Machine Learning Methods for Modeling Spectrum Occupancy

Principal Investigators: Dr. Michael Lemmon, Department of Electrical Engineering, and Dr. Bert Hochwald, Department of Electrical Engineering

Project Summary: We seek a student to help model spectrum occupancy over geographic regions as small as buildings and as large as cities. We are analyzing scenarios where a random number of wireless radio transmitters are randomly placed in unknown locations, and with the help of a few sensors that measure signal energy, we can estimate which areas are being reached by these transmitters. Such areas are called “occupied.”

It is especially important to estimate occupancy in the areas that have no sensors from our knowledge of the areas that have sensors. We have found that machine learning methods are especially helpful with this task, where encoder-decoder neural network structures are trained with a large database of scenarios of randomly-placed transmitters and sensors, and where the spectrum occupancy is known in advance, thus providing a training set. 

Student’s Role: We, therefore, seek a student that preferably has:

1) Basic knowledge of machine learning methods

2) Basic knowledge of wireless systems

3) Programming skills (Python, Matlab)

The student will be responsible for helping to train and assess the performance of the machine learning models, and possibly help build a dedicated GPU-based machine to minimize run time.