EMI

Ten Common EMI Problems

William D. Kimmel and Daryl D. Gerke

Attention to design fundamentals, old and new, can prevent the awkward later appearance of interference issues in equipment.

When the authors of this article some years ago wrote a paper entitled "Twenty Common EMI Problems," the primary electromagnetic interference issue was emissions. The European Union requirements were not yet in effect, and emissions were regulated primarily through 47 CFR Part 15 (the Federal Communications Commission). Things have changed a lot since those days. A wider variety of EMI problems are prevalent—emissions, yes, but many problems involve immunity and self-compatibility. Some of the problems encountered long ago have, by and large, become routine matters. However, others persist to this day. Plus, a lot of new problems have arisen.

This article discusses half the number of problems that the earlier one did. Some have been bedeviling designers for years, sometimes introducing a new wrinkle, and others are problems of more recent origin. They constitute something of a rogues gallery. But these 10 rogues can be tamed.

1. Ground Impedance

Ground impedance comes first because it occurs so often. The overwhelming majority of high-frequency problems, whether relating to emissions, self-compatibility, or immunity have high ground impedance at the root. These are not low-frequency ground loop issues, nor earth grounds. These are problems caused by local ground impedances such as are found on circuit boards or in cables. High-impedance ground paths are the principal contributor to cable shielding failure and common-mode currents.

Simply put, wires and traces are always high impedance. This is why a ground plane should be used at high frequencies, to keep the ground impedance as low as possible. How high is a "high frequency"? That depends on the application. But, with a wire or trace, the inductive impedance is higher than the resistance of the path already at audio frequencies, and it is significant at 1 MHz. At that point, designers should avoid using wires or pigtails for grounds. A good rule of thumb is that the inductance of a wire is about 20 nH per in. of length.

A 1-in. wire or trace has an impedance of Z = 2pfL = 12 W at 100 MHz—hardly a short circuit.

At radio frequencies (RF), any length of wire should be viewed with suspicion. A useful rule is to keep the path width at least one-fifth that of the length. Thus, a 5-in.-long ground strap should be at least 1 in. wide, and preferably wider.

2. Poor Cable Shielding

When an emission or RF immunity problem is encountered, it almost always involves the cable. Ground impedance, as just discussed, plays a key role in cable termination performance.

The best cable termination is the circumferential wrap, where the cable connector is grounded around the entire perimeter. A compromise is a clamp arrangement (see Figure 1). A wire or pigtail termination is unacceptable at any frequencies above audio.

Figure 1. A cable shield clamp is preferable to a pigtail ground.

A visual inspection will give a pretty good indication of cable shielding effectiveness. However, a problem may still be encountered in the field or on the test floor. There are several possibilities as to cause.

The cable shield may be single-point grounded. Cable shields must be grounded at both ends any time the cable length is 1/20 of a wavelength or longer. A quarter-wavelength is a worst-case condition. A cable purchased from the local computer store is almost assuredly grounded at one end only—and more than a few such cables are not grounded at either end. If the cable is grounded at both ends, it will be so labeled on the box. That is worth checking to be sure. Unfortunately, it is often difficult to tell if cable shields are terminated circumferentially. Many purchased cables use pigtails, which are bad news for EMC (see discussion above).

The cable shield may be damaged. Many cable shields are Mylar foil, which is not very robust. Even with careful handling, the shield could rupture somewhere along the cable, degrading the shielding effectiveness. The trouble with this problem source is that the rupture is not visible, nor will it be detectable with a digital voltmeter.

Finally, the cable shield could be grounded through the drain wire. This causes a problem if the drain wire is attached to a pin in the connector, making it almost completely ineffective, or to a screw post inside the connector, in which case shielding effectiveness is reduced. It is important to note here that this creates a pigtail.

3. Emissions from Switching Devices

The problem of emissions from switching devices goes back to the early days of emission requirements. The switching node—usually the collector or drain—of a power supply couples noise into the heat sink, creating a conducted emissions problem. This problem has spread with the proliferation of power-switching devices. Figure 2 depicts the situation with a motor drive, where the high-frequency noise capacitively couples to the motor housing.

Whenever encountering any electrical drives or relays, it is worthwhile to ask where the switching devices are. Surprises are possible. The authors recently encountered an emission problem from a relay.

Figure 2. The switched-power noise path in a motor drive.

