Figure 3

Representative examples of five decomposition methods. Examples of decomposition results into intersegmental weights (or DMD modes) in column 1 and time dynamics in the center three columns by conventional SVD-based method (row 1), exact DMD (row 2), companion-matrix DMD (row 3), row-type Hankel DMD (row 4) and column-type Hankel DMD (row 5) during a 2.0 km/h walk are shown. The rightmost column shows the reconstructed angle time-series data (the input data was computed by subtracting the temporal average values from the original data). SVD-based method shows similar two principal components in the previous study⁴⁴. Exact DMD shows that only two conjugate modes and one mode because of the data dimension d = 3. In companion-matrix DMD and the two Hankel DMDs, we show only three pairs of conjugate modes that show the highest VAF. Note that all DMDs compute complex-valued DMD modes and time dynamics, but here we only illustrate the absolute value and real part of them, respectively. VAF and reconstruction error of the dominant modes were higher and lower for SVD-based method, column-type Hankel DMD, row-type Hankel DMD, companion-matrix DMD and exact DMD in this order. Sampling frequency was 100 Hz.