Throughout the past several weeks, each of us have been working on a different approach to refining the algorithm, so this week we focused on integrating these different approaches into one final pipeline.

 

Overall, our final estimates for tilt were relatively close to the truth values provided by Verizon. Although the error was more than what would be ideal, our project’s purpose is ultimately to provide a proof of concept that engineers at Verizon could expand upon with more resources.

Although estimates for azimuth were further from the truth values, however we noticed that the predictions were groups together in the same groups as the true antennas. We believe that the reason for this is less with the data itself, and more how that data is presented. Our algorithm loses the information of which direction of the plane of best fit corresponds to the front of the antenna, causing there to be multiple ways to calculate the azimuth of a single plane. So planes with similar angles all get calculated the same way, but if they become too different, our algorithm uses a different calculation.

Ultimately, knowing that Verizon has their own team working on the same problem, but with a different approach, we decided that our time would best be spent improving the aspects of our process that are unique, as opposed to having to spend a large amount of time implementing something that Verizon engineers likely already have.

With our pipeline just needing the finishing touches, the last step is just preparing for FDR!