OUR DATA
We took a two-step approach to fulfill this requirement: collecting data ourselves with an EMG sensor and looking for online datasets. Because one of our team members (Tyler) had developed a custom-built EMG sensor for another engineering class, our team had access to data that we can collect ourselves. The data collected by the custom-built EMG sensor was first passed through a bandpass and a notch filter, followed by a root mean square operation, then lastly passed through a low-pass filter (which was designed in MATLAB using the FDA tool) in sequential order. The following lists show the filter specifications for the bandpass, notch, and low-pass filters applied to the data.
Band Pass and Notch Filtering:
60 Hz Notch Filter
50Hz - 150Hz Bandpass Filter
1000 samples/second
Low-Pass Filter:
48000 Hz Sampling Frequency
9600 Hz Passband
12000 Hz Stopband
We then used MATLAB to plot the data in the time domain and frequency domain, and also plotted a spectrogram as seen in Figure 2. While the bandpass and notch filters were carefully applied to the raw sensor output to provide meaningful data (by Tyler and her bio-engineering colleagues), the low-pass filter was designed and applied to the data to demonstrate what types of filtering our team would do as one of the later steps in our project. Consequently, the output of the low-pass filter shows a phase shifting of the RMS Bicep Flexion Data as seen in Figure 2, which is an undesirable behavior that our team will correct as the project progresses.
In addition to the data collected with the custom-built EMG sensor, we also looked for online datasets. In particular, one dataset that we found was collected by The University of Copenhagen, which consisted of 16 subjects with either normal or myopathy EMG data [1]. The data was all from a bicep performing isometric contractions. From these 16 patients we were able to collect 408 EMG signals. The number of subjects and trials provided in this dataset would be a great foundation for developing a classifier (see “Next Steps” section). This online dataset was used to complete the second task for the EMG signal processing project.
Figure 2
EMG data taken from a custom made EMG with bandpass and notch Filter (top), root mean square of EMG data (middle), and root mean square sent through a low-pass filter (bottom). The output of the low-pass filter is phase-shifted, which is an undesirable result, but the oscillations near the maximum magnitudes of the plot were nearly eliminated as expected.
