A cheap way to perform Conducted emission tests in-house ?

[Ashwin619]

Hi,

We had recently performed a conducted EMI emission test (as per CISPR 32) from one of the accredited Labs and we have found out that one of the DC to DC converters was not complying with it, as its emission magnitude was almost twice that of the ones specified under class B of the standard.
So now we are planning to perform an in-house conduction emission test so that we can do some changes in the design and somehow pass the standard.
The test setup needs to be cheap and quick to implement (accuracy and precision are not a priority). I am planning to make a DIY LISN for this purpose as there is quite some content online on how to make it.
However we don't have a spectrum analyzer and the closest one that I can think of is the FFT analysis of our DSO (Siglent SDS1072), although I am not completely sure whether it can truly replicate the waveform of a basic SA.
From what I have read about the difference between a DSO with FFT and a SA, from here ->
https://electronics.stackexchange.com/questions/50581/oscilloscope-with-fft-or-a-spectrum-analyzer
 
it seems that maybe I can get away with a regular FFT from a DSO to get some ballpark figures based on which then I can then make some changes to the input filter and wiring.

I have seen some cheap spectrum analyzers like these which seems like a really good deal for a SA

https://tinysa.org/wiki/pmwiki.php?n=Main.HomePage

I would appreciate it if someone could suggest to me whether the above setup would suffice or not in order to get a basic idea about the conducted EMI emission from a system.

Regards,
Ashwin619.   
 

[Miyuki]

Just take one of that cheap Chinese USB spectrum analyzers
They can do a reasonable job for a price point. Just need to check what model you are taking, some have issues or need some tweaks.
It is a relatively simple device using off the shelf components 
I'm also thinking about taking one for this purpose
Like this one, I do not know if this one particular is good, but some of similar one

[Ashwin619]

Hi Miyuki,

I had a look at this LTDZ 35-4400M device and it seems to do a good job for a spectrum analyzer, however, the frequency scan range seems to be quite high (35Mhz - 4.4Ghz) which is out of my range of interest. I believe most of the CISPR standards for conducted emission testing are defined for a frequency range of somewhere (150kHz - 30Mhz).
The tinySA seems like a good option for this range.
Also, what do you think about the FFT feature of a DSO, are they cut out for such a task.

Regards,
Ashwin619 

[erikka]

The tinySA is used by various organizations for pre-compliance testing.
The problem with a DSO could be the often 8 bit ADC at the input which limits the dynamic range in the FFT

[wilhe_jo]

The problem is, without the correct detectors, any measurement can fool you.

Cispr uses the cispr average (AV) and cispr quasi-peak (QP).
If your disturbance does not have QP=AV, it is very likely that you see the disturbance way higher with the cheap peak-only units. In this case it gets very difficult to judge if you'll meet the limit.

Try to get an SA with proper QP.

I tried to implement a QP on a hackrf, but the dynamic and accuracy of the hardware wasn't impressive so I let it go after the proof-of-concept.

73

[trobbins]

The OP does have a golden sample to allow comparison measurements.

[wilhe_jo]

Well, even with the golden sample approach, it's not like "10dB lower Pk = 10dB lower AVG"!

At least in general!

The same goes for the bandwidth options...
You want something which reasonably represents the measurement setup in the lab.

You can, of course, do it very cheaply...
eg. you can just tie + and - with 2 caps + 2x 200Ohms + together  and connect it to your 50Ohms "signal visualizer".
This is actually described in CISPR16 as an alternative measurement method as well.

I've done precompliance EMC on a fancy new high-res Lecroy using the resistor-method.
If that is what you have available, go for it!
To be honest, this was a stupid idea we had while having some beer which turned out to be actually a viable approach.
I'd say 10bits are fine for Class A but may be a little low for Class B stuff...

However, if you have the DC/DC measured using some LISN (and there are plenty of types as well!), that purely resistive approach will get slightly different results.

