Margaret Rooney

The control plane of a communications network relays information about the topology of nodes, routing paths of data, and the availability of frequency bands. The ability to identify control information can be useful in the development of cognitive radio networks where nodes can intelligently alter transmission parameters in order to optimize use of the electromagnetic spectrum. Knowledge of control information provides valuable insight into the behaviors of existing networks, enabling cognitive radios to integrate themselves into communication channels without causing interference to other users. In this work we present a method of distinguishing between data and control packets in 802.11ac wireless networks. The data used to develop our classifier is composed of signals collected in a range of indoor and outdoor locations, allowing our algorithm to be trained on inputs that reflect the intricate interactions between electromagnetic waves and the environment. We employ a series of time-frequency and time-scale transformations to convert the collected spread-spectrum radio frequency signals into higher dimensional binary images which form the basis for our machine learning feature vectors. We extract features from fingerprint-like objects in the binary images using image processing techniques, after which we perform statistical tests to aid in downselection. This allows us to optimize the composition of the feature vectors and avoid the curse of dimensionality. We evaluate the effectiveness of our feature vectors by comparing results from several classification algorithms.

Margaret Rooney is a fourth-year Ph.D. candidate in the Applied Science Department at William & Mary. Her dissertation research focuses using machine learning to determine various features of wireless communication networks, and characterizing the interaction of high frequency radio waves with the environment. She holds a B.S. in Mathematics from St. John’s University and an M.S. in Applied Science from William & Mary.