Fixing a product that is failing CE radiated emissions

[3Dogs]

Hi all.
This is my first post here, so please be kind.

I am a designer with 30 years of experience, mainly doing embedded controllers and some mixed-signal designs. I designed a very high-current motor control board for a battery-operated lift, which has been quite successful in terms of operation and reliability. At the time the design was initially started, FCC and CE testing were not a concern. Several years after the design was put into production, the product went through FCC testing, and with some tweaking (ferrite cores on the motor leads, and lowering the processor clock frequency), it passed.
Now the company is trying to get CE approval, and they have introduced another complication - they added a wired remote control tied into the existing front-panel control switches. The unit is failing CE radiated emissions, mostly in the sub-70MHz range. When the “pendant” is stretched out, the radiated emissions are much, much worse. I have been called back to see if i can help them fix this issue.
The first thing I want to do is to determine whether the radiation is coming from the H-Bridge FETs switching, or (as I am thinking), the broad-band noise from the brushed DC motor at 50-70 amps. The reason I don’t think it’s the FETs is: I modified a board with larger gate resistors and confirmed that the switching time was increased from 200nS to 500nS or so, and when the board was re-tested, they didn’t see any significant change.

I am proposing buying a Spectrum Analyzer and doing some basic scans to see if i can pinpoint the area where the noise is coming from, and when I do, using the SA to measure the effectiveness of any fixes I try. I have a number of questions about this (keep in mid that I am on a shoestring budget, under $2,000):

1) What SA to buy? I’m looking at the Siglent SSA3021X. If I’m going to spend the money on a new piece of test equipment, I’d like to buy the most capable machine. I’d like something that can measure above 900MHz, since I’m working with some ISM-band radios. I don’t know if it makes sense to get an SA that goes to 3GHz or more. There are some used machines on ebay that look good, but some of them are physically enormous (like an R&S FSB), and I’m concerned about buying someone else’s problem. What about the Anritsu MS2711D? There are several on ebay with warrantees, and they are cheaper that the Siglent.

2) What antenna to use? I’m not interested in spending a fortune for a bi-conic antenna. Also, those nice PCB antennas are all made for much higher frequencies than I’m concerned about. Should I just bare a piece of Coax? I don’t care about calibration, since I’m going to be making relative measurements.

3) How about near-field probes?

4) Can the Siglent still be unlocked? I’m on a budget, and having the EMC option would be nice.

I appreciate any insight into this that anyone can share.

[duak]

I can't offer any insight into a choice of a spectrum analyzer; however, if you can't measure something it's harder to make quantitive decisions.  A scope and some probes are just not good enough.  I remember that at the test range or chamber each scan took about 10 minutes wiith an hp analyzer and that there was something called "pseudo peak" but for the life of me, I can't remember exactly what that is.

I shepherded two products through the compliance process.  When bringing internal signals out of an EMI tight enclosure, ferrites are not always sufficient.  One product required replacing a twistlock bulkhead connector with a connector that mounted on a PCB containing 1n0 capacitors from each signal to the enclosure.  These caps bypassed much of the RF energy to ground right at the periphery.  These signals were DC power so a bit of capacitance didn't matter.

Best o' luck

[Berni]

Well for EMI precompliance testing you do need quite a bit of gear like a SA with a quasipeak detector, an EMI chaimber, calibrated wideband antenna etc..

For for just tracking down the problem you can use pretty much any spectrum analyzer and just stick on an antenna that's reasonably close to what you want to pick up since while antennas are tuned for specified frequencies the thing i they still work for everything around that, they are just not as sensitive. So you just want a SA with a reasonably low noise floor (so that you don't need much in the way of amplifiers to boost the signal out of the noise floor) that covers the frequencies you want, getting to 3GHz is nice so that it also covers the popular 2.4GHz band.

Its also helpful to use small coils as antennas since that lets you scan over around the DUT to precisely pinpoint the source.

And yes EMI can be pretty stubborn and get trough all sorts of filtering.

[TheUnnamedNewbie]

Have you considered getting outside help? A consultant that specializes in these kinds of things could cost you a lot less than the instruments needed and the engineer-hours it takes for people that are not experienced with this kind of stuff to try and learn and do a good job. Not saying that you shouldn't try, just that it might be more economical to bring in outside help.

[Pinkus]

Yes, the SSA3021x is fine for this and it still can be 'tweaked' afaik.
Near field probes are helpful to narrow down the search to specific areas / parts of the PCB, thus you will need them if you need to find a problem on the PCB. If the cause is the just motor, a regular antenna (piece of wire) would show you what you need to see. So you may skip purchasing the probes at the beginning but if the time is critical, purchase them together with the SA.

What I would do next:
1) Purchase a spectrum analyzer and a set of near field probes (or build one yourself - Google will help you).
2) Set up a test area where you can reproduce the measurements.
3) remove the DC motor and check again.

If the problem is gone:
4) replace the DC motor by some power resistors (a bunch of halogen light bulbs is easy to get/handle) and check again

If the problem is still existent, you will need to check the H-bridge in detail and make changes. This is where you need the near field probes the first time.
If the cause is the motor, you will need to add ferrits close to the motor to block this frequency. Maybe it would be enough to wrap each engine power line around a ferrit core a few time. Though I am wondering if such tests (using ferrits) were already done at the test facility (usually this should be the first the test engineer should propose/do and run a re-test).

Following this procedure, you can quickly nail down the problem.

