Anomaly Attack Detection in Wireless Networks Using DCNN

• The use of wireless devices in industrial sectors has increased due to its various advantages related to cost and flexibility. However, legitimate wireless communication systems are vulnerable to cybersecurity attacks, due to its inherent open nature. Detection of rogue devices therefore plays a crucial role in critical wireless applications. In this paper we design a deep convolutional neural network (DCNN) to classify legitimate and rogue devices using raw IQ samples as input data. An algorithm is presented to find the optimal number of convolutional layers and number of filters for each layer under an accuracy constraint, in order to enable fast prediction time. Furthermore, we investigate how wireless channel models affect the accuracy and prediction time of the designed DCNN model. Our obtained results are benchmarked against previous DCNN models. Moreover, we discuss how the systems should react to a detected rogue device, considering the IEC 62443 standard.

Subject headings

NATURVETENSKAP  -- Data- och informationsvetenskap -- Datavetenskap (hsv//swe)

NATURAL SCIENCES  -- Computer and Information Sciences -- Computer Sciences (hsv//eng)

Keyword

deep learning

finger-printing

IEC 62443

rogue device detection

Anomaly detection

Convolution

Cybersecurity

Deep neural networks

Neural network models

Wireless networks

Attack detection

Convolutional neural network

Finger printing

Industrial sector

Neural network model

Prediction time

Wireless devices

Convolutional neural networks

Publication and Content Type

ref (subject category)

kon (subject category)