EE - AI/ML Based Methods for Environment Classification Using Wireless Signals

 In many cases we are interested in a device automatically identifying whether it is indoors or outdoors. This may be so that it can transmit on certain frequency bands that are reserved for indoor usage, transmit at higher powers or receive notifications specific to being indoors. GPS coordinates cannot identify accurately whether a device is indoors. However, smartphones today contain a large number of radio interfaces that support the various wireless connectivity modes that we use on a daily basis: Bluetooth, Zigbee, Wi-Fi, and cellular being the main ones. These radio interfaces support a number of frequency bands: the unlicensed 900 MHz, 2.4 GHz, 5 GHz, and 6 GHz bands, and cellular bands in the low (< 1 GHz(, mid (1 – 6 GHz), and high (> 24 GHz) bands. Further, Android phones make available a number of Application Programming Interfaces (APIs) that allow one to extract the signal measurements made by these radios: signal strength, signal quality, bandwidth and band of operation etc. The PI has developed an app, SigCap, that extracts these measurements, exports them, and analyzes the resulting data. Please see https://sigcap.spectrumx.org/ for more details on SigCap and recent papers.

In recent work [1,2,3], we developed Machine Learning based classification methods that were trained on a data-set collected on various phones using SigCap. Our current project is building on that work by collecting new data sets in a variety of environments and enhancing the models to use newer deep-learning methods. We are also building new models to quantify building entry loss experienced by RF signals using the same measurements.

Student’s Role: The student will use the app on a number of different smartphones to collect data around the Notre Dame campus on 5G, CBRS, and Wi-Fi networks. The data will then be curated, cleaned, and extracted into CSV files for analysis of how these frequencies differ in indoor environments versus outdoor environments and examine building loss. By the end of the project, the student will have gained knowledge of deployed wireless networks, pertinent measurements, and analysis of the collected data. Depending on student interest, there will also be an opportunity to add features to the app and contribute to publications and demos.