Listen to the pop, sizzle, and hum. Look at the bars and wavy lines. These and other characteristics of electromagnetic interference (EMI) are very familiar.
RF energy is radiated from a culprit device and picked up by the antennas of a victim product; by power, control, and signal leads; and by direct propagation through the case into its sensitive electronic circuits. In addition, troublesome RF energy from a shared power line often is coupled into the victim through a common impedance. Small frequency-selective attenuators, more commonly called EMI filters, can help solve the problem.
Generally, EMC problems are just nuisances, but there are circumstances where EMI-induced malfunctions could jeopardize equipment, missions, and personnel. To control these problems, government organizations such as the U.S. Federal Communications Commission (FCC), the Canadian Standards Association (CSA), and the European Union (EU) mandate that manufacturers must meet their appropriate intersystems EMC standards.
Reducing Radiated and Conducted RF
The four aspects to be considered in EMC design are radiated and conducted emissions and radiated and conducted susceptibility. These can be divided into two categories: radiated RF and conducted RF. Very simplistically, after good PCB design practices have been implemented, shielding is used to solve radiated problems, and filtering addresses conducted problems.
The two approaches are somewhat synergistic. Shielding does not slow down the operation of high-speed systems, and it provides an isolated ground reference. In turn, the isolated ground reference effectively reduces internal crosstalk, circuit-path coupling, and common-mode coupling.
In many cases, shielding alone will eliminate the need for EMI filters. But if filters are required for conducted emissions, a shield enhances operation by providing the filter with a better ground sink and isolating its input from its output.
Even though good preliminary design combined with shielding typically reduces the need for filters, sometimes the choice becomes redesign or use a high-performance reflective filter to reduce conducted energy. Since redesign introduces new unknowns, the filter is the better choice. However, since reflective filters are made of inductors and capacitors, their installation should not be done indiscriminately because their phase shift and circuit timing delays often affect signal integrity.
Power supply circuits also are not immune from these effects. Power supply filters with their large-value components can significantly shift the power factor to the point where power-factor correction circuitry no longer will work. Even with these limitations, filtering remains an important technique for solving conducted emissions and susceptibility problems.
Filter Characteristics
A filter’s great advantage is its capability to attenuate a specific range of frequencies. This is made possible by the frequency resonance characteristics of sets of inductors and capacitors. Based on the attenuation-frequency response, a filter is categorized as low pass, high pass, passband, or stop band. The name describes its response characteristics. Figure 1 shows some simple filter circuits and their frequency responses.
Most filters used to solve EMC problems are low pass reflective filters. These attenuate the higher-frequency RF noise produced by circuits that create the system’s intentional signals. A filter reflects the noise back to the source out-of-phase where cancellation occurs.
Understanding how filters work is not difficult. Building, installing, and using them properly are different matters.
If you are really interested in learning the details about filters, a number of sophisticated computer synthesis programs can design filters based on equipment requirements. These programs synthesize any desired transfer function, though most EMC filters are Butterworth design because of its flat passband response.
Ideally, a low pass filter should pass all intended frequencies below its cutoff frequency without any loss or distortion and completely reject all unwanted frequencies above its cutoff frequency. Unfortunately, because a filter is constructed from inductors and capacitors, each LC combination with only 20-dB/octave frequency responses, this is not possible. Still, by combining more and more inductor and capacitor combinations, the attenuation will increase rapidly above the cutoff frequency.
Since the attenuation slope changes 20 dB/octave for each reactive component, a simple, three-element filter provides an attenuation of 60 dB/octave. A five-element filter would achieve 100 dB in the same range. This increase will continue until the parasitic inductance and capacitance begin to limit the attenuation.
Choice of a Cutoff Frequency
Generally, we don’t want to filter useful signals. As a result, the cutoff frequency for analog signals must be at least twice the frequency of the highest intentional signal component, that is Fco = 2 Fsig. For digital signals, the cutoff frequency must be greater than the pulse bandwidth frequency or
Fco = 1/p trise time
Additionally, the filter typically will be inserted into a transmission line with a characteristic impedance Zo or into a power supply with an input or output impedance Zo where maximum power transfer is required. This means the filter impedance below its cutoff frequency, where the intentional signals are located, should match Zo. This is approximately true when
For a signal transmission line, Zo can be relatively constant with frequency. The power-line impedance, on the other hand, generally varies a great deal. That’s why we use a line impedance stabilization network (LISN) for testing.
