2023_programme: On the Quickest Detection Problem for Authentication in Underwater Acoustic Channels

Back


  • Session: 19. Underwater Communications and Networking
    Organiser(s): Charalampos Tsimenidis, Paul Mitchell and Konstantinos Pelekanakis
  • Lecture: On the Quickest Detection Problem for Authentication in Underwater Acoustic Channels [invited]
    Paper ID: 2087
    Author(s): Cardillo Laura, Ardizzon Francesco, Tomasin Stefano
    Presenter: Ardizzon Francesco
    Abstract: Underwater acoustic communications are becoming increasingly popular, due to their\nuse in both military and civil applications. Indeed, many of these services require the communications\nto be authenticated, i.e., the receiver must detect if the received message was transmitted\nby an impersonating attacker. Several security solutions have been proposed to provide authentication\nfor underwater acoustic channels (UWACs). These typically test a security metric and\nare based on either cryptography or physical-layer security (PLS), i.e., evaluating the statistical\nproperties of the channel itself.\nIn this paper, we instigate the problem of quickest detection for physical-layer authentication\n(PLA) in UWACs, assuming that from a time instant all the signals are sent by the attacker and\nthe defender aims at detecting when an attack starts with the minimum possible delay, by examining\nthe characteristics of the UWAC over which the signal is received. We propose a solution\nbased on recurrent neural networks (RNNs). These neural networks (NNs) process the input\ntaking into account also its past realizations, and thus are suitable to detect anomalous changes\nthat can be tied to an impersonation attack. The proposed approach quickly detects when the\nattacker is starting its attack activity. Performance are evaluated using an authentication test on\nan experimental dataset, comparing the performance to the optimal cumulative sum (CUSUM)\ntest which however requires the prior knowledge of the UWAC statistics.
      Download the full paper
  • Corresponding author: Dr Francesco Ardizzon
    Affiliation: Department of Information Engineering, University of Padua
    Country: Italy
    e-mail:

Back