Clay Tech

"clay-works make things real"

translated from clazytech.com

Learn Unwanted Radiation Countermeasures In 20 Minutes

original Engineering

The following text was written for a certain occasion, but since it contains content of high public value, I decided to pull it out and publish it separately. I’m going to write about things that might be useful, from how radiation occurs to clues for countermeasures to practical, real-world operation. The content leans quite practical, so I’ve kept it fairly short (whether it’s a 20-minute read depends on the person). If you want the detailed theoretical side of electromagnetism, consult a specialist textbook. The base of my knowledge of electromagnetism was mostly this textbook I used in university.

Electromagnetism

My practical experience with unwanted radiation countermeasures is largely based on my time at a certain major company. (I spent many days shut up in an anechoic chamber, over and over again…) The power supply circuits were often in-house designs, and many of the products had board configurations that had been extended and added onto after the fact, so I gained experience on a fairly troublesome category of products in terms of unwanted radiation countermeasures.

First, about radiation

“Unwanted radiation” is also called “unintentional radiation.” Essentially it means radio waves are coming out that weren’t meant to come out. So where does it come from? More or less, it comes out from “every corner of the electronic circuit.” That said, in a system where current flows cleanly, radiation hardly occurs at all. Here, “flowing cleanly” means going out cleanly and coming back cleanly.

Any signal is established as a system where current supplied from the power source is converted into some form, sent out somewhere, caught at a distant location (this is where it gets read as a signal), and then returns to the power source through some other path. The selection of this system (which path the current flows through) is basically determined by wherever impedance is lowest. (The roots of this go down to quantum theory, actually. If you’re interested, it’s fun to study.) But on a typical PCB, current can’t go back and forth that cleanly (it gets forced into detours, or reflects off dead ends), and a bias occurs in the current density.

When I was in Silicon Valley, I had the chance to work with an NVIDIA layout designer and saw his designs. He skillfully used traces at subtle angles that were neither vertical (90 degrees), horizontal (0 degrees), nor diagonal (45 degrees) as commonly seen on typical boards, along with gentle curved traces. This is a design consideration to minimize bias in current density and avoid creating wasteful L components or reflections.

When current density becomes biased, some of the charge can’t cleanly return to the power source and jumps out looking for somewhere to go. Roughly speaking, this is the origin of unwanted radiation. The charge that jumps out looks for a way back, via the path with the lowest impedance. The location normally expected to have the lowest impedance relative to the negative side of the power supply is usually GND. So if there is GND near the area where current density is biased, the charge jumps there, and as a result the radiation ends up being resolved at a very small level.

But if the impedance from GND to the power source isn’t as low as expected, or if there’s no effective GND nearby, the electric field lines end up going all the way around and out, and an electric field is generated — that is, unwanted radiation is observed. Common techniques used against GND to prevent this include:

Roughly speaking, all of these are measures to lower the impedance and let the current return cleanly.

So, for now, what can be said about radiation is:

Clues for countermeasures

What can be done about unwanted radiation countermeasures really doesn’t change much, no matter how much the times change, no matter if we’re in the age of AI and big data, no matter if the singularity comes.

Hitting the source (genpatsu)

This is the idea of hitting the “source = the originating oscillator or clock.” This is an extremely fundamental countermeasure, so if this can be done successfully, the countermeasure is completely finished.

Concretely, “hitting” means:

This is the way of thinking.

The closer a signal gets to a square wave, the more harmonics appear. Even if the main clock is 10MHz, unwanted radiation failures often occur at 20MHz or 30MHz. This is because of harmonics. Generally, the higher the frequency, the harder countermeasures become, so lower frequencies are always better. Therefore, deliberately softening the waveform within an acceptable range can be one effective countermeasure.

Something similar happens when signal reflections (like ringing) occur due to impedance mismatch.

Also, since current density distribution is frequency-dependent (because various impedances have frequency characteristics), if harmonics remain, you can end up in a whack-a-mole situation — “we knocked down the 3rd harmonic and the 5th showed up” — which is a common occurrence in unwanted radiation countermeasures.

Concrete ideas for countermeasures:

When developing a product, it’s common to make a “clock table.” This is where you compile the frequencies of the originating crystal or PLL, the clocks used for communication, and so on. That becomes a clue when actually tackling countermeasures.

Common examples in general include:

and so on.

