Validation & Limitations
BioSignal Workstation has been cross-checked against MATLAB reference workflows and a supplemental independent SciPy exported-output suite for tested browser-based biosignal visualization and exploratory analysis workflows.
Core browser outputs were compared with independent reference workflows under matched settings.
Checks covered filtering, PSD, band power, statistics, spectrograms, and export behavior.
Selected public physiological recordings were included as practical signal-handling sanity checks.
The evidence supports educational and exploratory research use, not diagnosis or treatment decisions.
What was validated
Core DSP and export behavior
Filtering
Butterworth low-pass/high-pass design, filtfilt cutoff compensation, cookbook notch behavior, zero-phase timing, and waveform agreement.
Spectral analysis
Welch PSD scaling, band power, epoch screening, STFT/spectrogram power matrix behavior, and renderer contract checks.
Data quality
Missing-sample reconstruction, gap metadata, warning behavior, and blocking of unsafe nonuniform timestamp spectral analysis.
Reproducibility
Official CSV/HTML export metadata, large-file stress checks, real-data sanity checks, and automated regression checks.
Validation method
Matched references, not marketing numbers
MATLAB is the primary validation reference for the numerical phases. The checks use matched settings for the tool's actual algorithms, including cutoff compensation, periodic Hann Welch windows, per-segment mean removal, cookbook notch design, STFT scaling, and export metadata.
A supplemental SciPy package independently reproduces selected official exported outputs under matched tool-equivalent settings. This provides an additional cross-check for exported workflows, while MATLAB remains the primary validation reference.
The validation does not compare against default MATLAB or SciPy settings when those defaults are not the tool's intended method. That avoids false failures from method differences and focuses on whether the browser result matches the documented implementation.
Coverage snapshot
Validation at a glance
The public summary below is intentionally high-level. Detailed evidence is retained for release review and regression tracking, not as clinical evidence.
Filters and timing
LP/HP response, notch response, waveform agreement, and zero-phase peak timing passed.
PSD, bands, and STFT
Welch PSD, band power, statistics, and spectrogram power behavior passed matched-reference checks.
Gaps and time axes
Gap warnings, interpolation behavior, and nonuniform-time blocking passed safety checks.
Exports
Official CSV and HTML export paths preserved sample data and required reproducibility metadata.
Large-file handling
Stress checks covered large single-channel and multi-channel browser processing cases.
Regression checks
Automated reruns confirmed that the tested outputs remained consistent with accepted validation evidence.
Real signal sanity checks
Not synthetic-only
Public physiological datasets were included to confirm practical signal handling beyond synthetic cases. These checks are limited DSP consistency sanity checks, not clinical interpretation validation.
MIT-BIH ECG
Record 100, channels MLII and V5, first 5 minutes at 360 Hz.
EEGMMI EEG
S001R01, channels C3, Cz, and C4, first 60 seconds at 160 Hz.
EMGDB EMG
Healthy tibialis anterior EMG, first 10 seconds at 4000 Hz.
What this supports
A bounded, defensible claim
BioSignal Workstation is a MATLAB/SciPy-cross-checked browser-based tool for biosignal visualization and exploratory analysis under the tested workflows and versioned validation evidence.
What this does not support
No clinical claims
- -Not diagnostic and not clinical decision support.
- -Not a medical device and not regulatory-approved software.
- -Not proof that every possible file, browser, device, or future feature is correct.
- -Not a substitute for expert review, domain-specific validation, or institutional data-governance requirements.
Known limitations
Validation is a living process
The validation is strongest for the covered filter, spectral, export, gap-handling, stress, and regression paths. New algorithms, file formats, metrics, presets, or analysis workflows require new validation cases before they should be described as validated.
Real-data coverage includes selected ECG, EEG, and EMG sanity checks but does not represent the full diversity of physiological recordings, sensors, populations, artifacts, or clinical conditions.
Performance results depend on browser, hardware, operating system, and file size. Users should confirm sampling rates, units, channels, and warnings before interpreting any visual or exported result.
MATLAB and SciPy checks use matched methods when the tool intentionally differs from default library settings. This validates the documented tool behavior; it is not a claim that defaults from every package would produce identical output.