Conduct EMC Testing Economically In-House
To succeed in industry these days, most manufacturers have to maintain a strong position in the international marketplace. In Europe, that means products must qualify for the CE mark, which ensures compliance with all applicable European Union directives on electromagnetic compatibility (EMC). Designing for this stringent specification, which has been mandatory since January 1, 1996, requires knowledge about both design practices and test methods.
This article discusses equipment and methods electronics manufacturers can use to qualify their products. Included is an example of how Symbol Technologies, Holtsville, N.Y., tests its products, along with the practices the company uses to solve emissions problems. Several problems with test setup and configuration are also discussed. Finally, we will cover some possible surprises that engineers should be aware of when designing a product for EMI compliance.
From gun-shaped bar code scanners to hand-held PCs and wireless LANs, Symbol's products all have one thing in common: electronics. More specifically, they all incorporate embedded digital systems surrounded by analog circuitry. Thus, the products fall into the category of information technology equipment (ITE). The applicable standards include EN 55022 for radiated emissions and EN 61000-4-3 for radiated susceptibility. EN 55022 requires the use of an open area test site (OATS), a gigahertz transelectromagnetic (GTEM) cell, an anechoic chamber, or other alternate test setup to perform electromagnetic interference (EMI) testing.
For a number of reasons, Symbol chose to purchase a GTEM cell rather than build an OATS. A GTEM cell is immune to ambient noise conditions, and tests take up to 90% less time to run, making the system easier and faster to use. GTEM cells also perform susceptibility testing, something an OATS cannot accomplish. The equipment under test (EUT) fits inside the cell, a convenience not all companies have. Finally, the cost of GTEM cells versus an OATS is comparable, around $225,000, including the test equipment. Therefore, there's much more value for the dollar with a GTEM cell.
Despite all the advantages of a GTEM cell, Symbol is still considering the purchase of an OATS. One large disadvantage of a GTEM cell is its frequency limit. An OATS can measure beyond 25 GHz, while a GTEM cell peaks at 5 GHz. For most companies, this limit is fine. However, Symbol manufactures radio products that operate in the 2.4 GHz band, and approval of these radios requires testing up to the 10th harmonic.
The company's GTEM setup consists of an HP 8593E spectrum analyzer with a quasi-peak detector card, an Emco Boss manipulator, an HP 8648A RF signal generator, an IFI SMX-100 RF amplifier, and a PC. The entire setup is controlled by GTEM software running on the PC via the IEEE-488 (GPIB) interface (Fig. 1). The signal generator and amplifier are used for susceptibility testing and the spectrum analyzer is used for radiated emissions testing.
To perform an emissions test properly, the product (usually in prototype plastics) is securely mounted on the manipulator table with all the appropriate cables attached (Fig. 2). The test is run with the unit operated in a user-intended mode. For a bar code scanner, this means continuously scanning a bar code and transmitting the data to a host. To accomplish this, a pneumatic actuator pushes the trigger on the scanner, whose exit window aims at a bar code mounted a specified distance away. The power-supply cable plugs into an EMI-filtered ac outlet inside the chamber, and the data cable attaches to a host through a filtered or isolated connector.
Before the emissions test begins, the engineer enters specific standards information into the system's test software. This software is typically purchased from the manufacturer of the GTEM cell.
At the start of the testing, the manipulator rotates the unit to a 45° azimuth, 120° orthogonal angle to the ground plane inside the GTEM. The spectrum analyzer then sweeps the frequency range. When completed, the manipulator moves to a 45° azimuth, 0° orthogonal angle, and repeats the test. Finally, the manipulator rotates to a 45° azimuth, -120° orthogonal angle for the final radiated measurements. These positions, which are controlled by the software, are devised by the manufacturer of the GTEM cell to produce the desired test results.
