Sustainable AI for Wireless Spectrum Sensing at the Edge

Department of Electrical Engineering and Computer Science

Overview

Rapid understanding and labeling of signals in the radio spectrum in an autonomous way is key for radio interference monitoring, detecting malfunctioning devices, and for numerous regulatory and defense applications. Existing approaches require expensive, high-maintenance expert systems that rely on carefully handcrafted features extracted for specific signal types and properties under simplified hardware, propagation, and radio environment models. In this project, we implement, and test state-of-the-art algorithms and optimization techniques collectively known as deep learning to classify wireless signals from datasets that include a wide range of simulated analog and digitally modulated signals and different propagation scenarios. To enable training and inference on low-power wireless devices, we test progressive weight pruning methods to generate compressed network representations that fit on dedicated FPGA hardware and do not lose their accuracy.

Community Benefit: With billions of pervasively deployed and connected wireless devices, the oncoming IoT revolution will create a new paradigm for sensing and actuation at the network edge. While certain types of environmental, industrial and human-activity centric sensing require massive amounts of data to be transferred from the sensors to the remote cloud, there are many scenarios that can benefit from the deployment of sustainable AI solutions at the wireless edge. For example, rapid recognition of the modulation type of a signal can help a radio communication system rapidly adapt to radio interference. In military settings, friendly signals should be securely received, while hostile signals need to be efficiently recognized typically without prior information. If these steps can be accomplished at the edge, then we may save energy and communication overheads for power-constrained IoT devices.

Team Members

• Ethan Fisher
• Dariush Hassan
• Nicole Appleton Guerrero
  
• Raphael Valente
• Efrem Yohannes-Mason

Sponsor

FAU Center for Connected Autonomy and AI