Robert sheds light on the “dark” subject of electromagnetic interference. He covers everything from electromagnetic compatibility (EMC) issues to helpful design techniques. With a few modifications to your next design, you will be able to reduce the amount of unwanted radiation emission.
Welcome to The Darker Side! I am really excited to present this new column devoted to some of the lesserknown, more obscure aspects of electronic design. Scary, isn’t it? If you find the title strange for a magazine as serious as Circuit Cellar, note that I chose it after a brainstorming session with the magazine’s great editorial staff—so the responsibility is shared.
The purpose of this column is to highlight concepts too often perceived as expert-only subjects. If you read “analog” between these words, you will probably be right 90% of the time. I will do my best to help you understand what’s really going on while focusing on application-oriented explanations and pragmatic tips rather than painful theory and math (although some equations may show up from time to time). Lastly, because dark subjects may sometimes be understood differently from other points of view, and because I will probably be wrong or imprecise from time to time, I will be more than happy to receive your feedback. Just drop me an email and I will be in touch!
I want to start with one of the darkest possible subjects. It is something that can give headaches to engineering managers evaluating the risks on their projects. It is something that is often not learned until it is too late. It is electromagnetic interference, or EMI for short.
Electromagnetic compatibility (EMC) is about being reasonably sure that a given piece of equipment will work peacefully with other equipment in its neighborhood. “Peacefully” means the equipment should not be perturbed by others (EMI immunity), and it should not perturb them (EMI emission limitation).
Let’s look at the details. In the EMC/EMI world, equipment is usually classified into two categories: voluntary versus unintentional transmitters (or receivers). For example, a voluntary transmitter could be your remote control transmitter, while an unintentional transmitter could be a badly designed power supply generating plenty of EMI noise. Similarly, a voluntary receiver could be your FM radio, while an unintentional receiver could be your MP3 player picking up the RF noise transmitted by your cell phone.
EMI immunity relates to voluntary transmitters (generating strong RF fields, because they are designed to transmit something) and unintentional receivers (low-sensitivity receivers) that could be agitated by these strong RF fields. Reciprocally, the EMI emission limitation is a concern between unintentional transmitters (low-level but not expected) and voluntary receivers (high-sensitivity because they are designed to receive). Inspired by Tim Williams’s very good book, EMC for Product Designers, Figure 1 illustrates these two fundamental EMC gaps.
Because nobody knows which EMI environment a given product will be used in, legislation defined the absolute maximum admitted emission power and minimum immunity requirements. Legislation on EMC was initially limited to radio communications (i.e., voluntary transmitters and receivers), but it was extended to unintentional transmitters about 20 years ago, particularly in Europe. The first EC EMC directive, 89/336/EEC, was ratified in the early 1990s and was made mandatory in 1996. It covers both immunity and emissions limitations. In the United States, FCC Part 15 limits emissions of unintentional transmitters (spurious emissions, subpart B), as well as low-power, unlicensed transmitters (subparts C to G). But as far as I know, emission compliance with RF immunity standards is not mandatory for most products under the FCC’s guidelines, even if additional standards are usually requested by the marketplace. In all other countries, the international standards IEC61000-xxx from the IEC’s TC77 are usually translated into national standards, providing a similar framework.
Finally, electromagnetic signals can be either radiated (through the air) or conducted (through wires). Standards usually deal with these two aspects separately, even if they are often closely linked, as you will see.
You probably think that I was lying in my introduction when I promised to focus on application-oriented explanations and pragmatic tips rather than on long and painful theory. Agreed, I am going to the bench. Because the EMC subject could fill up thousands of pages, I will concentrate on only one topic: unintentional emissions. To demonstrate the basics of this nasty phenomenon I will not describe a rocket science project. I’ll focus on one of the simplest unintentional transmitters: a small TTL oscillator built around a 74HC00 chip and the 3.5795-MHz crystal I had on my desk, with a 1-kW load resistor simulating a clocked device on the board (see Figure 2).
Before I demonstrate the phenomenon, I need to introduce the EMC guy’s “pocket knife.”