However, a rough assumption for the power line is its source impedance, which at 50 to 60 Hz is equal to the source voltage divided by the delivered current or Zo = V/I. It may not be exact, but this at least gives us a starting point because, in the absence of any other information, we don’t know if the source impedance increases or decreases with frequency.
In an LC pair, resonance occurs at the frequency where XL = XC, which corresponds to Zo. For the analog filter design, the cutoff frequency must be twice the highest system operating frequency. As a result, the inductance must be smaller and the capacitance larger by a factor of two. After adjusting the reactance formulas accordingly, the component values can be determined from
This seems really simple—except that the inductors may turn out to be big enough to contain the equipment, and the capacitors really aren’t capacitors because of parasitics and lead inductance. Anyway, the inductor and capacitor values are related by the resonant frequency where XL = XC .
Solving for the resonant frequency gives the following equation:
For our simple filter, FCO is twice the highest operating frequency, so
which means
In this example, the cutoff frequency requirement is met for any combination of inductor-capacitor values whose product is equal to the constant k. As a result, filters with the same cutoff frequency are not necessarily created equal.
From a practical standpoint, filter capacitors and inductors typically can be made with values in the range of 1 pF to 10 µF and 0.1 µH to 10 H, respectively. Keep in mind that inductors are cheaper than capacitors. The difference arises from the higher cost to determine the performance characteristics of the filter.
Insertion Loss
In the quest for standardization and the customer’s desire for comparison, filter manufacturers use MIL-STD-220 or -220A to measure filter insertion loss. This is a very straightforward test. An RF signal generator is connected to the filter input and a wave analyzer/receiver to the output. Then the reduction in injected signal amplitude vs. frequency is determined.
The latest revision to the standard added a requirement to determine the effect of running the filter at full load with a DC power source. The test provides great data for the matched 50-W measurement conditions. Unfortunately, the installation very seldom is matched, and in the case of power lines, it never is 50 W. Both source and load impedances are critical to the performance of the filter.
As the circuit impedance is reduced, the filter cutoff frequency increases, and its attenuation decreases. The effect is more for filters with larger value capacitors than those with larger inductors. For this reason, a designer should use inductive filters in low-impedance circuits and capacitive filters in high-impedance circuits.
If the source and load impedances of the circuit are mismatched, the filter design must be inductive at the low impedance point and capacitive at its high impedance point, as illustrated in Figure 2. And yes, it will make a big difference in attenuation if it’s installed backwards.
Additionally, a filter used in a high-speed transmission line must match the characteristic impedance of the line to minimize reflections, waveform distortion, and delay. Even though a Simulation Program with Integrated Circuit Emphasis (SPICE) can be a great help in determining filter requirements, the best selection practice before buying is to discuss the circuit characteristics with the manufacturer, order some samples, and test them in the actual circuit. That is the only way that you can ensure that there will be no ringing, unacceptable reflections, or power-factor effects to upset system operation.
Installation
Lastly, to achieve maximum attenuation, proper installation is necessary to reduce parasitic capacitance that couples RF energy around the filter and degrades its performance at higher frequencies. At frequencies below approximately 30 to 50 MHz, parasitics aren’t a major problem, and filter components can be mounted directly on the PCB.
Above 50 MHz, however, the filter should be constructed in a shielded enclosure, mounted on a bulkhead, and grounded carefully through its faying surface and not through a lead. If the filter enclosure cannot be mounted through a bulkhead, the input and output leads should be located as far apart as possible.
Although each approach minimizes the effects of parasitic capacitance, bulkhead mounting is decidedly better. It can easily be argued that if a filter cannot be adequately grounded, it is better to leave the filter out of the design.
About the Authors
Ron Brewer is vice president of EMC technical services at Laird Technologies. He is NARTE-certified and has more than 25 years of experience in EMC/ESD/Tempest engineering. He has made almost 200 technical presentations in North America, Europe, Asia, and the Pacific and been named a Distinguished Lecturer by the IEEE EMC Society. Laird Technologies, Box 650, Delaware Water Gap, PA 18327, 570-424-8510, www.lairdtech.com
Return to EE Home Page
Published by EE-Evaluation Engineering
All contents © 2001 Nelson Publishing Inc.
No reprint, distribution, or reuse in any medium is permitted
without the express written consent of the publisher.
December 2001