He's just 2x the limit (+6dB).
IIRC, conducted emissions have an uncertainty budget of around 3 or 4dB in an accredited environment.
This means, in another measurement, he could be between 2-10dB over limit, just by the different test equipment and physical setup.

The more you deviate from the standards, the more you need to do over-engineering.
This may easily work out for a 2W DC/DC, but could be quite some hassle for 20kW

That's why I got some old Analyzers with proper QP and CISPR Bandwidth for my lab.
YMMV.

BTW, if you look at the rigols, siglents, gw-insteks, they all seem to have what would be required.
For more special reasons (radio stuff), I got me some reasonably cheap, old, heavy monsters instead of a new budget option.

73

[cdev]

I would try out a HackRF ($150) which is a passable receiver up to around 6 GHz. Also the LPDA antennas sold for location of RFI work well, and can help you locate the source, so can some homemade RFI probes.

They could give you a decent head start. Spending little for this little bit of insight.

Hi,

We had recently performed a conducted EMI emission test (as per CISPR 32) from one of the accredited Labs and we have found out that one of the DC to DC converters was not complying with it, as its emission magnitude was almost twice that of the ones specified under class B of the standard.
So now we are planning to perform an in-house conduction emission test so that we can do some changes in the design and somehow pass the standard.
The test setup needs to be cheap and quick to implement (accuracy and precision are not a priority). I am planning to make a DIY LISN for this purpose as there is quite some content online on how to make it.
However we don't have a spectrum analyzer and the closest one that I can think of is the FFT analysis of our DSO (Siglent SDS1072), although I am not completely sure whether it can truly replicate the waveform of a basic SA.
From what I have read about the difference between a DSO with FFT and a SA, from here ->
https://electronics.stackexchange.com/questions/50581/oscilloscope-with-fft-or-a-spectrum-analyzer
 
it seems that maybe I can get away with a regular FFT from a DSO to get some ballpark figures based on which then I can then make some changes to the input filter and wiring.

I have seen some cheap spectrum analyzers like these which seems like a really good deal for a SA

https://tinysa.org/wiki/pmwiki.php?n=Main.HomePage

I would appreciate it if someone could suggest to me whether the above setup would suffice or not in order to get a basic idea about the conducted EMI emission from a system.

Regards,
Ashwin619.   
 

Obviously its not going to comply with the requirements but it could minimize the money spent by making it so your actual hiring of an expensive lab was minimized as far as time. Because your time had been utilized when it cost less. First.

[strawberry]

FFT vs SA.
FFT for debugging good enough


could radiate known power level RF cal signal for to proof that EMC setup is working properly

[wilhe_jo]

Forget those bloody nearfield probes 95% of the time.

The results are asolutely not reproducable. Even if you mount them to a 3d printer to scan a pcb. BTDT!
You'll also enter the ghost hunting business if you rely on them.

Build yourself a RF current probe... Cables are the antennas for radiated and the exit-port for conducted emissions.

So troubleshooting both works with them.

These current probes are very versatile and with regard to reproducibility, they are between the resistor method and coupling networks.

Use nearfield probes  as your very last tool when you can't get a clue with all other options... To find your badly designed DC/DC on a PCB, a ground loop with your scope probe is good enough.

Btw I'm currently designing a very cheap diy current probe and calibration fixture... Just some ferrite, a short length of coax and a little bit of 3d printing...
I'm thinking of making it semi-open-source.

Regarding the hackrf, beware of the low input levels! That's an SDR.
So it's not built for huge dynamic ranges but for typical signal levels at an average antenna.

Where it shines is stuff like antenna characterization...