The solution is to provide differential-mode filtering as close to the offending elements as possible, and to maintain as much physical isolation of the noisy node as can be managed. Heat sinks should be well bonded to the appropriate reference plane, usually power common.

4. LCD Emissions

The increasing use of liquid-crystal displays (LCDs) has created a problem with emissions. Inevitably, the problem surfaces with the data cable, usually a flex cable from the circuit board driver to the LCD panel. The currents going up to the LCD do not all return on the cable; a small fraction remains as common-mode current, exciting the LCD. As discussed in section 1 above, ground impedance is again the problem here (see Figure 3).

A number of steps can be taken in this case. First, the currents need to be returned to the source—the driving circuit board—with as small a loop area as possible. The return-path impedance might be reduced by running a ground strap under the flex cable. But better would be to provide direct shunts from the LCD panel back to the circuit board. Effectively, this means grounding the panel to the circuit board, at all four corners if possible. Most LCDs have a metal plate at the back of the panel as part of the frame. If a particular display does not, the designer will need to add one. A ground plane is necessary to keep emissions from coming directly off the front of the panel, which may happen even if the perimeter is grounded.

Figure 3. Emissions from an LCD.

If a device has a metallic enclosure, then the designer should ground the LCD to the enclosure around the entire perimeter.

If for some reason a direct connection cannot be made, the designer should go with small (1- to 10-nF) capacitors at the mating points. This option is definitely second best, but it is better than nothing.

5. Stray Internal Coupling Paths

The problem mentioned in section 3 is not limited to switching devices: stray internal coupling paths are common in both emissions and immunity problems. Stopping EMI at the circuit board level whenever possible may be ideal, but the effectiveness of design measures is often undermined by stray internal coupling paths.

There are a couple of possibilities for dealing with this problem source. One is capacitive coupling from the inductor or ferrite. These components are usually inserted at inputs and outputs to block RF interference from entering or exiting. In such a role, the side of the inductor facing the noise source carries considerable high-frequency voltage, which will capacitively couple to any nearby metallic member, including ground planes, circuit board traces, and heat sinks. The solution is careful placement to avoid coupling to sensitive recipients.

Capacitive coupling from board elements to connector pins is a problem, too (see Figure 4). This path effectively bypasses any filter elements that may have been placed on the board. This is why a ferrite that is placed outside on the cable often works better than a ferrite placed inside on the circuit board. The solution is to intercept the coupling path with a Faraday shield, which can be located at the offending chip or at the connector. In either case, the shield connects to circuit ground.

Figure 4. Stray capacitance between the chip and connector can bypass on-board filters.

6. Component Parasitics

Parasitics are the stray reactive elements found in every component, whether a passive element or active device. Ideal components, such as those that may be characterized in the classroom, exist nowhere outside of the textbooks.

All capacitors have series inductance, creating a series resonant circuit. All wound inductors have interwinding capacitance, creating a parallel resonant circuit. These circuits resonate at much lower frequencies than might be thought, and are, of course, ineffective above resonance. Most capacitors resonate at well below 100 MHz, and most wound inductors resonate below 20 MHz. Even small transformers resonate below 5 MHz. So, designers should make sure their selected component is functioning at the frequency range of interest. It is virtually certain that filter and decoupling elements are being operated above resonance.

Turning from passive to active components, the same problem presents itself. This includes stray capacitance and inductance in the lead frame and bond wires, as well as in the die itself, which results in internal pin-to-pin crosstalk. This crosstalk problem is getting worse, what with decreasing pin pitches and higher speeds. These parasitic elements are also interacting with external circuit components, creating a dazzling array of resonances.

7. Intracable Crosstalk

Crosstalk is most often a self-compatibility problem that occurs when noisy power levels and signal lines are routed in the same bundle.

Power delivered to motors and other heavy loads tends to undergo transients during start and stop episodes. And, increasingly, switched-power motor drives are being developed that switch continuously at a high rate. These interfere with digital and analog signals. Left unchecked, power transients couple into digital lines, causing digital data errors. One of the most common of these resulting problems involves the pulse count from the encoder wheel, which is used to keep track of the rotor position or rotational speed. The count can be jammed by interfering signals and the encoder function lost. Similarly, video data lines often are susceptible to interference from motor drives.