As I understand, that not you but your employer or customer will pay the spectrum analyzer, go for it: having more testgear around is always helpful 

[AndyC_772]

To be radiating at <70 MHz there must be quite a long antenna, no doubt the cable joining the controller to the remote.

A pragmatic fix might just be to use a shielded cable, if it's not already. For "shielded", read "weapons-grade braid & foil shielded" if necessary; the engineers at your EMC lab may well be able to recommend something that they know to be highly effective.

Also make sure the shield is very well connected to the earthed metal box around your controller. (In this case, "well" implies a very solid, low inductance connection that doesn't involve using the drain wire, a connector pin, or other small, thin conductor).

For such a relatively low frequency I wouldn't personally go out to buy an SA straight away. Instead I'd just grab an oscilloscope, make a simple loop by connecting the ground clip of a probe to its tip, then wave that around the circuit. If something is generating a lot of interference, you'll see it.

[Benta]

I'm with UnnamedNewbie
Team up with a test lab, it's less expensive than you think and they know all the pitfalls and will have good suggestions and remedies. Also, they'll provide you with a test report, this is really important for your CE documentation (to keep you out of trouble).

[3Dogs]

Thanks for the suggestions.
I’m still leaning towards doing this myself, since I think it will be useful experience, and more convenient.

I’m planning on testing the motor by connecting it directly to the battery - this should tell me if it’s the brushes making the noise.

Any thoughts on what Spectrum Analyzer to get?

[jmelson]

The first thing I want to do is to determine whether the radiation is coming from the H-Bridge FETs switching, or (as I am thinking), the broad-band noise from the brushed DC motor at 50-70 amps. The reason I don’t think it’s the FETs is: I modified a board with larger gate resistors and confirmed that the switching time was increased from 200nS to 500nS or so, and when the board was re-tested, they didn’t see any significant change.

First test, does it fail when not lifting the load?  Or, does it only fail when the motor is running?  That would come close to answering the question, assuming the motor drive is not idling at 50% duty cycle all the time.

Also, you could likely pull a fuse or something to disable the motor drive, and see if that affects the EMI.

Also, did the test lab see a REALLY broadband peak or a narrow one?  A motor is not likely to produce a narrow peak, unless really high-Q circuits are on the leads.

Jon

[3Dogs]

The first thing I want to do is to determine whether the radiation is coming from the H-Bridge FETs switching, or (as I am thinking), the broad-band noise from the brushed DC motor at 50-70 amps. The reason I don’t think it’s the FETs is: I modified a board with larger gate resistors and confirmed that the switching time was increased from 200nS to 500nS or so, and when the board was re-tested, they didn’t see any significant change.

First test, does it fail when not lifting the load?  Or, does it only fail when the motor is running?  That would come close to answering the question, assuming the motor drive is not idling at 50% duty cycle all the time.

Also, you could likely pull a fuse or something to disable the motor drive, and see if that affects the EMI.

Also, did the test lab see a REALLY broadband peak or a narrow one?  A motor is not likely to produce a narrow peak, unless really high-Q circuits are on the leads.

Jon

I wasn’t at the testing, so I am getting everything 3rd (or 4th) hand.
My understanding is that the machine only fails when operating. That doesn’t mean that much, though, other than the µP emissions are under control. Once the winch is activated, it gets PWM-modulated by the H-bridge, and quickly ramps up to it’s maximum duty cycle (around 90%, since the H-bridge high-side drivers need some time to recharge).

Once I get some time alone with the machine, I think I can quickly determine if it’s FET switching or brush noise that is causing the problems.

And, to answer you 2nd question, there is a broad area from 10MHz to 70MHz that is over spec, with a number of stronger peaks. But, it’s pretty clearly not just harmonics of some clock, since the peaks are poorly defined.

[AndyC_772]

Is your motor PWM signal derived from a crystal, or some other stable time base? If so, I'd expect the peaks to be quite sharp.

Broad band noise, especially at the bottom end of the range (say, 30-50 MHz), is almost always from a switched mode power supply.

[nctnico]

There is a simple approach to fixing EMC problems:
1) Get a graph from the test lab which shows the frequencies versus amplitude and preferably a list with frequencies which make the device fail. Note how much damping you need to pass and add 3dB to have some room.
2) Use an SA and H probes to find the part of the circuit where these frequencies are generated. I can recommend the H field probes from 'RF explorer'.
3) Note the levels (or make screenshots) and position of the probe. This is you baseline measurement
4) Figure out how to change the circuit so the frequencies causing the fail are not fed to the outside. This is the hard part. Note that an electric current always flows in a circle!
5) Measure the levels with the modifications in place.
6) Repeat from 4 if you don't get enough damping.

[3Dogs]

I decided to bite the bullet and buy a used Anritsu MS2712E for $1199. It comes with a calibration certificate from about one year ago, and a 30 day warranty.
Assuming it is in good condition, it seems to be a somewhat more capable SA than the Siglent.

Based on the suggestions here, I also bought the Seeedstudio near field probes.

When all this stuff comes in, and I get access to the machine, I'll report back on what I find.

[max_torque]

If you need a short term fix, and you're confident it's the additional "aerial" of the remote control allowing the RF to leak out, then depending on what the coms between remote and the unit actually are, you may simply be able to low pass filter that additional wiring, using suitable ferrites and/or chokes.  If the data transfer is high bit rate, then things get trickier or course  (it may also be that data that is causing the RF issue!)