In cases like this, it’s preferable to limit your thinking to “things that appear externally as signals.” The reason is that if the clock is blowing directly from the silicon itself, countermeasures can only be taken internally, and the noise is probably going out externally via power/GND anyway, so that’s where you should be hitting.

Oscillator circuits

Places like oscillators are probably very simple in terms of circuitry. So what you need to watch for is:

That’s about it. In other words, there isn’t a lot you can do in cases like this.

Also, in cases where the generated clock is being sent on a long journey, it’s common to route it through an inner layer. This is a common countermeasure to strengthen it against noise by sandwiching it with GND. There’s some debate about this, and I personally think there’s also merit to the view that if it stays on the surface, countermeasures remain possible (like adding a damping resistor or a C to make a filter).

Communication circuits

I touched on communication clocks above, but if radiation does occur, USB is probably the toughest case. This is because USB needs to keep the eye pattern clean to guarantee communication performance. That means you can’t casually soften the signal (the simplest countermeasure is off the table). Conversely, there are also cases where the USB communication itself is noisy and blowing from there. In such cases, the root cause is mostly reflection, I think, but suppressing the reflection leads to a two-birds-one-stone outcome: “the waveform gets cleaner and communication errors decrease” and “unwanted radiation drops.” For example, checking the differential impedance using something like a TDR might reveal where the reflection is occurring.

Communication circuits are very often composed of multiple signals that share the same origin and destination. That means the signals mostly run in parallel. Generally, signals running in parallel are prone to noise coupling. For example, trouble can occur when an analog signal (like audio) and a clock run in parallel and noise gets coupled onto the audio. It’s the same idea. If you have a mental list of high-speed signals that warrant some caution, it’s good practice to keep them as far away as possible from noise-sensitive signals, and if that’s difficult, to design the board so that an R or C can be inserted later.

Strengthening power/GND countermeasures

What you need to think about is “how to gently return the noise generated in the power supply or signals back to GND.”

Reviewing the placement and values of decoupling capacitors

In addition to the GND-strengthening measures already given as examples, “reviewing the placement and values of decoupling capacitors” is also a kind of GND-strengthening measure. Capacitor impedance has frequency characteristics — basically, small-value capacitors pass high frequencies, while large-value ones don’t pass them as much. Roughly speaking, the former helps reduce AC-type noise, and the latter helps reduce DC-type noise.

If the band that’s failing unwanted radiation testing is already known, it can sometimes be effective to change the capacitor values specifically to encourage a “smooth return to GND” in that band. This is because changing the impedance changes the current density distribution. For example, MCUs often have multiple decoupling capacitors hanging off the same potential. Roughly speaking, these decoupling capacitors have two roles: one is supporting the power supply in a DC sense, and the other is lowering the power supply impedance against high-frequency noise. This type of decoupling capacitor is often connected quickly to a fairly large GND, so if this power supply is noisy, adjusting or adding decoupling capacitor values can sometimes lower the noise level.

Adding something in series to power/GND

Noise is current. So if you consume it with a resistor and turn it into heat, the noise itself disappears. That said, if you carelessly increase the impedance component that’s active at the frequencies used for DC or communication, problems can occur in operation. If the noise frequency is far higher than the communication frequency, inserting a filter or ferrite bead in series into the power/GND line to sharply cut that band can be very effective. When applying this idea, you also need to pay attention to whether the noise is coming from outside that power supply system or being generated internally. If it’s from outside, even if the noise frequency overlaps with the communication frequency, you can still cut it off sharply first, and then, inside the power system, build a mechanism that covers short-term power supply by adjusting the decoupling capacitor values. Happy ending. At that point, it also helps to think of a capacitor as something like a battery with an internal resistance that is frequency-dependent — impedance. In other words, a smaller value is suited for short-term, small supply, while a larger value is suited for medium-to-long-term, large supply. Beyond that, it’s a matter of fine-tuning through cut-and-try. If the cause is an AC adapter, USB bus power, or switching power supply noise, hitting it right there can be expected to bring things down quite a lot. It’s preferable to think about what to do primarily at the entry point of the device. On the other hand, from the perspective of the power supply IC side, having each IC contaminated with noise that then gets returned to the power supply IC side is something that should be avoided to the greatest extent possible. Concretely, the simple solution is to cut it with something like an LC filter. And when you think about “which power rail is likely to be dirty?”, the usual suspects that come to mind are the CPU, memory bus, DSP, FPGA, GPU, and so on.

Putting a lid on it

The last resort. This is about after-the-fact countermeasures.

