← Back to the chapter · Next detail →
Technical detail · Signal Preprocessing Contracts
Statistical changes and sampling limits
A moving average reduces variance. Interpolation creates synthetic samples, detrending removes low-frequency content, and downsampling removes the ability to represent frequencies above the new sampling limit. These may be useful changes, but each can bias the feature or hide uncertainty if treated as a direct measurement.
Aliasing occurs when sampled high-frequency content appears at a lower frequency. Downsampling a 100 Hz vibration component to 20 Hz without a low-pass filter can turn that energy into false evidence of slow behaviour. The pipeline may look smoother and cheaper while its feature becomes misleading. Filter out content the new sample rate cannot represent before reducing the rate; later cleaning cannot recover which slow-looking samples were aliases.
Upsampling invents points by interpolation and can imply detail never measured. Neither increasing nor decreasing the rate creates information that was not sampled. A wide smoothing window can erase genuine features just as a poorly chosen trend-removal step can remove relevant slow content. Match each operation to the event’s timing, shape, and frequency content.
Retain validated raw data because preprocessing is lossy and its choices will be revisited. Keeping the source lets you reprocess with different parameters instead of being stuck with one irreversible result. The aim is to clarify the signal without manufacturing one.