Threat Detection Model for WLAN of Simulated Data Using Deep Convolutional Neural Network

Abstract

Security identification solutions against WLAN network attacks according to straightforward digital detectors, such as SSID, IP addresses, and MAC addresses, are not efficient in identifying such hacking or router impersonation. These detectors can be simply mocked. Therefore, a further protected key uses new information by combining these simple digital identifiers with an RF signature of the radio link. In this work, a design of a convolutional neural network (CNN) based on fingerprinting radio frequency (RF) is developed with computer-generated data. The developed CNN is trained with beacon frames of a wireless local area network (WLAN) that is simulated as a result of identified and unidentified router nodes of fingerprinting RF. The proposed CNN is able to detect router impersonators by comparing the data pair of the MAC address and RF signature of the received signal from the known and unknown routers. ADAM optimizer, which is the extended version of stochastic gradient descent, has been used with a developed deep learning convolutional neural network containing three fully connected and two convolutional layers. According to the training progress graphic, the network converges to around 100% accuracy within the first epoch, which indicates that the developed architecture was efficient in detecting router impersonations.