Algorithm 1: Obtaining Refined Signals

For source input signal, reconstruct a phase portrait—obtain M vectors of state variables. Obtain clusters of M-dimensional vectors (e.g., for a 3D problem, a clustering vector is [position, velocity, acceleration]). For each cluster, select subclusters of near vectors, i.e., select the M-dimensional vectors closest to the center of their cluster. For each subcluster, select only those M-dimensional vectors that are distant enough among them in the time domain. Randomly select subclusters (containing at least two vectors). The number of selected subclusters should be large enough to satisfy the condition about the total size of samples. Generate refined signals—subsequences (substrings from the source input signal) that start from the initial vectors taken from selected subclusters.