Signal Processing
Signal processing is the craft of turning a noisy, contaminated signal into something trustworthy. Filtering removes unwanted components, fusion combines complementary sensors into an estimate better than any alone, and every step trades some information for clarity. Understanding what is removed and why is the difference between trusting a number and being misled by it, which is why this stage sits at the heart of how movement technology earns its accuracy.
Signal Processing
Filtering
Removing what you do not want
Filtering separates the signal of interest from noise and interference. A heart rate trace is smoothed to suppress motion artifacts; a motion signal is filtered to isolate the movement that matters. Filtering is powerful but never free: aggressive smoothing hides noise at the cost of also hiding real, rapid change.
Sensor fusion
Combining complementary sensors
Fusion combines several imperfect sensors into a single estimate better than any alone, weighting each by how credible it is at a given moment. Combining an accelerometer, gyroscope, and magnetometer yields orientation none could report on its own. Fusion improves on its inputs but cannot manufacture information they never captured.
The fidelity trade
Trading detail for clarity
Every processing step discards some information on purpose, exchanging fidelity for a clearer result. This trade is unavoidable and usually beneficial, but it is a choice with consequences. Knowing where in the chain detail was given up clarifies what a final metric can and cannot support, and guards against reading more into it than the signal holds.
Honest limits
What processing cannot recover
No amount of processing recovers information that the sensing stage never captured or that motion genuinely destroyed. Good signal processing makes the most of what is there and is honest about what is not. Treating it as a source of perfect truth, rather than a principled combination of flawed signals, leads to overconfidence in the result.