73

[trobbins]

The OP has clearly indicated a need for conducted, not radiated testing.   OP also needs a quick assessment of whether specific design changes make any measurable difference.  As long as the OP can clone a reasonable LISN and noise floor, and take practical efforts to suppress other external influences, and get reasonable measurement comparison with his golden sample and with other sample switchmodes as a base to start from, then step modification and testing of the other samples is imho a very quick and rudimentary path to take.  There will always be an element of risk in formal retesting, but if the modifications are based on well appreciated step changes that are backed up by scope measurement as well as conducted EMI plots that support the theory and scope measurements then that is a pretty reasonable in-house effort.

[strawberry]

OP asked about DSO (Siglent SDS1072) vs SA. doesnt matter what sensor is used, I guess.
some SA or DSO come with built in EMC features
SA is easy to blowup 

[Ashwin619]

A big thanks to everyone for providing such amazing insights into my problem.

My main objective is just to replicate the result of the CE test performed by the LAB with as limited resources as possible. Again accuracy is not a priority here. I agree with trobbins wrt to the fact that I do have some data to compare with my results.

I am planning to purchase the tinySA along with some signal attenuators to perform the CE test. I guess it should suffice for my requirements.

I am also considering making a DIY LISN like this one ->



Also, I am assuming that since my device is a battery-powered one, the requirements for the LISN shouldn't be much complicated. Some TVS diodes along with signal attenuators at the output of LISN shall be enough to protect the SA,

Meanwhile, I have just performed an FFT analysis of our converter by simply placing the positive and negative of the probe to the input caps with an x10 on my probe (I believe this may not be the right approach, thus any correction here is appreciated). The results are below ->







The result is in dBVrms which I first converted to Vrms and then to dBuV. The sampling frequency is at 12.5 MSa/sec.
I am sure this is not exactly one of those quasi-peak (QP) or average values of the emission amplitude but I guess it's a relatively good estimate.
The QP value of the peak at the switching frequency (i.e. 175KHz) was around 95 dBuV for the LAB tested one. Through my FFT analysis, I got a value of around 87.81 dBuV.

I'll try to perform a similar measurement once I receive the TinySA along with the LISN.
I would also like to add that I am quite new to this testing and validation thing so hopefully you guys will correct me if I have done something wrong 

Regards,
Ashwin619.


 

[Jay_Diddy_B]

Hi,
Here is my contribution to this topic. I will revisit some of the suggestions made by other posters.

LISN(s)

To measure conducted noise you will need one or two LISNs. Depending on the EMC specification you will need either a 5uH or a 50uH LISN(s). These can be made cheaply.

I shared my single LISN in this message:

https://www.eevblog.com/forum/projects/5uh-lisn-for-spectrum-analyzer-emcemi-work/msg404662/#msg404662



and my double LISN in this message:



https://www.eevblog.com/forum/projects/5uh-lisn-for-spectrum-analyzer-emcemi-work/msg3330352/#msg3330352

These LISNs have 10dB attenuators and transient limiters built in to protect the spectrum analyzer.

Spectrum Analyzer

Most of the time frequencies of interest are between 150kHz and 30 MHz. You may need to go to 108MHz (the FM radio band). Some software defined radios start at 30MHz, making them unsuitable for conducted EMC measurements.

The filter bandwidth required is 9kHz. It is 9kHz because this is the width of AM band broadcast channel. Spectrum analyzers may have an EMC bandwidth (RBW) setting of 9kHz. If they don't use the 10kHz setting. There is very little difference between using 9kHz or 10kHz.

Quasi Peak (QP)



Measuring receivers were widely used for EMC measurements until spectrum analyzers became affordable. These measuring receivers had analog meter to measure the signal level. The analog meter had a specified response time.



It should be noted that:

Average <= QP <= Peak

and that if the Emissions are continuous, like those from a switching power supply:

Average = QP = Peak

You only benefit from QP if the emissions are intermittent at a low frequency and low duty-cycle.

Regards,
Jay_Diddy_B

[Jay_Diddy_B]

Hi,

To convert from dBV to dBuV you simply add 120

-40dBv = 80dBuV

conversion factor is

 = 20 x Log 1E6 = 120

for compliance testing you need to measure around 50dBuV

60dBuV = 1mV
50dBuV = 316uV

You will probably in the oscilloscope's noise.

