Ghaderi, MajidMohammadi, Ali Teshnizi2023-09-152023-09-152023-09-12Mohammadi, A. T. (2023). Covert communication in autoencoder wireless systems (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.https://hdl.handle.net/1880/11702910.11575/PRISM/41872The broadcast nature of wireless communications presents security and privacy challenges. Covert communication is a wireless security practice that focuses on intentionally hiding transmitted information. Recently, wireless systems have experienced significant growth, including the emergence of autoencoder-based models. These models, like other DNN architectures, are vulnerable to adversarial attacks, highlighting the need to study their susceptibility to covert communication. While there is ample research on covert communication in traditional wireless systems, the investigation of autoencoder wireless systems remains scarce. Furthermore, many existing covert methods are either detectable analytically or difficult to adapt to diverse wireless systems. The first part of this thesis provides a comprehensive examination of autoencoder-based communication systems in various scenarios and channel conditions. It begins with an introduction to autoencoder communication systems, followed by a detailed discussion of our own implementation and evaluation results. This serves as a solid foundation for the subsequent part of the thesis, where we propose a GAN-based covert communication model. By treating the covert sender, covert receiver, and observer as generator, decoder, and discriminator neural networks, respectively, we conduct joint training in an adversarial setting to develop a covert communication scheme that can be integrated into any normal autoencoder. Our proposal minimizes the impact on ongoing normal communication, addressing previous works shortcomings. We also introduce a training algorithm that allows for the desired tradeoff between covertness and reliability. Numerical results demonstrate the establishment of a reliable and undetectable channel between covert users, regardless of the cover signal or channel condition, with minimal disruption to the normal system operation.enUniversity of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.Wireless securityCovert autoencoder systemsCommunicationWireless autoencoder systemsComputer ScienceArtificial Intelligencemaster thesis