Air transportation communication jamming recognition model based on deep learning (DL) can quickly and accurately identify and classify communication jamming, to improve the safety and reliability of air traffic. However, due to the vulnerability of deep learning, the jamming recognition model can be easily attacked by the attacker’s carefully designed adversarial examples. Although some defense methods have been proposed, they have strong pertinence to attacks. Thus, new attack methods are needed to improve the defense performance of the model. In this work, we improve the existing attack methods and propose a double level attack method. By constructing the dynamic iterative step size and analyzing the class characteristics of the signals, this method can use the adversarial losses of feature layer and decision layer to generate adversarial examples with stronger attack performance. In order to improve the robustness of the recognition model, we use adversarial examples to train the model, and transfer the knowledge learned from the model to the jamming recognition models in other wireless communication environments by transfer learning. Simulation results show that the proposed attack and defense methods have good performance.
22 December 2023 in IEEE Internet of Things Journal
Modulation recognition using deep learning (DL) can efficiently recognize modulated signals in cognitive radio-enabled Internet of Things (IoT). However, it is vulnerable to the attack of adversarial examples designed by attackers, leading to a decrease in its accuracy. Different adversarial techniques can be used for attacks, but these attacks have limited efficiency. This article proposes a double loop iterative method. Different from the traditional attack methods, the new method designs an additional external loop iteration for high efficiency. When generating adversarial examples, the initial conditions of each iteration can be updated as the number of iterations changes, so that the adversarial examples can cross the decision boundary of the model as much as possible. In addition, this article uses knowledge distillation to improve the traditional adversarial training defense, which improves the robustness of the model. Simulation results show that the proposed attack and defense methods have better performance than traditional methods.
31 December 2024 in IEEE Transactions on Cognitive Communications and Networking
Deep neural network (DNN) greatly improves the efficiency of modulation recognition in wireless communication, but it also suffers from attacks. Generative artificial intelligence (GAI) possesses powerful data generation capabilities, which can be used to defend against attacks in modulation recognition. In practical scenarios, black box attack can be implemented without information on the model. This is a great security threat. The existing defense methods are difficult to improve the robustness of the model while ensuring the recognition accuracy of the original signals. Therefore, this paper uses GAI to propose an adversarial decoupled defense method to protect modulation recognition. Firstly, for weak adversarial perturbations, the empirical mode decomposition (EMD) is used to highlight the high-frequency features in the signal, and the adversary detector is designed to detect the suspiciousness. Then, the signal is regenerated based on the generative adversarial network (GAN) to weaken the antagonism in the example. Further, the traditional adversarial training is decoupled into an original branch and an adversarial branch, and the outputs of the two branches are fused according to the suspiciousness. Simulation results show that the proposed defense method has high recognition accuracy for both original examples and adversarial examples even under attacks, and can effectively improve the robustness of the intelligent recognition model.
The problem of single-channel reception of global positioning system (GPS) communication waveforms makes passive sensing of aerial target difficult because of forward scatter. This paper proposes a novel aerial target passive sensing method based on linear canonical transformation (LCT) using the forward scattered satellite communication waveforms. The proposed method firstly preprocesses the received signal based on the characteristics of the traditional satellite tracking loop and the forward scattered satellite communication waveforms to effectively suppress the interference of the direct wave through DC removal. Then, the Gaussian noise and multipath interference in the channel are suppressed by applying a rectangular window to its linear canonical domain. Finally, aerial target sensing is performed based on the peak value of signals in the linear canonical transform domain. The characteristic signal is constructed by analyzing the satellite communication waveforms. Combining the linear canonical transform with the matched filter (MF) to estimate the target parameter. Simulation results show that the proposed method can effectively perform the aerial target sensing by using satellite communication waveforms in the forward scatter scenario.
16 October 2025 in IEEE Transactions on Cognitive Communications and Networking
Deep learning (DL) can enhance automatic modulation recognition in complex wireless communication environments by greatly improving the efficiency and accuracy of modulation recognition. However, the DL-based intelligent recognition model also has great vulnerability to the adversarial attack. To avoid this, a subtle adversarial perturbation can be added to the signal to disrupt the eavesdropper’s modulation recognition model while ensuring the normal demodulation at the receiving end. However, the transferability of existing attacks is weak, and it is difficult to disrupt the target model while ensuring concealment. This paper proposes a meta adversarial transfer attack method by using meta-training and meta-testing to enhance the transferability of adversarial attacks. In addition, for the transfer attack, this paper proposes an adversarial domain adaptation defense method, which uses the multi-class domain-invariant features of the examples in the source domain and the adversarial domain to improve the robustness of the model against attacks. Simulation results show that, compared with the traditional methods, the proposed attack method has better transferability for the eavesdropper’s model, and that the proposed defense method has better defense against the transfer attack.
24 July 2025 in IEEE Transactions on Information Forensics and Security
In wireless communications, the communication channel between the transmitter and receiver can be monitored by an eavesdropper. The eavesdropper uses deep learning (DL) to quickly identify the modulation parameters of signals and further disrupt legitimate communications. Since DL has been proven to be vulnerable to adversarial attacks, this paper proposes to attack the eavesdropper’s model by designing adversarial waveforms, preventing the eavesdropper from correctly identifying the modulation schemes used by legitimate users, and thereby preventing the eavesdropper from interfering with normal communications. This paper proposes an attention-based black-box attack method, which uses the prediction of different networks in the ensemble model to assign adversarial attention factors to each network. This greatly improves the transmission attack performance of the designed adversarial examples. In addition, by analysing the influence of the channel on the adversarial waveform, we further design the adversarial waveform that can be transmitted in the channel to improve the practicability of the attack algorithm. Finally, we theoretically derive the bounds of the adversarial risk increase that the attack brings to the target model. Simulation results show that the proposed method can improve the success rate of the attack on the eavesdropper’s modulation detection model, cause the model to misidentify the signal modulation type, and improve the security and reliability of legitimate transceivers in wireless communication systems.