Shielding

This is the method used in many communication modules. Naturally it reduces noise, and it also easily satisfies the requirement under the Radio Act (Japan) that a “module” be difficult to modify, which is why many modules come with shielding.

Sheet metal, if it can be added, is the strongest noise countermeasure part available. However, be sure never to forget to connect that sheet metal to GND. Sheet metal that is floating in potential works as a new antenna or reflector. A classic example of skillfully using something as a reflector is the Yagi antenna.

Making cables twisted pair

Twisted pair is effective when noise generated somewhere is riding on the cable. Cables are basically very capable antennas. Noise riding on a cable usually goes somewhere and causes trouble, so it should be removed. Detailed explanations of the effect of twisted pair are all over the internet, so I’ll leave that to them. Essentially, it’s a Columbus’s-egg trick: by making the electromotive force induced on the cable by noise point in opposite directions at close range, the two cancel each other out.

Other last-ditch tricks

These measures often work quite powerfully, but honestly they’re a headache because they involve more manual work, expensive countermeasure parts, and are often not desirable as a medium-to-long-term solution.

What to do at the start of countermeasures

Building on the ideas for countermeasures given above, what you should do at the start of countermeasures is to clarify, as precisely as possible, what the source is for the frequency that’s failing. Units that have multiple clocks at similar frequencies in multiple locations warrant particular caution.

The steps are:

  1. Enumerate the frequencies present on the board and make an educated guess
  2. If it’s possible to stop something via software, try stopping the clock (or clocking it down)
  3. If it can’t be stopped, cut off the circuit, for example by removing the component

A portable simple spectrum analyzer is handy during this process. The one I use is this one (cheap, but surprisingly usable):

RF Explorer Spectrum Analyzer, Simple Handheld Digital Pocket-sized Spectrum Analyzer (3G Combo)

Know-how for identifying the source location ① “Take it apart”

If you’re using a component where stopping the clock isn’t easy, you may have no choice but to observe it in a fully operating state. If it’s a small device, or if the boards are close together, you often can’t tell where it’s coming from. In such cases, it’s effective to stretch out the cables and physically separate the boards. Take them out of the enclosure, lay the boards out on a desk, and scan over them with the spectrum analyzer. Of course, this doesn’t guarantee a strictly accurate state, so it’s purely a method for pinpointing where the radio wave is blowing from.

Know-how for identifying the source location ② “Probe the spectrum analyzer ‘directly’”

A spectrum analyzer is generally understood as something you attach an antenna to. That’s true, but there’s actually a different way to use it: attach a probe (you can view it normally this way). By directly probing components or GND with a probe, you can observe disturbances in current density in quite fine detail. However, important cautions are as follows:

This might be a bit advanced.

Know-how for identifying the source location ③ “Touch it with your hand”

The human body is actually a fairly capable noise countermeasure part (it has capacitance, and in many cases is grounded), so sometimes you can observe radiation drop just by touching something with your hand. If you find such a spot, that’s the spot to fix. Similarly, sometimes just touching a cable causes a drop, or touching two specific spots at the same time causes a drop. These are big hints when it comes to countermeasures.

Cautions when identifying the source location

What to think about after finding out where the radio wave is blowing from

Once you’ve found where the radiation is coming out, before jumping straight to “okay, let’s fix it!” there’s something you should think through a bit more carefully.

The location of the emission versus the location of the source

First, a very important point is whether the location of the emission and the location of the source coincide.

If they do coincide, the available countermeasures are quite limited. For example, if a processor’s internal clock is emitting directly from the processor, often the only thing you can do is shielding (metal or radio wave absorbing material). That said, this kind of case isn’t seen much these days in CPUs and GPUs (manufacturers have already dealt with it), so basically all you need to keep in mind is that in this case, what can be done is very limited.

Communication path or power/GND

If it’s a communication path, it’s relatively good news. For example, it’s common for radiation to blow from an FFC cable’s wiring. In that case, just wrapping copper foil tape around it as shielding can bring it down quite a lot. If it’s wiring on the board, just applying copper foil connected to GND or a radio wave absorbing sheet can sometimes fix it perfectly. In these cases, the immediate countermeasure is relatively simple, but since the processing is labor-intensive or costly, the board should basically be fixed eventually. For example, for cable wiring, building an RC or LC filter around the connector, or inserting a countermeasure part like a common mode choke, could make it disappear cleanly, and for wiring on the board, appropriately setting up guard GND or inner-layer GND could resolve it. The most troublesome case is when it’s blowing from the power/GND system.

