Influence of embedding parameters and noise in center of pressure recurrence quantification analysis

Abstract

Recurrence quantification analysis (RQA) can extract the dynamics of postural control from center of pressure (CoP) data by quantifying the system's repeatability, complexity, and local dynamic stability through several variables. Computation of these variables requires the selection of suitable embedding parameters for state space reconstruction (i.e. time delay and embedding dimension); however, it is unclear how the parameters influence RQA variables when examining noisy CoP data. This study evaluated the sensitivity of RQA variables to embedding parameter values and noise level, and assessed methods of selecting embedding parameters for CoP data. Five healthy male subjects maintained quiet stance for 30 s while the anterior–posterior CoP was measured. The effect of noise was evaluated by adding uniform white noise of increasing amplitude to the raw CoP signal. The magnitude of all RQA variables decreased with increasing noise amplitude for all subjects. A sensitivity analysis was performed by systematically altering the embedding parameters for the raw data with and without a selected level of added noise. The key result was that, for all subjects, the RQA variables were sensitive to the embedding parameter values and the level of noise in the CoP data. Finally, the performance of false nearest neighbors and average displacement algorithms for choosing embedding parameters was evaluated. Both methods gave clear and consistent results for all subjects with either raw or noisy data. The results suggest that careful selection of embedding parameters is essential when using RQA to examine postural control based on noisy CoP data.