5 tips for reducing EMI: even the best high-speed design plans can fail FCC certification.

FIXING ALL THE signal integrity problems is no guarantee that a product will pass an FCC EMC certification test, because it only takes a few nanowatts of radiated power, within the roughly 100 kHz bandwidth of the test, to fail. Passing Class B certification, for example, corresponds to having a field strength of less than 100 [micro]V/m, at a distance of 3 m, in a frequency range around 88 MHz.

While it is true that differential currents associated with signals and their return paths do radiate, the common currents that flow up and down planes and external cables provide a much greater source of emissions. These common currents look like an electric dipole antenna and are typically driven by voltages on the planes (FIGURE 1).

[FIGURE 1 OMITTED]

The planes' voltage that drives the common currents is known as ground bounce. Even if the ground bounce voltage is below a level that will cause signal integrity problems, it can still cause radiated emissions that will fail FCC certification. For example, if the ground bounce voltage is 100 mV, and there is a 10 cm-long shielded cable attached to the ground plane, the radiated emissions can exceed 400 [micro]V/m at a distance of 3 m.

An important first step in reducing radiated emissions and passing an EMC certification test is reducing ground bounce. Ground bounce is voltage created between two different points on the ground return path. It is almost always due to a dI/dt through the total inductance of the return path, usually the planes. Reducing ground bounce is all about reducing the dI/dt and reducing the total inductance of the return path.

There are five important methods for reducing ground bounce and radiated emissions:

1. Use differential signaling for I/O wherever possible. This may, in some cases, reduce the dI/dt in the power and ground distribution and the dI/dt in return paths in the planes.

2. Use a signal topology that makes the return current look symmetrical around the signal path. There must be no ground bounce in a coax cable because there is no total inductance in the return path. The magnetic field lines outside of the coax from the signal current have exactly the same distribution, but in the opposite direction, as the loops of magnetic field lines from the return currents. The field lines cancel and there are no external magnetic field lines around a coax, so there can be no total inductance in its return path and no ground bounce. It's not practical to use coax signal lines in circuit boards, but stripline topology is a pretty good approximation of coax. Microstrip topology's total inductance in the return path is more than 10 times that found in stripline topology.

3. Make the return path as wide as possible, i.e., use planes for the ground returns. The total inductance of the return path is reduced if the partial self-inductance of the return path is reduced. The No. 1 way of doing this is permitting the return current to spread apart as much as possible.

4. Bring the first current as close to the return current as possible. The total inductance of the return path is reduced if the partial mutual inductance between the two current paths is increased. For signal currents, the impedance the signal "sees" will be too low and impedance discontinuity problems will arise if the signal path is brought too close to the return path. However, how close the power path can be brought to the ground path is limited only by the price you are willing to pay.

5. Avoid discontinuities in the return path. Anything that keeps the return current from flowing as closely underneath the signal path as possible will increase the total inductance of the return path, thus increasing the ground bounce voltage. After you've put forth the effort of designing for low total inductance, don't screw it up by putting gaps to the way.

These tips are not essential to every design, of course, but will increase the pass rate for EMC tests.

Ed.: The December No Myths contained a formula error. The correct formula is:

[k.sub.b] = 0.5 x (Zeven - Zodd)/(Zeven + Zodd)

REFERENCES

(1.) This column is taken from a lecture Charles Grasso and I gave at a meeting of the Rocky Mountain Chapter of the IEEE EMC Society on Dec. 4, 2003. Copies of the presentation are available on the Web site: http://www.ewh.ieee.org/r5/denver/rockymountainemc/archivehub.htm

(2.) Many of the details on ground bounce and inductance can be found in my new book Signal Integrity--Simplified, published by Prentice Hall.

ERIC BOGATIN is CTO of Synergetix (synergetix.com). He can be reached at [e-mail].