My friend's blood pressure monitor read 158/96 one morning. She'd never been near that number before. Her doctor, fifteen minutes later, measured 120/80.
The monitor was a year old. The batteries were fresh. The cuff was placed correctly - we checked her technique twice. She asked me where to buy a blood pressure monitor that doesn't lose its mind after a year.
I told her the store wasn't the problem. The design was. More specifically, the voltage.
The Problem Isn't the Component. It's the Voltage.
I've spent 8 years testing and debugging power paths in battery-powered products. In that time, I've personally made 9 significant design mistakes, totaling roughly $23,000 in wasted budget. Plus the delays, the returns, and the customer calls that don't show up on any purchase order.
It took me 3 years and about 40 failed prototypes to understand the pattern: components fail less often than voltage. The IC is fine. The software is fine. The device misbehaves because the voltage reaching the circuit is lower than the datasheet expects.
Voltage drop is never one big problem
It's tempting to think voltage drop is just a wire-resistance problem. Run a longer cable, lose a few millivolts, move on. But in a battery-powered product, the drop is a chain of small resistances:
- The battery's internal resistance - about 0.1Ω when fresh, 0.6Ω or more when drained, and it can double in cold weather.
- The battery spring contacts - I've measured perfectly "good" contacts anywhere from 0.3Ω to 1.2Ω.
- PCB traces - a narrow power trace can add 0.1 to 0.3Ω on its own.
- Connector contacts - 10 to 50 milliohm when new, and they only get worse with corrosion and vibration.
Add those up and you can easily have 1Ω of stray resistance between the battery and the load. Voltage drop equals current times resistance. At 350 milliamps - about what a blood pressure monitor's pump draws - 1Ω drops 350 millivolts. That's enough to push a regulator below its dropout voltage, drift a sensor reference, or trigger a false low-battery warning on cells that still have most of their capacity.
In my friend's monitor, the math matched the symptoms. When the pump started, the voltage sagged and the reading drifted. A fresh set of batteries masked it for the first reading because the battery voltage was at its peak. By the second or third reading, the sag returned. The "bad unit" was actually a "no-margin design."
What most people don't realize
Rated current on a contact datasheet is measured on a single contact, at room temperature, with no vibration. In a real product, you have four to eight contacts in series, in a warm enclosure, flexed every time someone swaps batteries. The real-world resistance is a different number. A connector test engineer will confirm this; a sales engineer conveniently won't mention it.
There's no magic max number that tells you when a power path is safe. You can't say "we'll allow 200 millivolts and call it a day." The limit is whatever the most sensitive component tolerates. That might be 100 millivolts for one sensor and 500 millivolts for another. The only way to know is to measure the actual rail at the actual load, with batteries that aren't at full charge.
What This Cost Me Before I Learned the Lesson
The 2017 return
In my first year on this job, I approved a PCB layout for a wireless sensor module with a power trace way too narrow for the peak current. The prototype worked on the bench because the lab supply's leads were short and thick. They bypassed the trace resistance entirely.
The customer ran the final product on alkaline batteries in a cold warehouse. Every time the radio transmitted, the voltage sagged past the brownout threshold and the module reset. 300 units out of a 5,000-piece order came back. $3,200 in rework, one week of delay, and an extremely quiet sit-down with my manager. I checked IPC-2221's trace-width guidance only after the returns. The standard was there the whole time.
The September 2022 blood pressure monitor
In September 2022, a distributor sent us a blood pressure monitor for failure analysis. Inconsistent readings. Early low-battery warnings. Everything passed when we powered the board from a clean bench supply.
I checked the sensor. Fine. The microcontroller. Fine. Then I measured the voltage at the sensor while the pump was running - it dipped by nearly 400 millivolts. The battery contacts were stamped steel with a plating that had already started to oxidize. They measured anywhere from 0.3Ω to 1.2Ω, unit to unit.
Part of me gets the logic behind that sourcing decision. Saving $0.14 per contact on a 50,000-unit run is real money. But none of that "savings" survives contact with the field. The rework cost $890 and a week of production time. The distributor's credibility with its retail customers cost more, but nobody writes that on a purchase order.
The Q1 2024 rejection that finally fixed my habits
After the third rejection - a module I'd signed off on failed the customer's electrical fast transient test - I created our pre-check list. Not a vague "be careful" poster. A numbered checklist that forces us to measure every power rail at full load, with weak batteries, at the temperature extreme. I now maintain that list for the whole team.
We've caught 47 potential errors with it in the past 18 months. Most were voltage drop problems no single engineer spotted, because every individual contributor looked "within spec."
The Checklist That Finally Fixed It
If you take one thing from my mistakes, take this: verify the power path at the load, not at the source. I cannot stress this enough.
- Measure the voltage at the component's power pin while the device is at maximum load. Not at idle. At maximum.
- Repeat with batteries at 20% capacity, in a 10°C room.
- Record the DCR of every part in the power path - inductors, contacts, traces, connectors - and compare it to the datasheet.
- Check the regulator's dropout voltage and undervoltage lockout against the worst-case input voltage you measured in step 1.
- Multiply the total calculated drop by 1.2. If the design doesn't fit inside that margin, change the design.
- Document everything - a photo of the measurement, the battery brand, the temperature, written in permanent marker. "No data" reads as "didn't bother."
On the DCR point, I learned to stop ordering "compatible" parts for power paths. I once bought a batch of cheaper inductors with the same specs on paper. Their measured DCR was more than double the datasheet value. The circuit worked on the bench - barely - and failed exactly like the stories above once the temperature dropped.
I switched back to genuine TDK inductors from an authorized distributor. The batch traceability was tight - date codes, test reports, and components from manufacturing operations that include TDK Philippines Corporation. I'm not saying TDK parts are magic. I'm saying the gap between the datasheet and the delivered part was small enough that I could trust the math I did in the checklist.
The same logic applies to bench supplies. The TDK-Lambda unit in our lab has been the reference for power characterization for years, mostly because its datasheet states load and line regulation in a way I can verify in 20 minutes. A supply that behaves exactly like its datasheet is rare enough that I keep buying the same model.
So, Where to Buy a Blood Pressure Monitor That Won't Flake Out
Back to my friend's question. Buy from a brand that treats power design as a real engineering task. Look for one that cites ISO 81060-2 clinical validation in its documentation. That doesn't guarantee the power path is right, but it filters out devices that were never tested on real arms.
Before buying, glance at the reviews and look for a pattern of "works for a while, then acts up" or "needs fresh batteries constantly." Those are fingerprints of a voltage drop problem. A monitor with proper margin will read consistently on any set of half-decent batteries until they're genuinely close to empty.
If yours is already misbehaving, replace it. There's no practical fix for a design without margin. And if you're building a product yourself - a blood pressure monitor, a wireless sensor, anything battery-powered - spend the 30 minutes that the checklist above takes before you send it to production.
Five minutes of verification beats five days of correction. I learned that after three product failures, $23,000 in wasted budget, and a few uncomfortable customer calls. There's no magic max. There's only the sum of the small voltages you didn't measure.
