Analysis of the Impact of Varying Statistical RF Fingerprint Features on IoT Device Classification

ABSTRACT

Rapid advancements in technology are having a vast impact on today's society. The number of technological devices available to consumers is continuously growing and is projected to reach 55 billion by 2025. The interconnection of these devices and systems has become more prevalent and is commonly known as the Internet of Things (IoT). The increase in these devices has led to the exploitation of vulnerabilities in these systems, thus creating a need for security mechanisms suitable for the IoT domain. RF Fingerprinting is a promising security technique that uses the physical characteristics of a wireless signal to identify a device. These characteristics, known as features, are combined to create an RF fingerprint and are then run through a machine learning classification algorithm for device identification. This work performs an analysis on the impact that varying features for each device have on classification using only statistical RF Fingerprint features. Experiments are performed on IoT home automation devices with variance, skewness, and kurtosis features extracted from the wireless signals and tested with common machine learning classifier models. Results show improvements in device identification when using features specific to a device for each individual fingerprint.

BIO

Asia Mason received her Bachelor's and Master's degrees in Electrical Engineering from Morgan State University and is currently pursuing her Doctor of Engineering in Electrical Engineering. Ms. Mason is a member of the Advanced RF Microwave Measurement and Electronic Design Laboratory (ARMMED) and the Center for Reverse Engineering and Assured Microelectronics (CREAM). Her areas of focus are wireless communication and physical layer security for IoT devices.