There’s a project I bring up almost every time a client waves off certification as a formality for the end. It was an environmental sensor. Solid product. It worked flawlessly on every bench I put it on. Then it went to the certification lab and failed. We changed something, sent it back, it failed again. Then again.
The product was fine. The problem was that it radiated like a small radio station the moment it sat in the test chamber.
It took real digging to find the cause. It was the display cabling. The ribbon feeding the screen carried a set of parallel digital lines, and at the screen’s refresh rate the whole harness behaved like a tuned antenna. We made several hardware changes. But the fix that finally moved the measurement was in firmware: we added a slight randomization to the display refresh so the emitted energy stopped stacking up at one resonant frequency and smeared across a wider band instead.
Nobody designs a ribbon cable to be an antenna. That is the whole point of this article.
What EMC certification is actually checking
EMC stands for electromagnetic compatibility. Strip away the acronym and it is two promises: your product does not flood the airwaves with noise, and it keeps working when other devices flood the airwaves around it. The first half is emissions. The second is immunity. Most first products get stuck on emissions, so that is where I will spend the words.
Emissions split into two categories, and the difference drives a lot of decisions.
Intentional emissions come from radios. WiFi, Bluetooth, LoRa, anything that transmits on purpose is an intentional radiator and falls under the strict part of the rules (FCC Part 15C in the US, the RSS standards from ISED in Canada).
Unintentional emissions are the ones that catch people. Every digital circuit you build is a small transmitter you did not mean to design. Clock lines, switching regulators, display buses, motor drivers, they all leak energy. My sensor passed every functional test and still failed, because a cable quietly turned into an antenna. That is the normal failure mode, not the exotic one.
There is a detail here that changes the math: if a product has no radio at all, it only has to clear the unintentional-radiator rules, which are lighter and in many cases can be handled by a supplier’s declaration instead of a full certification campaign. Disabling or removing the wireless function genuinely changes which rulebook applies to you.
When you actually need it
This is where I see clients split into two camps, and both are partly right.
One camp wants to keep EMC for the very end. Certification gates commercial sale, not engineering, so technically they can develop the whole product, build prototypes, test on the bench for months, and never book a lab until launch is in sight. They have a real point: you do not need certification to prototype, and chasing it too early burns money testing a design that is still moving under you.
The other camp wants to be fully certified from day one. They have usually been burned before, or they have a hard launch date and no room for a surprise. They also have a real point: a late EMC failure can freeze a launch for months, and by then the design is too rigid to fix cheaply.
Neither extreme is a strategy on its own. The right timing depends on the radio content, the launch pressure, how much the design has stabilized, and how much certification risk the product is carrying. A simple sensor with a pre-certified module is a very different bet from a dense board full of fast digital and a custom power stage.
What I do for most projects is keep EMC on the backburner from day one without letting it dominate the early work. The early effort goes into design choices that cost nothing now and save a respin later. Then certification gets scheduled deliberately, when the design is stable enough that a pass actually means something, and early enough that a failure still leaves room to react.
What the numbers look like
In the greater Montréal area, a day in a certification lab runs around 1200 CAD. A full emissions sequence, if nothing major goes wrong, lands near 10k CAD. The “if nothing major goes wrong” is doing heavy lifting in that sentence. My environmental sensor was several failed days stacked on top of each other.
And once you pass, the design is frozen. No more BOM changes, no swapping that regulator, no layout tweaks. The certified configuration is the configuration. A late component substitution can quietly invalidate the whole report. This is also why the “certify on day one” camp can get burned: certify too early, change one part, and you may have certified a product you no longer ship.
The moves that stack the odds
Use pre-certified modules. An ESP32 module with a modular grant means the radio is already certified. You inherit that approval instead of testing the radio yourself. Same logic for power: buy an AC-DC brick that already carries its own certification rather than designing your own offline supply and owning that risk.
Design in mitigations before you need them. This is where an experienced team earns its rate. We are not magicians, and nobody guarantees a first-pass result. What we do is shift the probability. We have seen the failure modes, so we plan for them: zero-ohm resistors placed in series on suspect lines so they can be swapped for 22-ohm parts to slow down edges, empty footprints left for ferrite beads, layout choices made early to handle ESD. None of that costs anything if it is on the board from the start. All of it is painful to add after a failed test.
Run pre-compliance early. You do not need a full accredited lab to get a strong signal on where you stand before you book the expensive chamber time.
And one move that surprises people: certification reports are public. The FCC maintains a searchable database of every certified product. Test reports, internal and external photos, often block diagrams. You can look up a competitor’s device by its FCC ID and see how they built it and how it performed. It is one of the most useful and least used research tools in hardware.
The honest version
EMC is not a formality you bolt on at the end, and it is not a box you need stamped before the design has settled either. It is a property of the design, decided mostly by choices made long before anyone books a lab. The clients who struggle are usually the ones who picked a camp, end or day one, and stopped thinking about it. The ones who pass cleanly, on the schedule they wanted, treated the timing as a decision they made on purpose, not a default they inherited.