The GTEM software then performs calculations that correlate the data to produce a plot of radiated emissions. The software uses many parameters during its correlation: the height of the EUT's center from the ground plane, the distance from the GTEM antenna to the EUT, the separation between the ground plane and the septum, and the distance between the EUT and the septum, just to name a few. By locating the EUT in these three positions and performing the calculations, Symbol has consistently shown a correlation between its GTEM and an OATS. The company can and does certify non-RF products in-house with this setup.
A good example of EMC design and test procedures is a project currently in the works: a cordless scanner that uses a base station for charging and data transfer. Regulations insist that a product be tested under worst-case conditions. One such configuration consists of the scanner mounted in the base, charging and transferring data. Running an emissions test on this system checks two interconnected microprocessors that transfer data between each other and a host. Charging adds to the emissions because the charging circuitry inside the product uses a switch-mode power supply. In order to obtain the worst case, a completely dead battery is used to draw the most charging current.
Symbol wants to certify the new scanner to the CISPR B standard in Europe. Although things could have been worse, a graph of the results for the initial prototypes of this product shows that the system did fail to meet the CISPR B limit at a few frequencies.
Without understanding why an electrical device radiates, an EMI problem like this is an engineer's worst nightmare. It means staying in the lab until midnight the whole week, trying everything and anything to get under the CISPR B line. Before cutting one pc-board trace or lifting one pin, however, a thorough analysis of the results is in order.
The thin spikes that jump over the CISPR B line are caused by a very distinct periodic signal somewhere in the system. We assumed that the peak at 44.2 MHz was the fourth harmonic of the 11.059-MHz crystal running the microprocessor in the base. We also assumed the peak at 73.7 MHz was the third harmonic of the 24.576-MHz crystal running the microprocessor in the scanner. The peak at 221.3 MHz is the ninth harmonic of the latter crystal.
Looking at the average emissions level across the spectrum reveals the broader resonances of the system. The resonance around 40 MHz is caused by cable effects, something we'll discuss later. The resonance that bothered us the most was the one around 140 MHz. It was very strong and almost exceeded the limit. Had the resonance not been there, those little peaks that exceed the limit would be way down at an acceptable level.
In order to locate the source of the 140-MHz noise, we used a device called EMScan. The EMScan system, from Amplifier Research, Souderton, Penn., provides a spatial view of a pc-board, showing emissions "hot spots." Suspecting that the noise originated from the scanner, we removed the pc-boards from the plastics, spread them flat on EMScan's bed of antennas, and plotted the results at 140 MHz. A hot spot was picked up dead center on the switching transistor inside the charging circuitry. We lifted the open-collector output pin on the chip and inserted a ferrite bead between the pin and the pc-board pad. Subsequent scans showed that the resonance at 140 MHz was completely gone.
To reduce the spikes at 73.7 MHz and 221.3 MHz, 33-Ω resistors were placed in series with the read, write, and three chip select signals going from the microprocessor to the ROM and RAM inside the scanner. This drastically reduced the overshoot and undershoot on these signals, lowering the harmonics that were showing up during the emissions test.
Removing the 44-MHz spike was not as easy. In the end, we had to turn another base pc-board using four layers rather than two. We simply added ground and VCC planes onto the already existing two-layer artwork, and made some improvements near the data connector.
The final product, tested in the same configuration as the prototype, met the specification. The addition of two plane layers in the base drastically reduced the 40 MHz spike, and the ferrite bead in the charging circuit removed the resonance at 140 MHz.
Keep in mind that these graphs show the peak amplitudes of electromagnetic radiation. Regulatory requirements specify their limits using a quasi-peak measurement. Most preliminary emissions tests use the peak measurement because it's faster, and provides enough information to troubleshoot any EMI problems.
Further, a system below regulatory limits during a peak test will be under the limits during quasi-peak. Normally, if a spike at a certain frequency is only a few dB over the limit line (during a peak measurement), the GTEM software is instructed to perform a quasi-peak measurement on only that frequency. If the offending signal is non-periodic, or a transient, the quasi-peak measurement can reduce the results by approximately 2 to 3 dB.