THE SPECTRUM ANALYZER
You will need a way to see and measure RF because EMI is mainly about RF signals covering a large frequency range (from nearly DC up to several gigahertz) and a large signal amplitude range (from watts down to nanowatts or lower). A spectrum analyzer, an expensive but invaluable tool, must be used. Oscilloscopes are currently available far above the gigahertz region (even up to 18 GHz for those with really deep pockets), and high-end digital models provide a spectrum analyzer using a fast Fourier transform of the time-acquired signal. Don’t be confused. This is often a useful feature, but by no means equivalent to an RF spectrum analyzer. The reason is the dynamic range. The average highspeed oscilloscope uses an 8-bit ADC in the giga-samples-per-second range and may be a 10-bit ADC for higher-range scopes. This provides, at best, seven or eight effective bits. From MSB to LSB, this provides, at best, a dynamic range of 1 to 256 (i.e., 28). This is in terms of voltage. Because power is proportional to voltage squared, this translates to a dynamic range of 1 to 65,536 (i.e., 2562) in terms of signal power. This is not bad, but it represents only 48 dB (i.e., 10 log(65,536)). The worst real spectrum analyzer provides at least 80 or even 100 dB of dynamic range.
A spectrum analyzer is nothing more than a tuned, swept, very selective receiver. The architecture of a simple RF super-heterodyne spectrum analyzer can bee seen in Figure 3. The measured signal, coming from a wire or an antenna, first goes through a step attenuator to avoid saturation. A low-pass filter gives it a clean signal from kilohertz up to the maximum frequency of the equipment. Assume the maximum is 3 GHz. The signal is then mixed with a clean high-frequency swept local oscillator covering a 3-GHz range (from 3.5 to 6.5 GHz). It is bandpass filtered around a frequency just higher than the maximum usable frequency (3.5 GHz). This architecture allows it to limit spurious receptions. For example, when the local oscillator is set at 4.223 GHz, the only input frequency that will provide a signal going through the 3.5-GHz filter will be at 723 MHz (i.e., 4.223 GHz – 3.5 GHz).
The 3.5-GHz intermediate frequency signal is amplified and down-converted through several mixing stages to a low intermediate frequency. Then, one of the most important parts of the spectrum analyzer comes in: the resolution filter. This is a very selective, calibrated, and variable band-pass filter that allows it to set the measurement bandwidth, which has a strong influence on sensitivity, frequency discrimination, and measurement speed. The signal is sent through a logarithmic amplifier, allowing the analyzer to display a power level in a logarithmic scale over several decades. Finally, its amplitude is detected, filtered again to remove noise by averaging, and displayed.
This presentation is simplified because digital techniques are heavily used from the latest intermediate frequency blocks down to detection through digital resolution filters and FFT. Moreover, the microwave equipment that has a frequency range far above a few gigahertz uses a different conversion scheme for its input section. They use exotic Yigbased preselection band-pass filters and harmonic mixers. Anyway, this simplified architecture fairly matches the architecture of 20-yearold spectrum analyzers, such as my Hewlett-Packard HP8569. If you want to know more about spectrum analyzers, the bible is “Spectrum Analysis Basics” (application note AN150). Now that you have a spectrum analyzer, connect a wideband antenna, or better yet, a preamplified electric field probe to its input. Now you have a real-time view of the surrounding RF environment (see Photo 1). Such a test setup is not adequate for EMC formal assessments requiring calibrated antennas and shielded anechoic chambers, but it is a very useful relative measurement tool, as you will see.
Photo 1 was taken with a “clean” environment, but what happens when you switch on the test crystal oscillator? I built it on a small prototyping board, even if you don’t agree that it is the nicest idea for a good RF design (see Photo 2a). I then dropped it on a desk 2 m away from the receiving antenna (that’s around 6¢or 7¢for those living on the wrong side of the ocean), and powered it from a 5-V bench supply. Photo 2b shows what was displayed on the spectrum analyzer’s screen. I am not joking! The separation between peaks is close to 3.5795 MHz, which is not a surprise. You can see all the harmonics of your crystal oscillator from 20 MHz up to a few gigahertz. That’s a good illustration of unintentional emissions, isn’t it? Note the power of the maximum peak on the first test as a reference. The top of the screen is –7 dBm. The vertical scale is 10 dBm/div. I found the maximum peak at about –34 dBm (including a gain of the preamplified antenna, air path loss through the 2 m between the test oscillator and the antenna, and so on).
