Frequency Band Analysis of Multiple Stationary Time Series

Figures

• 

  FIGURE 1. (Top) Detrended, variance‐standardized stride interval time series from four participants in the gait analysis study, one from each group that corresponds to a neurological condition: ALS, Huntington's disease, Parkinson's disease, or healthy controls. (Bottom) Estimated normalized power spectra for the detrended stride interval time series from each participant, sorted by the corresponding neurological condition in the top row.

• 

  FIGURE 2. (Top) Underlying spectra from a sample realization of each model. (Bottom) A representative time series realization from each subpopulation from the corresponding model in the top row.

• 

  FIGURE 3. Distribution of estimated frequency band boundaries colored according to subpopulation assignment within the corresponding model in the top row. Here, T=500, Kj=20, J=3 and L=3 for Model 1 and 2(a) through 2(c) and L=4 for Model 3.

• 

  FIGURE 4. Multi‐taper estimated normalized power spectra (see Section 1) sorted by subpopulation along with the associated frequency bands determined under the second mode of summarization described in the text.

• 

  FIGURE 5. (Left) All 61 multi‐taper estimates along with the frequency bands and average summary measures suggested by FBAM to characterize the entire population. (Middle) Low‐frequency (LF) replicate‐specific collapsed measures of power by neurological condition computed using the frequency bands shown in the left plot. (Right) Logarithm of Kruskal–Wallis p‐value as a function of frequency band boundary. The dashed horizontal line corresponds to p=0.05, and the solid vertical line marks the FBAM boundary seen in the left plot.