gf]

I suggest to read this one: https://holometer.fnal.gov/GH_FFT.pdf

[Glenn0010]

I suggest to read this one: https://holometer.fnal.gov/GH_FFT.pdf

Hi Gf, thanks so much for this. I'll look into it next week and will try it. The FFt was done onboard the scope but I have the time domain data so can still make this direct comparison

[gf]

The FFt was done onboard the scope

Built-in scope FFT ususally displays the power spectrum (not the raw complex Fourier spectrum, which spans positive and negative frequencies). But I don't know if the same applies to the frequency domain data you can download from this particular scope.

[Glenn0010]

Using the time domain data, I've ran an FFT using matlab and accounted for the power in the negative frequencies and it gave the same results as before. So I'm not sure where the variance is coming from.

This is the time domain data (20 ms)



This is matlab fft vs receiver



Maybe its due to the sampling time or the attenuator is reducing my SNR. Am not sure what else it could be