The Messy, Beautiful Challenge of Processing Wearable Signals

If you've ever built anything around wearable devices, you'll know the feeling: the data looks great in a slide deck, the graphs are smooth, the metrics line up nicely. But then you get your hands on the raw signals and reality comes rushing in. Suddenly it's not a neat line — it's a noisy jungle. A wrist twist masquerades as tachycardia. A loose strap looks like an ectopic beat. Bluetooth drops out just when the interesting thing happens.

And yet, that's where the real engineering challenge — and the fun — begins.

Unlike hospital-grade devices, wearables live in the wild. They're with people in the gym, at the supermarket, in the rain, on the sofa. They're being tugged at by kids, bumped against desks, worn tighter on one wrist than the other. If you're an engineer, you can't control any of this. What you can do is build systems that make sense of chaos.

Motion and the tyranny of artifacts

Movement is the enemy of clean signals. A PPG sensor shining light into skin is a delicate measurement at the best of times. Throw in a jog, or even just a wrist flick to check a notification, and the clean waveform you saw in the lab collapses into spaghetti.

You might think, "Fine, just use the accelerometer to filter it out." But motion sensors themselves are imperfect: they pick up gravity, orientation, and micro-tremors in different ways. Sometimes the accelerometer says "still," while the PPG says "chaotic," and both are right — because the user is holding a shopping bag, standing still, but clenching their fist.

Building a system that can tell the difference between genuine physiology and the noise of everyday movement is one of the core challenges of wearable engineering. And the kicker is: users never sit perfectly still just because you'd like them to.

Physiological variability: no two humans alike

One of the humbling things about wearables is how personal they are. Two people can wear the same device in the same setting and get very different signals.

Skin tone matters. Vascular structure matters. Temperature, hydration, even the angle of the strap matters. Some people just have better peripheral circulation than others, and the signal reflects that. For engineers, this variability makes it really tough to build "one-size-fits-all" algorithms.

What looks like an outlier in one dataset might be completely normal for a given individual. Which means you need systems that adapt, that recognise "this is what normal looks like for this person," while still being able to spot when something genuinely unusual is happening.

Sampling vs. power: the eternal trade-off

This one is as old as electronics: do you sample fast and get beautiful, information-rich signals at the cost of battery life, or do you sample slowly and risk missing something important?

For consumer devices, battery life wins most arguments. Nobody wants to charge their watch three times a day just so the HRV graph is a bit smoother. But in healthcare contexts, there's real tension. Clinicians care about subtle shifts that only show up if you catch the detail. Miss a few beats, or smooth too aggressively, and you can wash out the signal you actually needed.

The art is in making these trade-offs invisible to the user while preserving enough fidelity for downstream analysis. Easier said than done.

Data integrity: the gaps we can't ignore

In the real world, data streams drop out. Straps loosen. Bluetooth cuts out. Batteries die mid-run.

Engineers have to decide: do we throw out whole chunks of data? Do we interpolate and risk inventing false continuity? Do we mark gaps and let downstream systems decide how to handle them?

None of these choices are free. Throwing out data risks missing an event. Filling in gaps risks bias. Passing on the mess pushes complexity downstream. In clinical contexts, those decisions matter even more, because the stakes are higher than whether someone hit their step goal.

Sensor fusion: when more isn't always better

Modern wearables don't stop at heart rate. They measure movement, oxygen saturation, temperature, sometimes even blood pressure surrogates. In theory, the more data the better. In practice, synchronisation and calibration drift turn multi-sensor setups into headaches.

An accelerometer might timestamp in one way, the PPG in another, and by the time you line them up the signals don't quite match. Over hours or days, tiny errors add up. What looks like a cross-signal correlation might just be clock drift.

And then there's the philosophical problem: what do you do when two sensors disagree? If the PPG says "heart rate 90" and the accelerometer suggests "running flat out," which do you believe?

Consumer noise vs. clinical signal

Perhaps the biggest challenge isn't technical at all, but cultural. In fitness apps, trendlines are often "good enough." If your watch says your VO₂ max is improving, you don't mind if it's a bit fuzzy. But in healthcare, fuzziness isn't good enough.

If you're flagging an arrhythmia to a GP, or building a report for a patient's record, you need to be able to stand behind the signal. That doesn't mean every wearable needs to become a medical-grade device. It does mean engineers have to build with a different kind of responsibility. It's not just about making the graphs look nice; it's about ensuring that what you pass on is defensible.

Why this excites us

For engineers, this is the kind of messy, interdisciplinary space that makes you want to dive in. You've got embedded systems, signal processing, human physiology, machine learning, user experience, all tangled together. You're building bridges between silicon and skin, between noisy real-world signals and the sober demands of healthcare.

And while there are challenges everywhere you look — from motion artifacts to clock drift — there's also opportunity. Every time you solve a piece of this puzzle, you bring wearables a little closer to fulfilling their promise: giving people insight into their health in everyday life, and giving clinicians data they can actually use.

That's why we spend so much time thinking about these issues. Not because they're easy, but because they're worth it.