Chemical Sensors Notes FL2010

• Team Members: Jessi Mischel, Anisha Rastogi
• PhD Supervisor: Vanessa Tidwell
• Faculty Supervisor: Arye Nehorai
• Goal:
• Meetings
  • Thursday 8:30 AM - 10:00 AM

---

• Status

• • Week 1 -- created a modified Elvis subVI to replace a lost subVI used during the summer; created boolean indicators in sixsensors VI to enable us to determine our location in the programming loop; began running experiments in order to determine optimal sensor temperatures at which chemical sensing data can be collected.

• • Week 2 -- This week we are working with modifying the data analysis matlab programs created this summer to analyze our collected data. It is now able to call files from this semester's folder, however we are still going through the code, commenting, and making sense of it all.

• • Week 3 -- This week was a sad week. After fixing some glitches in our data processing code, we have determined that our current method of data analysis gives completely insignificant results. Leaving the lab with heavy hearts, we have decided that for next week we are to try a new method of data analysis.

• • Week 4 -- We started out the week by redefining our code to find peaks rather than zero crossings. we did this by using the standard deviation of the signal and determining how many peaks cross this threshold. We are still playing with this method, as that some signals are showing no peaks while we know that for every signal there must be at least the one initial peak. We edited our method to look at the signal rather than the derivative and eliminated this problem. However, now we are seeing far too many peaks since the signal is so absurdly noisy. We are currently working on designing a filter that accurately smooths our signal without losing any data to eliminate this issue. This is a challenge.

• • Week 5
  • Week 6
  • Week 7

• • Interlude: After several weeks of work, we developed a peak detection system in Labview instead of in Matlab because Labview had more peak-detection tools to offer. We then decided to determine which sensor in a pair peaked first, rather than determining which of the two sensors in a pair had more peaks, in order to detect which side of the sensor array the alcohol source was located. The former approach seemed more logical, and may be reliable. We attempted to perform this processing in Matlab by creating our own processing algorithm. Since the results didn't look that promising, we used a different algorithm, which uses Dynamic Time Warping. We are still working out the glitches in the Labview and the Matlab code. Our goal is to use the results from the Dynamic Time Warping algorithm to determine whether errors arise in the signal due to Labview problems, or due to the signals themselves. Once we determine this, we hope to switch back to our first processing algorithm, if possible.

• • Week ? --

• • • We de-bugged our Labview peak-detection algorithm to see why it wasn't detecting peaks. It turns out that the peak detector was detecting some troughs, and the valley detector was detecting some peaks. We temporarily corrected for this by increasing our window size and decreasing tolerance, but we want to put in a high-pass filter into the Labview code to eliminate high-frequency noise and improve the peak detection permanently.

• • • We analyzed Dynamic Time Warp results, and determined that some sensor pairs always produced positive delays, and others always produced negative delays. For sensor pair 3, it always produced positive delays for old, but only produced negative delay for new for closer distances.

• • Week ?? --

• • • Implemented a band-pass filter into our Labview code to eliminate high-frequency noise and get better peak detection. We managed to arrive at filter parameters that preserved the shape of the signal: Low cutoff: 0.5 High cutoff: 0.55 Taps: 301 This filter compromised the amplitude of the filtered signal, but Ed told us that this wouldn't distort the peaks and troughs, so we thought the reduced amplitude would not have a significant effect on our peak detection. We still have to confirm whether the updated Labview code detects peaks properly.

• • • Replaced old processed files with new processed files created from the updated Labview code.

• • • Ran the Dynamic Time Warp algorithm on our new processed files. We are still observing the same patterns as before in the results from this algorithm.