Classifying Wi-Fi from Raw Power Measurements using a Neural Network Adapted from WaveNet
Robert J. Estes, Derrick Willis, Hazem H. Refai · 2021
Research into identifying coexisting wireless devices and technologies has been growing in importance as the number of wireless devices increase every year. While many various methods have been proposed over the years, most of them usually rely on using features from the frequency domain. These can usually be distinguishable among different wireless standards. The feature space in the frequency domain is usually limited but less complex than those in the time domain. However, when trying to identify wireless devices that use different specifications of the same Wi-Fi standard, utilizing the frequency domain feature is difficult as the features are too similar between specifications.In this paper, we investigate utilizing a neural network to identify the Wi-Fi standard using raw power measurements. More specifically, we adapt the WaveNet model to take advantage of its capability to handle time series data with thousands of timesteps, an advantage that both recurrent neural networks (RNN) and long short-term memory (LSTM) networks do not share. Wi-Fi signals are collected across versions 802.11n, 802.11ac, and 802.11ax both individually and with multiple standards coexisting across a range of different throughputs. The data is then pre-processed and used to train a neural network adapted from the WaveNet model. Results indicate that high accuracy detection can be achieved by utilizing this method.