Figure 4. Process to derive the K matrix from the time series of key Hold Time (HT).

The hold time signal a[t] is first divided by a square time window into vectors containing a variable number of hold time samples. These vectors are stored in the set B only if contain more than S samples. For each vector in B a normalized histogram is computed as an approximation of the hold time probability. Finally, the histograms are vectorized and used as columns of the matrix K.