Analyse a balance trace

Prepare ML/AP displacement data, choose processing settings and export a reproducible analysis.

1. Try the example

Select “Try recorded example”. The example contains 7,200 samples at 100 Hz. Run it unfiltered, then select Symlet-8 wavelet filtering and run again. Compare the path lengths while keeping the graph scales fixed.

2. Prepare your file

Use a CSV, semicolon-separated or tab-separated text file. Select the columns containing medial–lateral (ML) and anterior–posterior (AP) displacement. Specify whether the first row contains column names. Check the sample rate and units against the acquisition record.

ML,AP
1.2,2.1
1.3,2.0
1.1,2.2

Format illustration only; analysis needs at least 16 rows, wavelet filtering 211, the four wavelet bands 512 and DFA 128.

Samples must be evenly spaced. The tool does not infer a sampling rate from timestamps, repair gaps or resample irregular Wii Balance Board output. Export or resample to a uniform time grid using a documented acquisition workflow before importing.

Displacement units may be mm, cm or m. Analysis results and exports use mm; the interactive plots display cm. Vertical force and other unused columns are not analysed. Maximum file size: 5 MB and 100,000 samples.

3. Choose processing

• None: unfiltered displacement.
• Moving average: centred odd-length window with reflected endpoints.
• Butterworth: order 6, forward/backward; choose a cutoff below half the sample rate.
• Symlet-8: undecimated level-3 approximation. Its nominal upper band is 6.25 Hz at 100 Hz; it changes with sample rate.

The four wavelet-band traces use a separate, documented Symlet-8 decomposition. DFA uses the original unfiltered displacement, so changing the display filter does not silently change the DFA input.

4. Read the plots and results

The COP sway trace shows ML horizontally and AP vertically, with equal spatial scaling. Axes initially fit the data, with equal ML/AP spans; you can switch to −5 to +5 cm. Time-series graphs show the selected displacement axis against seconds. Changing displacement limits updates every relevant graph.

Path length sums the distance between samples. Mean speed divides that path by (N−1)/sample rate. Filtering can shorten the path by suppressing rapid fluctuations; this is a processing effect, not evidence that a person’s balance improved.

5. Retain the analysis

Download the filtered CSV and results JSON. The reproducibility record includes the input checksum, software and method versions, processing settings and numerical results. Keep the original file with these exports. The workspace itself does not retain files after you leave.

Check your interpretation

A filter reduces the recorded trace’s path length. What does that establish?

Scope and sources

This is a new research implementation, not a validated reproduction of the original LabVIEW software. Wavelet boundary treatment and DFA settings are stated explicitly. Neither a wavelet band nor a DFA exponent identifies an individual physiological deficit.

• Original SeeSway paper (2018)
• Wavelet application in ACL reconstruction (2014)
• Input/output specification
• Python calculation source