Frequency Response Analysis 

We analysed the frequency domain of data from 6 different users collected in identical physical environment. The data is visualized in the following graph where amplitudes of the signals are normalized, such that the maximum amplitude from each gesture is 1. The graph represents the distinctness of our 6 tested gestures in the frequency domain. For "pluck" gesture, one observes a good response in the frequency range of 25 to 45 Hz, which corresponds to the range of frequencies wherein lies the frequency of first harmonic, given the length of the plucked part of the sensor. This response in pluck gesture promises future application of self-powered pickup in string based musical instruments.

In all gestures apart from stretch and pluck, one notices a peak at 60Hz, which is due to the omnipresent electrical power transmission lines. The amplitude of the signal in pluck and stretch is much higher than the signal intercepted from the 60Hz power line, because of which the 60Hz signal is suppressed in the visualization when the data is normalized.

Spectrograms for revising classification algorithm

In the process to debug and improve the machine learning classification, we graphed spectrograms of different gestures to observe which gestures are best distinguished using that feature. Spectrograms also influenced our decision about whether a gesture should be kept in the set of distinguishable gestures for this first study of its kind. Therefore, visualizing features influenced gesture design and vice versa. We are constantly improving our classification algorithm to increase the number of gestures in this first set of 6 gestures.