Weiguo Shen

Interference recognition serves as the preprocessing technology for interference suppression. Researchers typically employ the method based on the time-frequency graph - deep convolutional neural network (CNN) in pursuit of recognition accuracy. Nevertheless, certain time-frequency analysis processing approaches or network models with excessive parameters make it challenging to apply this technology to devices with limited resources. In this paper, a weight pruning and parameter quantization assisted multi-Scale lightweight network named WPMNet is proposed, which can directly take the raw time series data as input. The network model comprises three multi-scale convolutional modules, an attention mechanism, and fully connected layers. The multi-scale convolutional modules are composed of convolutional kernels of different sizes in parallel, aiming to extract deep-level features and features of various granularities simultaneously. To compress the model, we introduce an adaptive search algorithm for determining the optimal pruning threshold for weight pruning, ensuring that the accuracy decline is within an acceptable range. And the model is further compressed through parameter quantization. The experimental outcomes demonstrate that prior to the lightweighting of the model, the average recognition accuracy was 97.6%. After the lightweight processing, the compression ratio of the network attained 84.5%, while the average recognition accuracy merely declined slightly to 95.08%. This substantiates that the proposed approach is applicable to scenarios that simultaneously demand both high precision and model deployability, such as embedded systems or mobile devices.

In this work, we investigate the age of information (AoI) in aerial intelligent reflecting surfaces (IRS) assisted covert data collection in Internet of Things (IoT) networks. Operating with the unmanned aerial vehicle (UAV) and IRS can improve the data transmission covertness, as well as the data freshness by reconstructing the wireless propagation environment. Specifically, we consider a scenario in which ground IoT sensors transmit confidential information to the legitimate receiver (Bob), through IRS assisted UAV relay, at the same time, an eavesdropper passively listens to and intercepts the confidential information sent by sensors. To minimize the average AoI of the system, a joint UAV trajectory and IRS phase shift optimization problem is formulated under the constraints of covertness requirement. The minimum error detection probability and optimal detection threshold are first derived at Willie, which represents the worst-case situation for the legitimate transmission. The constructed problem is a mixed integer programming NP-Hard problem, which is more complex using traditional convex optimization methods. Therefore, the online learning, i.e.,deep reinforcement learning is leveraged to obtain the near-optimal solution. Details, the Deep Q Network (DQN) and Deep Deterministic Policy Gradient (DDPG) methods are utilized. Numerical results show that deep reinforcement learning can improve the information freshness of the system by reasonably designing UAV trajectory and IRS phase shift under a given covertness constraint.

Internet of Things (IoT) play a paramount role in every aspect of our daily lives. Due to more diverse human needs, the future IoT networks are expected to be highly dynamic and heterogeneous with assistance of mobile nodes, such as Unmanned Aerial Vehicles (UAVs). However, the broadcast and openness nature of wireless communication and high-mobile characteristic of UAV can cause security threats to UAV-assisted IoT systems. For this sake, we consider exploiting covert communication to provide such a system with a higher level communication security, which can prevent the legitimate transmission being detected by the adversary monitor. Specifically, this paper jointly optimizes the transmit power and hovering location of the UAV to guarantee communication security in the IoT system. We maximize the signal-to-noise ratio (SNR) of the legitimate receiver in presence of a malicious warden, with constraints of communication covertness, the UAV’s spatial location and maximum transmit power. Particularly, the UAV’s location is represented in terms of angles, rather than the commonly used distance, in most of the literature. The optimal location is determined in two steps. The simulation results show that the proposed optimization schemes can effectively find the optimal hovering location and transmit power of the UAV to maximize the receiver’s SNR under the covertness constraint. The optimal hovering location is directly above the line connecting the legitimate receiver and the warden, and in close proximity within the small region directly above the receiver, which has implications for other analogous research scenarios or practical applications of UAV.

A wireless communications system usually consists of a transmitter which transmits the information and a receiver which recovers the original information from the received distorted signal. Deep learning (DL) has been used to improve the performance of the receiver in complicated channel environments and state-of-the-art (SOTA) performance has been achieved. However, its robustness has not been investigated. In order to evaluate the robustness of DL-based information recovery models under adversarial circumstances, we investigate adversarial attacks on the SOTA DL-based information recovery model, i.e., DeepReceiver. We formulate the problem as an optimization problem with power and peak-to-average power ratio (PAPR) constraints. We design different adversarial attack methods according to the adversary’s knowledge of DeepReceiver’s model and/or testing samples. Extensive experiments show that the DeepReceiver is vulnerable to the designed attack methods in all of the considered scenarios. Even in the scenario of both model and test sample restricted, the adversary can attack the DeepReceiver and increase its bit error rate (BER) above 10%. It can also be found that the DeepReceiver is vulnerable to adversarial perturbations even with very low power and limited PAPR. These results suggest that defense measures should be taken to enhance the robustness of DeepReceiver.

Automatic modulation recognition (AMR) of radio signal is an important research topic in the area of non-cooperative communication and cognitive radio. Recently deep learning (DL) techniques enable significant progress in AMR. However, the techniques of adversarial machine learning cause the threats of adversarial attacks in DL-based AMR. In this paper, we aim to make AMR model robust, accurate and lightweight, thus propose a multi-distillation mechanism for robust training of DL-based AMR models, namely Adversarial Multi-Distillation (AMD). In the framework of AMD, by knowledge distillation, two powerful teacher models transfer the learned classification knowledge and defense knowledge, respectively, to the student model to form robust training. Our experiments with public dataset RML2016.10a show that the proposed method can significantly improve the defense of AMR models to against adversarial perturbations and keep relatively high classification accuracy, which enables robust decision making with lightweight models under adversarial attacks.

Direction-of-arrival (DOA) estimation is a vital research topic in array signal processing, with extensive applications in many fields. In recent years, deep learning has been applied to DOA estimation to improve the performance. However, most existing deep learning-based DOA estimation methods extract DOA information from the covariance matrix (CM) input. In this paper, we introduce a novel deep learning-based DOA estimation scheme that utilizes the raw in-phase (I) and quadrature (Q) components of the signal as the input. We formulate the problem as single-label classification and multi-label classification based on the number of signal sources. We design a convolutional neural network to solve the problems and to adapt to different number of snapshots. We also propose a deep learning regression-based method to overcome the limitations of classify-based methods in dealing with off-grid angles. We conduct extensive experiments with simulations and over-the-air collected signals to analyze the performance of the proposed method in various scenarios including different SNRs, additive generalized Gaussian noise (AGGN) and extreme multi-source DOA estimation. Results demonstrate that our proposed method outperforms the existing deep learning-based DOA estimation methods.

Recently, backdoor attacks posed a new security threat against radio signals modulation models. The attacked model performs well on benign samples, whereas performs abnormally on backdoor samples. Current backdoor attacks usually inject static trigger into benign samples, which can be easily detected with static trigger patterns and large perturbation. In this paper, inspired by the Generative Adversarial Network (GAN), we proposed a backdoor attack against radio signals modulation based on adaptive trigger (AT-GAN) to learn the patterns of adaptive trigger, which can generate different trigger for specific radio signal. AT-GAN is composed of Adaptive Backdoor Generate Network (A-BaN), discriminator and radio signals modulation model. A-BaN is trained to generate adaptive triggers, discriminator is used to limit the perturbation of triggers and modulation model is poisoned based on generated poisoned samples. By a three-player game, A-BaN can generate adaptive triggers with small perturbation. The experimental results on two benchmark datasets demonstrate that our method can achieve better attack performance and reduce the detection rate of backdoor attacks compared to baseline backdoor attacks.