Symbol also uses its GTEM cell to perform susceptibility testing, with the EUT installed in the chamber in the same configuration used for emissions testing. However, an RF amplifier is connected to the GTEM antenna. The manipulator places the EUT in the first position (described earlier), and the test begins. An RF signal generator sweeps the frequency range, using an AM signal modulated to 80% with a 1-kHz signal. The GTEM is configured to simulate the antenna being placed 3 m from the EUT. The amplitude of the RF signal is adjusted at each frequency to keep a constant electric field around the EUT.
The objective is to expose the EUT to an electromagnetic field with a uniform power density across the entire spectrum. At each frequency, we operate the EUT (by pneumatic trigger) and check it for correct operation. The same test is performed in the remaining two manipulator positions. For certain products, a camera (connected by optical fiber) installed within the chamber allows the technician to view an LED or LCD during exposure to the electric field.
As mentioned earlier, an EMI problem can take a very long time to fix. Emissions and susceptibility test results depend not only on the equipment under test, but on the cables and power supplies. It's very important to use the exact data cables intended to for sale with the product. Testing with shielded cables and selling with unshielded ones is not a good idea.
The power supply also plays a large a role in the emissions results. Switching supplies bring their own headaches into the picture. A power supply with its own CE marking will decrease the probability of having a power supply-related emissions problem. Remember, when testing a product for European certification, the European version of the power supply must be used and powered by 220 V ac, 50 Hz. If your lab is in the U.S., watch out for EMI problems caused by the 220 V ac generator itself.
The types of power supplies and cables are not the only factors; their positions inside the GTEM or OATS also affect test results. Cables are huge antennas that amplify the noise within the product to which they're connected. As with any antenna, position and length (among other things) determine the resonance characteristics. We try to move the cables into a position that causes a worst-case emissions result. The cables are held in this position to maintain repeatability from test to test. Tests of another scanner illustrate this point.
At first, the results looked very bad, with plenty of failure points. However, an identical setup, but without the cable, indicated the true noise source was at 180 MHz, which is the 10th harmonic of the ALE in the digital system. The cable spread and amplified the 180-MHz noise across the spectrum, making the results look very bad. In both instances, the scanner was powered by a 9-V battery. Large resonances around 30 to 50 MHz are usually caused by cables. The noise at those frequencies often disappears when the cable is removed.
To determine if an emissions problem is actually a cable problem, try placing snap-on ferrites on the cables. The ferrites will shift the resonance point of the cables, producing a noticeable effect. The position of the ferrite on the cable is critical, and will have to be optimized to remove the problem. Some companies (PC monitor makers, for example) sell their products with ferrites molded into the cable just for this reason. Placing common-mode filters or LC filters in series with pc-board connectors can reduce EMI caused by cabling.
Surprising as it may sound, the effects of software could pose serious issues when testing for emissions and susceptibility. We have seen noticeable differences in EMI performance with different software architectures.
When you think about it, this makes sense. When EMI testing first begins on a prototype, the system typically runs a stripped-down version of the system software. Then, when the product is submitted for final qualification, the full operating system is ready. Radiation is primarily generated by the switching of CMOS gates within microprocessors, and ASICs and their associated interconnections. The released software may toggle certain pins differently (a watchdog timer, for instance) than the prototype software. An operating system tends to make periodic calls to certain functions, as opposed to a single threaded system. These factors sometimes affect the frequency content on the address and data buses.
Engineers designing products that must meet EMC regulations need be aware of the big picture before they start designing. Designers must understand which certifications the product is required to meet, and the exact configurations that will be tested. Then EMI development and test time must be factored into the schedule. It's worth spending an extra week designing a pc-board to save weeks in the EMI lab.
For companies contemplating buying a GTEM cell of their own, the math is simple. Symbol has enough projects to keep its GTEM cell running every day. And, the time-saving advantage of having a way to measure EMI in your own lab is priceless. For the specific types of products that we manufacture, the FCC has recognized that a GTEM cell produces results that correlate to those of an OATS. Thus, the company has the advantage of certifying its own products.