Go back to the reference spectrum in Photo 1. In the same setup, the FM broadcast peaks were measured at –63 dBm. That means the EMI field transmitted by the small crystal oscillator is 29 dB (i.e., 63 – 34) more powerful than the FM broadcast signals, even if the crystal oscillator is 2 m away from the receiving antenna. And, if you are still reluctant to use decibels, this means that this interference is no less than 794 times (i.e., 1029/10) more powerful than the radio station you want to listen to!
What happened? Any AC current going through a conducting loop generates an electromagnetic field that will propagate in the space with an amplitude roughly proportional to the current, to the surface of the loop, and to the square of the frequency. Because your crystal oscillator is more or less using square waves, all harmonics of the clock signal are transmitted and unfortunately well received, up to the maximum frequency of the HC00 gates.
What should you do? First, try to reduce the area of the current loops. Photos 3a and 3b show what happened when I routed the wires differently, connecting the 1-kW resistor differently. The spectrum looks the same, but the maximum peak is now –39 dBm. That seems like a small improvement, but remember that you are using a logarithmic wide dynamic scale.
That’s a 5-dB improvement (from –34 down to –39 dBm). It means the power of the received signal is 105/10 lower, so it was reduced by a factor of 3.2 just by moving a wire. Not so bad. So, the golden rule is to limit all loops carrying high-frequency signals or low-frequency signals with abrupt slopes like logic signals. Because the return-current path is usually the ground, this recommendation translates into the most important EMC tip: build your circuits with a full ground plane on one layer of your PCB. That will ensure that there is always a return path just below every wire. Some will argue that split ground planes could be good, but I’m sure that 99.9% of the designs will work better with a full ground plane rather than with more complex schemes (except for some audio designs well packed in a shielded enclosure and not really affected by radiated EMI). For example, if you design a board using two ground planes joined only by a small bridge, then there will be current loops, except if all other wires going from one area to the other are passing exactly over the bridge, which is not easy to do.
How do you go further with minimal changes to the design? The level of harmonics generated by our test project is as high as it is because the signals are steep square signals, but is this steepness actually needed? Why use a 74HC00 with a propagation delay of 7 ns, high output current, and more importantly heavy current peaks at each transition due to its CMOS technology to build a simple 3.5-MHz clock? I dug into my shelves and swapped the HC00 with an old LS00. The maximum peak was immediately reduced to –45 dBm, which is a 6-dB incremental improvement. Compared with the initial situation, these two small changes reduced the transmitted noise by 11 dB, or more than a factor of 10, just by moving a wire and swapping a chip with an older one! Let me summarize the lesson learned by this second experience. Always use the slowest possible technology for a given project, with the slowest possible output currents and the slowest possible voltages, particularly in digital signals. That’s obvious, but it’s too late when you remember it after the design is complete.
I didn’t want to talk about conducted EMI in this column, but I have to. Assume that the noise level is still higher than acceptable. You can continue to improve the design every where you can and see the result on the transmitted noise 2 m away, but that could be quite time consuming. It would be easier if you could pinpoint the precise location of the transmitted noise source. Luckily, this is also possible with a spectrum analyzer. The only additional tool needed is a magnetic field probe (H-field probe). The reason is that an H-field is decreasing as the cube of the distance, while the E-field is decreasing as its square. So, using an H-field probe is far more effective to locate a transmitter at very short ranges because the H-field is decreasing very quickly when you leave the immediate surroundings of the emitter. You can buy good, lowcost, preamplified H-field probes like the Hameg HZ530 probe set that I’m using, but after buying the spectrum analyzer your pockets will be empty. Fortunately, a simple H-field probe can be easily built using a small 50-W shielded wire. Just strip the cable a few millimeters from the end, turn it to make a small loop, and solder the inner connector back to the shield braid. The end of the braid should not be connected to itself in order to provide a gap in the shield. That’s it. Now you have a good magnetic field probe.
