Figure 7

Equation 1 is used to create a synthetic signal y ₁ with sampling frequency of 8 kHz: (a) Signal y ₁ , ( α,σ,μ,a ) for each component from 1 to 7 is as following: (6,0.001,0.42,2Π3600), (1,0.05,0.68,2Π2600), (1,0.05,0.68,2Π600), (6,0.001,0.80,2Π1700), (1,0.04,0.93,2Π1000), (1,0.03,1.18,2Π2600), (6,0.001,1.27,2Π1700). (b) Signal y₂is generated by replacing the 5th and 6th components of signal y₁with two chirps with slopes of 1 and 0.5 kHz/s, respectively. (c) TFD of y₁. (d) TFD of y₂. (e) TF feature plane including the feature points. Although NMF of order 10 was applied to TF decomposition of each signal, 3 of the y₁decompositions ($\left[{\overrightarrow{w}}_{(}i\right){\overrightarrow{h}}_1^{\text{T}}\left(i\right)]$) did not have any significant content ($\mathrm{energ}{y}_i/\mathrm{energ}{y}_{y_1}<1\%$) so they were excluded from the analysis. D_{y 1}and D_{y 2}were the clusters identified as the discriminant features in each signal. (f, g) The TFD structures selected as the discriminant patterns in y₁and y₂, respectively.