ACHIEVEMENTS
SELECT SENSORS
We have been able to select the best sensors for our purposes after comparing a variety of different options and comparing outputs between the options best fit for our design. After subjecting the sensor to rigorous conditions, we have found that an IR line detector paired with an EMG sensor is the best fit for control at the bicep or forearm and will allow for a tremendous decrease in price and increase in quality.
DATA COLLECTION
Several sets of data were collected with the IR sensor at 20 samples per second. Each scenario with different possible sources of noise. The data was plotted and analysed on the Data section of the website. Data has been collected with the EMG sensor as well. The sensor setup looks promising!
DATA ANALYSIS
Analysis of the signals in the frequency domain allowed us to better understand the content of our data. While we discovered that the Fourier transform itself was not directly useful, this analysis paved the way for the DSP tools we eventually decided to use.
CHOOSING DSP TOOLS
There are a few tools available to us that we've looked into. First, we've found that the exponential moving average gets rid of low amplitude, high frequency noise. In addition, the leaky integrator has been a useful tool in removing the noise from our EMG data.
FLEX DETECTION
We delved into one of the papers recommended by the professor about quickest change detection. However, we decided against pursuing the algorithms outlined there. Instead, we are detecting change using a couple different methods. For the IR sensor, change is detected using a "threshold" method, and for the EMG sensor change is detected using the cumulative sum algorithm. Both of these algorithms are discussed in depth on earlier pages.
FILTERING OUT NOISES
Through trial and error and searching for various papers, we were able to discover the most suitable filtering techniques that effectively smooth the signals and enable the threshold method. Some techniques were introduced in class and some are from the papers we found. See the filtering page for more detailed reasoning.
CHALLENGES
Struggles We Encountered
COLLECTING DATA WITH A SPONGE AND PHOTOSENSOR
We encountered some hurdles when collecting data with a sponge and the photosensors. Since the apparatus involves a strap around the entire arm, the detecting process is not localized to a singular muscle. For instance, when the arm is bent, other parts of the arm, such as Brachioradialis, also flexes. This causes a problem because when the user is not sending out any muscle signals deliberately the sensor will still pick the signals up and implement them. Most of the time, this is not an issue, however, while collecting the data, it was noted that bending at the elbow also caused a compression of the sponge, and thus an unintended signal. We found that the best way to overcome this was to place the sensor on the outside of the arm around the upper tricep. Having the sensor in this location allows the user to bend their elbow without having our DSP tools recognize this as a flex.
