Fixing a radiated emissions failure before the lab test

The board works on the bench. Then it goes to a test lab, and the scan shows a forest of peaks above the limit line. The lab can tell you which frequencies failed. It cannot tell you which trace, converter or cable is responsible, and every re-test costs schedule.

Most of that diagnosis can be done at your own bench with modest tools, before and after the lab visit. This note covers what the rules actually ask of a typical digital product, where the energy usually comes from, and the order we would try fixes in.

First, find out which rules apply to your product

Do this before touching the layout. The answer changes how much margin you need.

In the US, a product containing digital electronics is an unintentional radiator under FCC Part 15, and it must be authorized before it is marketed. 47 CFR 15.101 allows either Supplier’s Declaration of Conformity or a grant of certification for most digital devices. A product with a radio in it is a separate matter: the transmitter is an intentional radiator, and a pre-certified module does not remove the digital side of the product from the picture.

Then check the class. Under 47 CFR 15.3, a Class B digital device is one marketed for a residential environment. A Class A device is one marketed for commercial, industrial or business use, excluding anything sold to the general public or intended for the home. The limits in 47 CFR 15.109 differ by class, and the two classes are specified at different measurement distances. If you are unsure which class you are, that is a question for your test lab before you design to a number.

Also check whether an exemption applies. 47 CFR 15.103 lists exempted digital devices, such as those used exclusively in transportation vehicles or exclusively as test equipment, and it is narrow. A device that is partly exempt, or that has a charger that works while the product runs, may not qualify. Read the text for your case instead of relying on a forum summary.

One more fact worth knowing: the frequency range the lab must scan depends on the highest frequency in the product. 47 CFR 15.33 sets the upper scan frequency from the highest frequency generated or used. A product with a fast processor clock gets scanned higher than a product with only a slow one. That is a reason to know every clock on the board, including the ones inside modules.

Where the energy usually comes from

Emissions failures are rarely exotic. They tend to come from a short list.

Switching converters. The hot loop of a buck or boost converter, formed by the switch, the diode or synchronous FET, and the input capacitor, carries fast current edges. A large loop radiates directly. A switch node with a big copper pour acts as a small antenna. We covered a related layout problem in why a boost converter fails when the Wi-Fi radio starts, and the same loop discipline applies here.

Clocks and their harmonics. A crystal or oscillator is a narrow source, but a square-wave clock has harmonics well above its fundamental. Peaks that sit at regular multiples of one frequency point at one clock. Peaks that march across the band at a fixed spacing often point at a bus, such as an SDRAM, an SD card or a display interface.

Cables acting as antennas. This is the most common one. The board itself is quiet, but noise on a ground reference drives common-mode current onto an attached cable, and the cable radiates. USB cables, display ribbons, sensor leads and power cords all do this. A product that passes with no cables attached and fails with them attached is telling you exactly where to look.

Return path breaks. A fast signal that crosses a slot or a split in the reference plane forces its return current to detour. The larger loop radiates more. Layout tools do not flag this, because the connections are all technically complete.

Display and memory buses. A parallel RGB or memory interface running at tens of MHz with unterminated, unmatched traces and long runs is a broadband source. Touchscreen products are especially exposed because the display cable is long, and it leaves the board.

Cheap diagnosis at your own bench

You do not need a chamber to learn which of these is your problem. You need a way to compare before and after, and the discipline to change one thing at a time.

1. Near-field probes. A set of small loop and stub probes connected to a spectrum analyzer or a low-cost software-defined radio lets you sniff individual components. The peak you saw at the lab will light up much brighter near the source. Probe the converter inductor, the crystal, the connector, and the cable.
2. Cable test. Clamp a ferrite on the suspect cable, or remove the cable, and watch whether the peaks drop. A big change means common-mode current on that cable.
3. Power the board from a battery or a linear supply. If the noise disappears, a switching converter was the source.
4. Disable one function at a time in firmware. Turn off the display, stop the SD clock, drop the radio. Firmware can tell you which subsystem radiates, if you build the test switches in.
5. Record everything. Photograph the probe position and note the setup. Near-field readings are only useful as relative comparisons. They are not a prediction of the lab number, and we would not quote one as such.

A near-field probe sweep can show you which component is the source. It cannot tell you whether you will pass at the lab’s distance. Treat it as a way to rank suspects and to confirm that a fix did something.

Fixes, in the order we would try them

Cheap and reversible first.

• Cable and connector. Add or move a common-mode choke or ferrite, and check the shield termination at the connector. A shield tied to ground at the connector, with a short low-inductance connection, is usually better than one tied through a long wire. Check the actual mechanical layout, because an enclosure can change the result.
• Converter loop. Shrink the hot loop by placing the input capacitor against the switch pins and keeping the switch node copper small. Slow the switch edge with a small series resistor or a snubber only if the datasheet and efficiency budget allow it.
• Clock edges. Add a small series resistor near the clock source to soften the edge. Many microcontrollers also allow lower drive strength on I/O pins, which is a free firmware change. Where the part supports it, spread-spectrum clocking reduces peak levels, but check that your peripherals tolerate it.
• Plane and return path. Fix any slot or split under a fast trace. Add stitching vias where a signal changes layers so the return current has a path.
• Stackup. If the board is two layers and fast signals are everywhere, a four-layer respin with a solid ground plane is often the real answer. We would say that plainly instead of piling on ferrites.

Changes made at the lab on a bench-top prototype may not carry over to a production board, which may have different cables, a different enclosure and a different assembler’s parts. Re-check after any change in the supply chain.

When it is worth planning for this on day one

The cheapest emissions fix is a layout that did not have the problem. The practices that matter most are dull: an unbroken ground plane under fast signals, short converter loops, connectors placed together on one edge, filtered I/O at the point where cables leave the board, and footprints for a few optional parts such as ferrites, series resistors and a common-mode choke that you can populate or leave empty. Leaving those footprints costs a little board area and saves a respin.

It also helps to plan the test itself. Know the class, the exemption question, the highest clock, and which cables and accessories ship with the product, and give the lab the real configuration. A scan of a stripped-down unit proves little about the product you sell.

How we approach it

We review the layout for the radiators above, place optional filter footprints where they are cheapest, and suggest the order to try fixes in. If you have a failing scan or a board heading to a lab, send the schematic, the layout files, the lab’s frequency list and a note on which cables ship with the product.