As cybersecurity breaches become an increasing threat to many in today’s technology-driven world, School of Computing Assistant Professor Nirnimesh Ghose is exploring how to improve security through fingerprinting — of devices, not humans.
With a new grant from the National Science Foundation, Ghose will continue research in the expanding field of radio fingerprinting, a physical-layer authentication technique that plays a critical role in identifying individual technology devices.
Ghose will serve as principal investigator and collaborate with Boyang Wang of the University of Cincinnati on the project, which will be jointly supported by the Secure and Trustworthy Cyberspace program and the Established Program to Stimulate Competitive Research.
Radio fingerprinting can distinguish wireless devices using radio frequency signals, as unique device hardware imperfections are carried in such signals. Through their research, Ghose and his collaborators aim to develop new methods to promote the robustness, scalability and resilience of radio fingerprinting by synergizing deep learning and signal processing.
“In a wireless network connection, the verification server can verify a device’s radio fingerprint. This will prevent a malicious device from authenticating someone who has compromised the password,” Ghose said. “It can also be used to detect malicious devices, such as blacklisted UAVs that attempt to forge credentials to evade detection.”
Ghose, whose research focuses primarily on network security, said while radio fingerprinting is more than a decade old, using machine learning for radio fingerprinting is a new innovation in the field.
“This has improved the performance of fingerprinting, and we will be investigating how to make it difficult for an adversary to forge a fingerprint,” Ghose said.
Ghose and his team will approach the project by developing three new methods of radio fingerprinting: designing complex-valued triplet neural networks; building physical-layer-assisted generative adversarial networks; and building neural networks across the time domain, frequency domain and time-frequency domain. Each new method will improve accuracy, recognition and resilience by targeting and guarding against various types of attacks.
According to Ghose, advancements in radio fingerprinting techniques could greatly enhance protection of the most vulnerable aspects of technology.
“In general, endpoints are the weakest link in the cyberspace, so this is an effort to secure them,” Ghose said. “I’m excited to build new ways to make the cyberspace secure.”