Assuming that you have an H-field probe connected to the spectrum analyzer, move it around the design, and in particular, around the wires connecting it to the 5-V power supply (see Photos 4a and 4b). Bingo! You can see that the power cables are radiating a strong RF field in particular in the 50- to 100-MHz range. The current peaks drawn by the logic chip show up as conducted current peaks on the power lines because of insufficient decoupling, even though the ubiquitous 100-nF capacitor is in place on the power rails. The conducted EMI is then radiated due to the long unshielded power wires. That’s why conducted and radiated EMI are often linked.
You must reduce the conducted EMI through the power lines. I’ve modified the test oscillator with the addition of a 1-mH coil plus a ferrite bead both on the GND and 5-V lines, as well as the addition of a 10-μF low-ESR capacitor in parallel to the 100-nF one (see Photo 5a). One caution regarding filtering coils: in high frequencies, a higher-value coil is not always better, because parasitic capacitances can quickly make it useless. Technology does matter, and that’s why ferrite coils are usually far more efficient for EMI reduction than standard coils of the same value. This subject will need a column by itself.
Back to the spectrum analyzer. The maximum received peak is now at only –51 dBm if you consider only the HF contributors above 10 MHz, another 6-dB improvement (see Photo 5b)! As a last improvement, I tried moving the 1-kW resistor closer to the chip (see Photo 6a), and the received EMI level is down to –55 dBm, 4 dB better (see Photo 6b).
IS IT ALL?
Using one of the most simplistic designs (a crystal oscillator) built on the most simplistic platform (a protoboard), I found a couple of very simple design modifications that allowed me to reduce its unwanted emissions by 21 dB (i.e., 55 – 34), effectively reducing the spurious noise power by 99.2% (1021/10 = 125, 1/125 = 0.8%)! With these changes, the FM broadcast signals are now clearly visible on the spectrum analyzer, which means that you will be able to hear your favorite songs. And more importantly, these design changes were really minor, including rerouting some wires, an HC00 replaced by an LS00, coils, and ferrites on the power supply. Impressive, isn’t it?
In this short article, I have highlighted some key points of one aspect of EMC, unwanted radiated emissions, and some design tips that will make your life easier. The EMCminded designer will also need to take care of immunity, conducted EMI, common mode noise, ESD, and the regulation side of the EMC. Anyway, I hope that I have successfully demonstrated that the EMC is not black magic, even if it is occasionally on the darker side. CC
Robert Lacoste lives near Paris, France. He has 18 years of experience working on embedded systems, analog designs, and wireless telecommunications. He has won prizes in more than 15 international design contests. In 2003, Robert started a consulting company, ALCIOM, to share his passion for innovative mixed-signal designs. You can reach him at email@example.com. Don’t forget to write “Darker Side” in the subject line to bypass his spam filters.
Agilent Technologies, “Spectrum Analysis Basics,” application note AN150, www.metrictest.com/resource_center/pdfs/agl_spec_analyzer_basics.pdf.
T. Williams, EMC for Product Designers, Newnes, 2007
HP8569 Spectrum analyzer
Agilent Technologies, Inc.
HZ530 Probe set
PUBLISHED IN CIRCUIT CELLAR MAGAZINE • AUGUST 2007 #205 – Get a PDF of the issueSponsor this Article
Robert Lacoste lives in France, between Paris and Versailles. He has more than 30 years of experience in RF systems, analog designs and high-speed electronics. Robert has won prizes in more than 15 international design contests. In 2003 he started a consulting company, ALCIOM, to share his passion for innovative mixed-signal designs. Robert is now an R&D consultant, mentor and trainer. Robert’s bimonthly Darker Side column has been published in Circuit Cellar since 2007. You can reach him at firstname.lastname@example.org.