The basic problem is the very close proximity of the noise source to the recipient circuit along the entire cable run. The easiest solution is simply to route the signals and power in separate cables. Crosstalk falls off fairly quickly, so, generally, not much separation is necessary to solve the problem.

If power and signal must be routed in the same cable, the designer should use shielding and magnetic-field cancellation techniques. Shielded twisted pair is effective for both electric- and magnetic-field coupling. Certainly, at a minimum, the signal lines should be shielded. Shielding of the power feed may be necessary as well. It is up to the designer to decide whether to ground the shield at one end or at both; the treatment will depend on the edge rates of the interfering source. Typically, power disturbances will want a single-point ground. Audio-frequency shields should usually be single-point grounded.

8. Inadequate Signal Returns

A common situation with cables, especially intrasystem cables, is that there are not enough return lines. The design thinking has been that anything more than one wire for ground return is waste.

This is another case where high ground impedance is the source of a problem. The signal goes out one line, and most of it comes back on the return, as intended. But unless the return path impedance is zero, a small portion of the current returns on a stray path. That portion may be as small as one part in a thousand, but even that is often enough to create a common-mode cable emissions problem.

In the extreme case, the voltage drop on the return is enough to cause data errors, creating a signal integrity problem.

How many returns are needed depends on data rates and run length. For high speeds, those above 100 MHz, designers should figure on one return for each signal line. For lower speeds of less than 10 MHz, one return for every five signal lines is recommended.

9. Discontinuous Return Paths

Many printed circuit board (PCB) problems can be traced to a discontinuous return path. Grounding plays a key role in all aspects of electromagnetic compatibility (EMC), including in PCBs. The EMC problem in PCBs is becoming more acute as speeds rise.

The problem revolves around the signal current loop. Ideally, the signal goes out a trace and returns immediately under the trace. The laws of physics dictate that the current follows the minimum-energy path, which usually means the smallest loop area. But all too often the return path is broken by a discontinuity; that is, the trace crosses a cut in the plane or passes through a via and changes reference planes. When something like that happens, it creates a discontinuity, energizing the slot and causing signal reflections.

A good design rule is to spend as much time considering the return path as is spent on creating the signal path. Designers should avoid crossing slots, should drop a ground via if ground planes must be switched, and should put in a capacitor near the via if they are using Vcc as a reference plane. Clearly, this cannot be done for all signals. The best thing is to concentrate on the critical signals, clocks, and data buses in the case of emissions, and on strobe lines for immunity.

10. ESD in Metallized Enclosures

The problem of electrostatic discharge (ESD) often occurs as the result of an afterthought. People who use plastic enclosures add metallization only when forced to, in order to minimize emissions or RF immunity problems. To be effective, the metallization needs to be brought right up to the seams so as to provide conductive continuity between mating surfaces. Unfortunately, this creates a whole new set of ESD contact points. The designer may create a new problem while solving another one.

This can be a tough nut to crack. The designer has essentially three choices: to redesign the equipment to eliminate the need for metallization, to recess the metallization so that the arc cannot get to it, or to take extra care to close the gaps in the shield. Basically, the problem arises not from the presence of the metallization but from the gaps in the shield. It is hard to get good closure with a metallized box. But once the gaps are closed, the shield should perform well for ESD.

Far and away the best defense against ESD is to prevent discharge in the first place, leaving only indirect discharge as an effect needing to be handled. That means that the designer has to recess metal members beyond the reach of static discharge, something that is not always feasible.

If discharge cannot be avoided, then it will take a very good shield to cope with ESD.

Conclusion

Equipment designers run into a lot of EMI problems, some of them quite subtle. But surprisingly often, EMI problems occur because the fundamentals have been ignored; some of the problems described in this article have been around for decades. Others have appeared recently and are growing increasingly common.

The problems analyzed here are central to equipment design, and all are manageable. Designers who work to avoid them will avoid a lot of headaches.

William D. Kimmel, PE, and Daryl D. Gerke, PE, are partners in Kimmel Gerke Associates Ltd., an EMC consulting firm with offices in St. Paul, MN, and Mesa, AZ, that specializes in EMC design, troubleshooting, and seminars. Together, they have more than 70 years of EMC experience. The NARTE-certified EMC and ESD engineers may be reached by phone at 888-EMI-GURU or via their Web site at www.emiguru.com.