Identity Theft Detection Based on Neural Network Non-Linearity Identification in OFDM System

Confidentiality of the communication is extremely important in wireless networks due to the broadcast nature of the radio channel. The secrecy of the communication can be achieved via secure PHY layers or by using crypto techniques which need a pre-shared secret between the sender and the legal receiver. A technique has been proposed in literature to avoid preshared secrets based on the noise-pattern of the radio channel. However, when the channel frequency selectivity is low or the eavesdropper is located near to the victim, the key exchange may fail to achieve security. The eavesdropper can gain access to the secure communication and steal the identity of one of the two communicating nodes. In this paper, a technique for identity theft detection is proposed: the receiving node can detect unintended change of sender´s identity. A Feed-Foward Neural Network is used to identify the non-linearities of the OFDM radio transceiver which are unique and characteristic of the transmitting and receiving devices. The simulation results show that the NN correctly detects the non linearities coefficients for pedestrian and high mobility terminals in the presence of severe fading and noise, providing a strong mean for radio signal identification to detect broken security.