Saturday, May 1, 2004
System Design: Frugal Testing For Interference
RECENTLY, SOME FRIENDS SENT ME AN SOS e-mail about interference that was being detected from a little automotive microprocessor device they had been shipping for years with no problems. They couldn’t detect the interference with their new spectrum analyzer and a short monopole probe. But the end customer was quite vocal about the strength of the interference in the company’s new automotive FM/stereo systems.
The situation was a classic electromagnetic interference (EMI) problem, so its solution is relevant and applicable to the avionics world, as well. The universal first steps are always to define the “victim” system, and to measure and classify the emissions from the “offender.”
The victim system was a new Japanese FM radio, which has a different bandsplit from North American systems, reaching down to 75 MHz (marker beacon frequency, just for interest) and up to 95 MHz. The product that created the problem had only a 3.58-MHz crystal in the microcontroller circuit, so no obvious relationship seemed to exist.
What was going on and how did the interference present itself? The microcontroller was doing almost nothing and had no outboard memory or other bus-related subsystems to explain the significant emissions. So what was the problem?
Monitoring the offending unit with a simple monopole or other e-field antenna did not reveal anything useful, and ambient signals made resolution of the unwanted emissions difficult. Having run into this problem before, I used a small H-field loop antenna, and the offending emissions suddenly exploded into sharp relief on the spectrum analyzer. The test antenna is not hard to make. It’s just two turns of insulated, 22-gauge wire wrapped around a convenient half-inch (1.3-cm)-diameter screwdriver for shaping, and then soldered to a female BNC (UG-1094A/U) connector. I put heatshrink over the antenna to avoid deforming the loop and altering measurements between tests. This lets me connect the antenna easily to an 18-inch (46-cm), 50-ohm coax cable, and then to the spectrum analyzer without significant variation between tests.
The loop sensor has a wide bandwidth (to 1 GHz) and, with its tight focus, is able to pinpoint problems. Running the loop antenna over the problem unit (in a plastic housing) showed that the emissions were right at the clock oscillator. Emissions also were present on the power and control wires leading to and from the offending unit, even though that wiring was well bypassed, according to the schematic. The spectral content was different on the wiring, with little emission below 50 MHz. (The bypassing was actually partially effective.) But there was still significant emission above 50 MHz, our victim system’s problem area.
The test loop probe also picks up lots of ambient radio frequency (RF) noise, just like any dipole or monopole. To really cut off such pickup, I long ago made a small chamber out of an unwanted 19-inch (48-cm), rack-mounted piece of test gear. I made the rear panel hinged and added three antennas to the inner surface for e-field pickup. I can throw in any small offending object and either sample with a loop at any point or monitor the object via the antennas on each side. It’s a useful and cheap way to troll for problems without spending $10,000 for a shielded chamber and giving up 100 square feet (9.3 square meters) of unavailable floor space.
The currents in the microprocessor’s clock circuit were really the problem. They were square waves, with extensive harmonic content, reaching to well over 100 MHz with significant energy. This was a low-power microcontroller, consuming only a few milliamps at 5 volts DC, but the victim effects were significant, a common RF situation.
A second test unit, using a resonator instead of a crystal, produced sine waves at the external measurement points. I expected a reduction in emissions. But tests showed it was every bit as bad an emitter since the internal circuitry squared up that sine wave, although the distributed harmonic relationships were different. A final test unit could go into sleep mode on command, shutting off the system clock oscillator. The emissions vanished as soon as that unit went into sleep mode.
Just to be sure the emissions came from the test unit and had a strong e-field component, I wrapped it in tinfoil. As expected, all emissions vanished all over the housing. But the wiring still had interference, indicating that the power supply bypassing was inadequate to quench the clock artifacts.
The key elements of this problem were now clear. The microcontroller clock was the problem, belching out harmonics well into the discomfort zone of our victim radio and disturbing the power supply rails enough to reradiate back through them. It was evident from the PC board layout that the local bypass had a long lead length back to the critical points and did not do a good job suppressing current spikes internally on the 5-volt bus. A series choke also would have helped to decouple the interference from the outside world. An important lesson to take from this example is that the a surface examination seemed to show a well-designed circuit, with no obvious relationship to the victim system.
This problem for the end customer was actually solved in software by making the unit sleep when not needed. Testing clearly showed that sleep mode is a zero-emission state. In the avionics world, this system probably would need to do something, so that technique wouldn’t help. Simple case shielding was shown to be 100 percent effective in suppressing both E- and H- field emissions, so that would be an essential part of a viable avionics solution. The microcontroller bypassing and power supply decoupling could easily be improved to stop the conducted spurious emissions back through the power supply and control lines, and could easily be checked for success. That would have been the basic hardware game plan.
A useful near-field probe costs only a few dollars, and many others can be made or bought. A simple chamber represents maybe $100 in time and effort, and a scrap unit of suitable size could serve as the case. Current spectrum analyzers are expensive, anywhere from $5,000 for a used one to $60,000 or more for a new one. But searching for interference is really not a Cadillac situation; often a used Ford pickup version will do an excellent job for about $500 to $1,000. I use an older but still effective HP 141T system with different plug-ins (HP 8552B, 8553A, 8554A, 8555A) to cover a full sweep from 11 MHz to greater than 10 GHz. Add a few lucky finds like a tracking generator or two and a tracking preselector, and the result is a full-blown swept network analysis system with emission tracking into deep microwaves, for very little money. It was no small irony in this case that the new system did not find it, but the old beast did.
Can’t afford a spectrum analyzer? A good broadband radio also will work, but it can be more time-consuming to tune to look for problems. Keep in mind, the original mil-specs called for a noise receiver to do the emission tests. You can buy the same detection functionality, although uncalibrated, for maybe $100 at Radio Shack.
Fixes and techniques can take many forms, and the solutions are as varied as are engineers and designers. But the first step is always being able to find the problem, and the final step to confirm it is actually eliminated. All the rest is art—and talent.
Walter Shawlee 2 may be reached by e-mail at firstname.lastname@example.org.