It seems to be coming from a wide area (multiple locations), and the power or GND plane is the culprit

When it’s coming out of a plane, it’s genuinely troublesome. The reasons are as follows:

“Disperse it” means spreading the noise out as much as possible. This means creating a tight connection structure so that current flows as cleanly as possible through the power supply and GND, as already discussed above. There’s no magic solution that resolves this in one shot; it’s steady, grinding work (whack-a-mole).

To avoid falling into whack-a-mole, simulation has long been used.

Murata’s Femtet is a fairly monstrous solution, yet quite reasonably priced. https://www.muratasoftware.com/

“Seal it off” is the “put a lid on the smell” approach already discussed. No further supplementary notes needed.

There’s also a slightly different curveball approach: “deliberately create a bias and hit it there.” Concretely, this means aggressively cutting off the power/GND at high frequencies. By doing this, you can limit where the noise can go, and then hit it there. However, since each power/GND rail gets weaker as a result, doing this on a small board can leave you with no moves left and stuck. It’s quite difficult.

Overall summary

Unwanted radiation countermeasures are often constrained by the stage, schedule, budget, and so on. If there’s enough time and budget, remaking the board is the proper approach. It’s not a bad idea to run a simulation at that stage either (ideally you’d do it from the very start, but it’s hard to spare that effort until the risk actually materializes). If you foresee the risk in advance (say, you’re carrying over the board or module design of a model that gave you trouble before), I have, on a few occasions, gone ahead and thrown in 0Ω resistors wherever I could think of. Then, when unwanted radiation actually fails, you swap those spots for countermeasure parts. So, what usually causes headaches with unwanted radiation countermeasures is mostly the case where you don’t have that kind of margin or advance measures (which makes sense). In that case, you tend to fall into “let’s just stick stuff everywhere and see what happens!” (I did that plenty of times myself when I was starting out), but knowing the principles and overall picture described above can be useful in many ways.

And so on. The countermeasure work itself will involve some amount of feeling around in the dark, but there’s a big difference between groping around with this kind of knowledge in your head versus groping around blindly.

Bonus

L or R

If you’re going to insert something in series to suppress noise, the options that come up are L (= ferrite beads, etc.) or R (= resistor). L has frequency characteristics, so it’s very well suited to the purpose of “cutting a specific frequency.” R itself can become a source of DC-level voltage drop, so at first glance it seems tricky to use.

However, when you use L in series, whether intended or not, and regardless of whether there’s a decoupling capacitor present, an LC filter ends up being formed. This is because the input capacitance of the IC on the far side of the L, the capacitance of the board, and various other C’s are always attached. The frequency characteristics of an LC filter always show resonance behavior.

This means that at some location, the filter can actually work to strengthen the noise (series resonance). What you intended to weaken at one spot ends up strengthened at another spot. The device to soften this effect is the ferrite bead. For example, take this Murata item: https://www.murata.com/ja-jp/api/pdfdownloadapi?cate=cgsubChipFerriBead&partno=BLM03AG102SN1%23 Its impedance characteristics look like this.

Image 1

If you bring this frequency close to the resonant frequency (by choosing the right ferrite bead), you can cleanly cut the target frequency through the combined effect. Think it through properly, and cut it down properly. Conversely, R can be thought of very simply — just calculate the cutoff frequency and build an RC filter, and the noise drops quite obediently. But “obedient” is all you get; you can’t do anything special with it (like creating a peak at a specific location, for example). Figure out the advantages and disadvantages and use each accordingly.

Handling it lot by lot in sequence

It’s a common story to handle countermeasures sequentially across production lots. As you can probably easily imagine, there are situations like: “The 1st lot ships next month, so the only countermeasure we can deploy is radio wave absorbing material. But there’s about half a year until the 2nd lot, so we can take proper countermeasures by then.” In that case, it might be good to form two teams (or maybe just two people in charge), if possible, since the work involved is different. Short-term whack-a-mole is a battle of brute force and stamina. Long-term countermeasures, on the other hand, are an approach of solving it in one or two shots through careful calculation (including simulation) and redoing prototypes, rather than sheer number of attempts. Skillfully combining both approaches, with an eye on the timeline, is know-how for the development field.

Originally published in Japanese at https://clazytech.com/2020/10/364/. Translated with LLM assistance and reviewed before publication.