TECHNIQUES
We utilized four in-class and three DSP tools throughout the process of this project
IN CLASS TOOLS
MOVING AVERAGE
Noise filtering
This is the first technique we employed so as to level our signals and obtain a clear picture of the flexes and noises. It did not work well because the signals in their entirety were unforeseeably attenuated, not just the noises. We thereby decided to forgo this method and explore something else.
FAST FOURIER TRANSFORM
Bring signals to frequency
This classic tool helped us visualize the signal in the frequency domain. The initial purpose of this tool was to isolate the noises. However, we gradually discovered that the noises did not have a specific frequency range but instead were distributed throughout the signal graph. Although this made noise-canceling more challenging, it actually paves the way to our discovery of exponential moving average.
HIGH PASS FILTER
Isolation of meaningful data
This simple yet powerful technique introduced in the class assisted us effectively to isolate the sections of the EMG data where flexes occurred. In our EMG data, we discovered that the highest frequencies only occurred during flexes. Implementing a high pass filter, then, isolated the areas where flexes occurred, allowing us to effectively accomplish our goals.
DSP TOOLS
EXPONENTIAL MOVING AVERAGE
Reduce the noise
This technique effectively allowed us to cancel out the unwanted noise. The exponential moving average is a special subcategory of moving average that assigns more weight to most recent values and exponentially (hence the name) diminishes the weights for previous values. The resulting plots displayed manifest peaks and troughs with smoother intermediate parts.
CUMULATIVE SUM
Electromyography Sensor
We employed this technique mainly for the electromyography sensor in our prosthetic arm design. This method is primarily used for detecting flexes, which are then fed to the algorithm that directs the arm. Please see the Cusum page for more detailed outline of this technique.
WEIGHTED MOVING AVERAGE
level the plateau areas
This technique, despite being slightly superior to the standard moving average, did not quite help us achieve the goal of canceling out noise. We tried the weighted moving average with weight being 1/5 and discovered that although it effectually reduced some of the wiggles in the plateau areas, the peaks were also attenuated, which is not ideal for our purposes. Therefore, we opted for an exponential moving average.