Position-Based Cross-Layer Authentication for Industrial Communications

We consider a robot (Alice) moving in an industrial environment while transmitting messages to nearby endpoints through fixed access-points (APs). An intruder robot (Trudy) aims at transmitting malicious messages to the endpoints, impersonating Alice. We aim at detecting Trudy's transmissions by comparing the expected position of the transmitter with two estimates of it obtained from a) the channel-state-information (CSI) estimated on the signals received by the APs, and b) the traffic information in the network. Such estimates are obtained with convolutional neural-network (CNN) and support vector regressor (SVR) models along with Kalman filters to exploit the trajectory evolution. Numerical results obtained using the DICHASSUS dataset confirm the effectiveness of our proposed solution.