Ghost in the Machine, Part 2

We are back to exorcising the EMI demons that were the topic of the last weekıs article. Last time I left you with your face on the keyboard as you dozed off with thoughts of conducted EMI, radiated EMI, ground loops and juju beads. EMI can be a tough topic to follow. But follow it we will with even more hints and tips to deal with EMIıs ghost. Donıt forget to read my trouble shooting course Shotgun Wedding. Itıs a great help in locating the offending spirit.

Repeatability
Remember that EMI is caused by some sort of electromagnetic field either conducted or radiated. Often times this only occurs on an occasional basis. That in itself can make it hard to track down. So we will review the concept of repeatability. If you ever think you have solved a particular problem, you will need to remove the solution, and see if the problem comes back. Case in point: One time I was trying to eliminate a flickering problem on a display we were using. As I worked out what was going on I tried putting a ferrite on the wire harness. The problem went away. Thinking I had it solved, I instructed the production line to install ferrites on all the machines. You can probably guess what happened. Shortly after the line started up again, the flicker was back. I later discovered the problem was caused by motor brush arcing, I just happened to put the ferrite on when the motor brushes ıburned inı eliminating the noise source. Now I will always remove and reinstall the fix several times to be sure the problem returns and is eliminated consistently. The first thing I ask any engineer when he/she returns with a fix is did you remove it and make sure the problem is still there?

Controlled Environment
Every engineer knows the importance of a controlled environment to determine the validity of a test. Yet, I see this concern overlooked often when trying to track down an EMI problem. Maybe it is because EMI is so difficult to reproduce. There are some standard techniques for reproducing EMI in a test environment. If you have be dealing with the European CE requirements, you may be familiar with some of them, such as EN 61000-4-4. This references one test that I find particularly useful, the EFTBN test. It stands for Extremely Fast Transient Burst Noise. This is a great test for finding immunity problems with a given design. The history of it goes back to the ı60s and ı70s. Some IC based clocks that were being developed seemed to get inaccurate during use. No one ever really located the source of the noise, but they found if the clocks could past this test they developed, they kept time correctly. That eventually became the EFTBN test.

Now you can buy an EFT machine, but they are expensive. A cheap alternative that you might try (but its not as controlled or as safe) is to plug you equipment in, and in the same socket plug in a wire with a rusty file attached to one prong, and a wire with a bare end attached to the other prong. Now CAREFULLY rub the bare wire up and down the file. This is a poor manıs version of the showering arc test. (The showering arch test was used by UL for some time before it was replaced by the EFT test) I take no responsibility for injury caused by being so poor that you have to use the rusty file test.

Timing is Everything
Another item I have learned is to track down the sick circuit right when it is failing, often times you may leave it till you have time to research it. Then when you go looking you canıt find it cause itıs working now. You have to catch it in the act so to speak. So when it happens, donıt wait, grab your ıjuju kitı and go ghost hunting. Donıt be surprised if something happens on the production line that you canıt get to repeat in the lab, go to the line and try to figure it out. Amazing amounts of noise can be found on the production floor. (Especially if you have welders out there) Our production line once found a metal table that would mess up a CD player whenever it was within about 2 inches of the table surface. The table was grounded to a steel post holding the ceiling up. I learned you can have upwards of 50 volts of noise between ground in the outlet and the steel in a building that is tied to that ground. I tied the table to the outlet ground and the problem went away. (I made sure to repeat the problem and solution several times, as I had learned by then.)

Under Pressure
Sometimes we are under pressure to develop a solution fast. To do that you might try throwing everything youıve got at it at once. If you solve it, then try removing one piece at a time. EMI problems are often combinations of various things. And if you try on fix at a time you may overlook a combination of fixes that would have solved your dilemma. You might need that 0.1uf cap on the AC line and the ferrite clamp on the data harness. As often as not, you will need more than one fix to solve the case.

Be Prepared for Surprises
An across the line AC cap will do great things to filter out noise coming into you system. Thatıs why the put them in surge suppressors. That was an absolute truth for me till just a few weeks ago when I was tracking down a noise problem on a communications harness. And I noticed something funny. I was observing the noise on the communication lines when I asked one of my engineers to plug the unit under test into a surge suppressor, instead of directly into the wall. The noise got worse. Iım still not sure why, but we used it to improve our filtering and the reliability of the data. Donıt make any assumptions. Test everything.

Not All Components are Created Equal
What is Xc for a 1uf cap and 0.01 uf cap at a frequency of 1 Mhz? Letıs see, Xc = 1/(2 * 3.14 * 10M * C), so multiply, cancel the exponents, mumble, mumble, grunt, grunt. You get 0.016 ohms and 1.6 ohms respectively. The larger cap should effectively short more noise to ground. Too bad this isnıt a perfect world, or that would be the case. Take a look at a regulator data book, what are the recommended capacitors? There are one large and one small one right? The reason is the larger capacitors often do not work like smaller caps at higher frequencies. A perfect cap would but alas, there are no perfect caps, only perfect calculations, sigh. K Hint, select a cap with a roll off close to the frequency you are trying to clamp. ı3 db is a good place to start.

One other thing, the capacitance printed on the case is only legitimate when used at the operating voltage on the case of the cap. The moral of the story, you may have the right component, but the wrong value. Nothing a little experimentation canıt solve.

I Dream of Juju
Experience is of great value in the battle against EMI. But you donıt have to learn all the courses the hard way. You can learn from otherıs mistakes. But I do suggest you document your reasons and conclusions for a given fix you have found, you will use it in the future. Like the suit guy says, ıI guarantee it!ı

Snore, snore, sniffle, snort! Oh yeah, till next time. Iım gonna take a break before I get ıqwertyı face.

Product Engineering Archive

Guides and Experts   Analog Avenue   EDA Tools   PLD   DSP   EDA   Embedded Systems   Power   Test

		Click here to get your listing up.