What problem does it solve?
Preparing raw signal data for analysis or machine learning in MATLAB involves many error-prone steps: filling gaps, removing drift and outliers, denoising, resampling to a uniform rate, aligning channels, deriving labels, splitting datasets, and framing long signals. This Skill routes each task to the correct Signal Processing Toolbox function and workflow, preventing hallucinated function names and hand-rolled loops.
Core Features & Use Cases
- Signal Conditioning: Fill NaN gaps with fillmissing or fillgaps, detrend polynomial drift, remove outliers with filloutliers or hampel, denoise with smoothdata or sgolayfilt, and resample or align channels with resample, retime, and synchronize.
- ML Data Pipelines: Build signalDatastore pipelines, derive labels from filenames or folders with filenames2labels and folders2labels, create stratified train/val/test splits with splitlabels, and frame long signals with framesig and framelbl.
- Structured Labeling: Create labeledSignalSet containers with attribute, ROI, point, and time-frequency ROI labels for Signal Labeler export and trainnet hand-off.
- Use Case: Given a folder of sensor .mat files with class names encoded in subfolders, build a signalDatastore, derive labels with folders2labels, split stratified with splitlabels, and hand off combined datastores to trainnet.
Quick Start
Ask the agent to clean a noisy signal by removing outliers and drift, then load a folder of signal files into a labeled, split datastore ready for trainnet training.
