I stood over 800 connectors in ESD trays, the smell of solder flux still in the air, and thought: this one is on me. If I'd checked the calibration sticker on the network analyzer before our first article test, none of this would have happened.
Here's my role. I'm the quality manager at a contract electronics manufacturer that builds wireless-enabled medical devices. I review every deliverable before it reaches customers—roughly 200 unique items per year, from PCBA first articles to full production batches. In 2024, I rejected about 7% of first deliveries due to specification mismatches. Most of those were caught early in pre-shipment inspection. This one wasn't.
Let me rephrase that: it could have been caught with a 30-second glance at the calibration status on a Keysight PNA. Instead, it turned into a $22,000 lesson about prevention.
How It Started
Last September, we took on a contract to produce a wireless-enabled blood pressure monitor for a new client. The product is fairly straightforward: an arm cuff, a pump, a pressure sensor, and a Wi-Fi module that transfers readings to an app. The interesting part was the RF front end—specifically, the 50-ohm edge-launch SMA connector joining the module's antenna trace to the external antenna. The fixture also involved a short U.FL-to-SMA pigtail from a new supplier, which was one more variable on the table.
The design spec required a return loss of -20 dB or better from 2.4 to 2.5 GHz. If the connector's impedance transition is sloppy, RF energy reflects back into the module. That doesn't just degrade performance; it can create radiated emissions that fail regulatory requirements.
Our test plan covered all the bases. We'd validate the first article with a Keysight PNA, run a pre-compliance scan with a Keysight spectrum analyzer, inspect solder joints under the microscope, and document everything. The lab had three benches, two oscilloscopes, and one network analyzer we had to book time on. The spectrum analyzer usually sat idle that month because the other active programs were all baseband work. A pre-scan would have taken maybe 30 minutes to set up. I didn't bother.
I had two realistic options. Option A: complete first-article validation—dimensional inspection, PNA measurement with fresh calibration, radiated emissions pre-scan. That's two to three days of lab time. Option B: spot-check three boards on the PNA and compare with the RF engineer's simulations. The simulations predicted -18 dB worst-case, so a quick check felt reasonable.
I went back and forth for maybe an hour. Option A was the right call—I knew that. But the timeline pressure from management was real, and I convinced myself the production batch would get a thorough pass later. I chose Option B.
And here's the part I'd rather not say out loud: I didn't check whether the PNA was calibrated. There was a certificate somewhere in the lab, but I never looked at the instrument's calibration sticker. In hindsight, it was likely overdue. Seven months, maybe eight. I'd probably have to pull the log to know for sure.
The Failure
The three samples passed. My log showed return loss values of -17.8, -18.2, and -17.5 dB at center frequency. Close enough to simulation. I wrote PASS and approved the first article.
Six weeks later, the certification lab called. The device failed radiated emissions at 2.44 GHz, with the harmonic signature tracing back to the connector area.
I brought a failed unit back to the lab and re-ran everything. This time I checked the calibration status first, ran a full two-port calibration, and then measured. The return loss at the connector was -8 dB at 2.44 GHz—a disaster. The uncalibrated PNA had been displaying -15 dB. A pass with margin was masking a defective solder process.
The production reflow had been lifting one corner of the connector's ground tab, creating a hairline gap in the ground plane. Just enough for the connector to resonate in the worst possible band.
I used the spectrum analyzer Keysight had provided under our service agreement and ran a near-field scan. Same harmonics, -3 dB over the limit. No ambiguity. We had built 800 units with a bad solder process, and I'd signed the approval.
Now came the decision. Ship the 800 units with a client-side risk acceptance, or rework all of them?
The upside of shipping was hitting the launch date. The risk was a second cert failure: roughly $22,000 in extra test cycles, plus the client relationship. I kept asking myself whether hitting the deadline was worth losing a medical device client we'd spent six months courting. To be fair, the deadline math was actually simple: two weeks late was survivable; a second failure could kill the project. I signed the rework order, and my hand shook a little when I did.
We reworked every unit.
What I Changed
We set up a station on the production floor. Each connector was re-soldered by hand, inspected under the microscope, then checked for ground continuity with a Keysight DuraForce Pro 2 handheld multimeter. Not a glamorous instrument, but exactly the right tool for 800 repetitive checks. I also added the DuraForce Pro 2 itself to the calibration register. It's the kind of tool you stop thinking about because it just works—and that's exactly when it bites you.
After that, I created a 12-point pre-measurement checklist that we still use today:
- Verify the calibration sticker and certificate on every instrument before taking data.
- Run a known-good reference (thru line or calibration standard) before recording any result.
- Log the calibration check in the measurement record, not just the final numbers.
- Do a radiated pre-scan on first articles, even when the simulations look clean.
- Inspect at least five solder joints per lot under the microscope.
- Confirm the connector supplier's inspection certificate matches our incoming goods record.
It sounds boring. That's the point. The checklist has already caught two issues: a bad cable batch in November and a pre-amplifier drift last month. Both would have reached customers otherwise.
Around the same time, a colleague from customer support asked me a question she'd been googling all morning: how to calibrate blood pressure monitor units at home. The search results were contradictory, and she figured I'd know.
The answer is simple. A blood pressure monitor is a measuring instrument—it measures pressure in millimeters of mercury. Like any measuring instrument, it needs to be checked against a known reference on a regular schedule. Once a year is a reasonable guideline. If the reading is off by more than a few mmHg, you recalibrate it or replace it. That's it.
A network analyzer and a blood pressure monitor are enormously different. But the principle is identical: you can't trust a measurement from a device whose accuracy hasn't been verified. The Keysight PNA, the spectrum analyzer, the DuraForce Pro 2, the blood pressure monitor—they all sit on the same side of that statement.
The same colleague asked a few weeks later whether the speed display on her home treadmill could be calibrated. I laughed, but the answer is essentially the same: compare it against a known reference—a stopwatch and a measured distance—and adjust if needed. Calibration is everywhere once you start looking for it.
The Real Lesson
I still get tempted to skip the calibration check on Monday mornings. I catch myself thinking, the lab was fine last week. That's exactly the thought that cost us $22,000.
What I've learned is to trust the checklist over my gut. Prevention isn't a grand strategy; it's a set of unglamorous, repetitive checks that stop small oversights from becoming expensive failures.
5 minutes of verification beats 5 days of correction. The hard part is remembering to take the 5 minutes.
I used to think quality managers are people who knock on your door after the mistake. The role is actually the opposite: my job is to build the door frame so the mistake can't walk in. It's less heroic than a rescue, and the results are invisible when they work. But invisible results still matter.
If you're in production or quality, here's where to start: go look at the sticker on your network analyzer. Check that your reference standards aren't scratched. Confirm the calibration logbook is current. If you find a gap, close it before your next measurement—not after the cert lab call.
It takes 15 minutes. In my experience, that's the cheapest insurance you'll buy all year. And if you don't have a calibration log yet, today is a good day to start one.