With your x10 probe you would only have 31.6uV at the oscilloscope input.

Jay_Diddy_B

[wilhe_jo]

To measure conducted noise you will need one or two LISNs. Depending on the EMC specification you will need either a 5uH or a 50uH LISN(s). These can be made cheaply.

Nope, dc/dc converters in CISPR32 should be measured at 150Ohms.
That's done with a 150ohms CDN or simmilar, a current probe, the mentioned 100-resistor+50-ohm analyzer methods,... .

You can, of course use any other stabilizing method...but as I explained, the further you move away from the labs test-setup, the more overengineering you will need to consider...

Also, I am assuming that since my device is a battery-powered one, the requirements for the LISN shouldn't be much complicated. Some TVS diodes along with signal attenuators at the output of LISN shall be enough to protect the SA,

Meanwhile, I have just performed an FFT analysis of our converter by simply placing the positive and negative of the probe to the input caps with an x10 on my probe (I believe this may not be the right approach, thus any correction here is appreciated).

There is not "the" CE-test. there are various and there are various limits which require you to use the corresponding test instrumentation.

LISN is in most instances the wrong thing to use for dc/dc's... this is most of the time the thing you use for a.c port measurements... wrong common-mode/diff-mode impedance!

Your aproach using the scope-probes isn't too bad actually... connect + and - at the input with 2x + 200ohms + maybe 100nF-1uF (you want the cap be low impedance from 150k-30MHz so that the resistors define the overall impedance at you input).
I tried to sketch this in ascii art...

Code: [Select]

                    │ │
                    │ │                                                                                │ │
                    │ │      ┌─────────────────────┐                ┌──────────────────────┐           │ │
     +   ───────────┤ ├──────┤       200           ├──────┬─────────┤         200          ├───────────┤ ├───────────── -
                    │ │      └─────────────────────┘      │         └──────────────────────┘           │ │
                    │ │                                   │                                            │ │
                    │ │                                   │
                                                          │
                                                          │
                                                          │
                                                         ┌┴─┐
                                                         │  │
                                                         │  │
                                                         │  │
                                                         │  │ 50
                                                         │  │
                                                         │  │
                                                         └┬─┘
                                                          │
                                                          │
                                                   ───────┴───────

                                                      ──────────

From the center of these resistors,  place 50 ohms to ground (earth, or whatever you want to call it). At this 50 ohms resistor, you get your disturbance reading.

Next, you need to make your DC lines pretty high impedance (for the common-mode). I'd suggest (if you don't have an network analyzer) to add more and more clamp on-ferrites until the peaks don't change anymore... You'd want around 1.5kohms at around 1MHz IIRC...


73

[wilhe_jo]

Little more information: I just studied again some aspects of CISPR32...

Maybe the LISN confusion comes because of the definitions.
Basically, they say that a dc-port (where you plug-in some ac/dc adapter) should be treated as an ac-port.

However, for the methodology it clearly refers to cispr16-2-1 which calls for 150ohms for low voltage dc and the common LISNs for low voltage ac.
This is in-deed a little ambiguous and confusion if you don't have the possibility to read both document.
Especially wrt. the limits to be used...

Maybe this should be clarified.
I'll take some note and try to remember this when I comment on the next edition...

73

[Ashwin619]

Hello Jay_Diddy_B,

Thanks for your contribution. Ill try to implement a DIY LISN for my application. Will refer to your circuit if needed     

[Ashwin619]

Hi wilhe_jo,

Thanks for your wonderful insights regarding the need for LISN and the input stabilizing circuit. Ill try to attach this RC network across the input and then let you know the results of the FFT.

Regards,
Ashwin619.

[ademuri]

Ashwin619, did you try the TinySA? Any other updates you can share?