Recurrence plots are a useful preprocessing tool since they provide a
comprehensive view of the dynamic courses within a signal. A
recurrence plot is a 2-dimensional N x N pattern of points where N is the
number of embedding vectors obtained from the
delay coordinates of the input signal.
A point i,j in this plot is set if
with being the set of M nearest
neighbours of ; i.e. if is one of the
nearest neighbours of and if it lies within a
sphere of radius r around the current reference point.
Since the coordinates i and j represent points on a time axis,
recurrence plots give information about the temporal correlation of
phase space points.
From the occurrence of lines parallel to the diagonal in the
recurrence plot it can be seen how fast neighboured trajectories
diverge in phase space. Therefore, the average length of these lines is a
measure of the reciprocal of the largest positive Lyapunov
Recurrence plots help revealing phase transitions and instationarities.
Visible rectangular block structures with a higher
density of points in the recurrence plot indicate phase transitions
within the signal. If the texture of the pattern within such a block
is homogeneous, stationarity can be assumed for the given signal
within the corresponding period of time.
- number M of neighbours
- step size s (i.e. size-1 samples are skipped between successive columns of the plot)
- embedding dimension m
- relative radius r
Usually, the relative radius ranges from 0.05 to 0.2, and M is
typically between 10 and 50.
Applying the function to an continuously instationary signal, you will get a plot
with points accumulating near the diagonal. (e.g. Recurrence plot of the sample signal sinchirp.dpa)
y = RecurrPlot(x,M,s,m,r);
Note that y is a 2D plot.
False nearest neighbors (FNN), Largest Lyapunov exponent (LLE), Mutual information.
Eckmann et al.