With the increasingly complex electromagnetic environment and the intelligent development of radar, the jammer, as opposed to radar, urgently needs to improve its ability to recognize threat targets and make jamming decisions. In this paper, we first introduce the concepts and systems of cognitive electronic warfare (CEW) and summarize its research status. Through analysis of the existing CEW systems, we propose a CEW model suitable for the cluster confrontation scenarios. Then, for the radar jamming decision-making (RJDM) namely a crucial part of CEW, we discuss the advantages, disadvantages, and applications of the traditional methods and analyze the machine-learning based methods including Markov decision processing, the newest Q-Learning, Deep Q-Learning (DQN), Double Deep Q-Learning (DDQN), A3C algorithms, and their improved algorithms etc. We build radar adversarial models and verify the effectiveness of reinforcement learning (RL) algorithm and the superiority of deep RL by simulating both the underlying Q-Learning and DQN algorithms. Finally, the research trends of CEW are discussed.

In order to improve the efficiency and adaptability of cognitive radar jamming decision-making, a fusion algorithm (Ant-QL) based on ant colony and Q-Learning is proposed in this paper. The algorithm does not rely on a priori information and enhances adaptability through real-time interactions between the jammer and the target radar. At the same time, it can be applied to single jammer and multiple jammer countermeasure scenarios with high jamming effects. First, traditional Q-Learning and DQN algorithms are discussed, and a radar jamming decision-making model is built for the simulation verification of each algorithm. Then, an improved Q-Learning algorithm is proposed to address the shortcomings of both algorithms. By introducing the pheromone mechanism of ant colony algorithms in Q-Learning and using the $\epsilon$-greedy algorithm to balance the contradictory relationship between exploration and exploitation, the algorithm greatly avoids falling into a local optimum, thus accelerating the convergence speed of the algorithm with good stability and robustness in the convergence process. In order to better adapt to the cluster countermeasure environment in future battlefields, the algorithm and model are extended to cluster cooperative jamming decision-making. We map each jammer in the cluster to an intelligent ant searching for the optimal path, and multiple jammers interact with each other to obtain information. During the process of confrontation, the method greatly improves the convergence speed and stability and reduces the need for hardware and power resources of the jammer. Assuming that the number of jammers is three, the experimental simulation results of the convergence speed of the Ant-QL algorithm improve by 85.4%, 80.56% and 72% compared with the Q-Learning, DQN and improved Q-Learning algorithms, respectively. During the convergence process, the Ant-QL algorithm is very stable and efficient, and the algorithm complexity is low. After the algorithms converge, the average response times of the four algorithms are 6.99 × 10⁻⁴ s, 2.234 × 10⁻³ s, 2.21 × 10⁻⁴ s and 1.7 × 10⁻⁴ s, respectively. The results show that the improved Q-Learning algorithm and Ant-QL algorithm also have more advantages in terms of average response time after convergence.

Spectrum prediction is a promising technique of improving the spectrum exploitation in cognitive radio networks (CRNs). Accurate spectrum prediction can assist in reducing the energy consumption of spectrum sensing and improving the network throughput, etc. Recently, a significant amount of research efforts has been devoted to this area, especially deep learning (DL) methods, and greatly advanced spectrum prediction abilities. This paper reviews the state-of-the-art developments in DL for spectrum prediction. Specifically, we first summarize the existing spectrum prediction methods and give a taxonomy from different aspects. Second, we formulate long-term, multi-dimensional, and non-ideal spectrum prediction problems to provide theoretical support for spectrum prediction models. Third, we divide the DL models into four categories: basic models, models for the long-term domain, models for the multi-dimensional domain, and models for the non-ideal domain base on application scenarios and offer an in-depth tutorial on these models. Fourth, we comprehensively collect and organize widely used common spectrum data sources in the existing literature and also overview the spectrum data simulation methods to facilitate other researchers. Furthermore, we summarize and analyze the selection of evaluation metrics for spectrum prediction models based on DL. Finally, we summarize the research trends and highlight the critical research challenges.

Spectrum prediction is a promising technique to release spectrum resources and plays an essential role in cognitive radio networks and spectrum situation generating. Traditional algorithms normally focus on one-dimensional or predict spectrum values in a slot-by-slot manner and thus cannot fully perceive the spectrum states in complex environments and lack timeliness. In this paper, a deep learning-based prediction method with a simple structure is developed for temporal–spectral and multi-slot spectrum prediction simultaneously. Specifically, we first analyze and construct spectrum data suitable for the model to simultaneously achieve long-term and multi-dimensional spectrum prediction. Then, a hierarchical spectrum prediction system is developed that takes advantage of the advanced Bi-ConvLSTM and the seq2seq framework. The Bi-ConvLSTM captures time–frequency characteristics of spectrum data, and the seq2seq framework is used for long-term spectrum prediction. Furthermore, the attention mechanism is used to address the limitations of the seq2seq framework that compresses all inputs into fixed-length vectors, resulting in information loss. Finally, the experimental results have shown that the proposed model has a significant advantage over the benchmark schemes. Especially, the proposed spectrum prediction model achieves 6.15%, 0.7749, 1.0978, and 0.9628 in MAPE, MAE, RMSE, and $R^2$, respectively, which is better than all the baseline deep learning models.

Radio frequency (RF) fingerprint (RFF) refers to a unique hardware-related feature of RF devices, characterized by non-reproducible physical characteristics. This feature makes RFF suitable for equipment identity authentication in wireless communication, particularly in the Internet of Things (IoT). Therefore, we propose a more generalized and intuitive definition of the RFF, namely the differences between the received and ideal RF signals. The ideal RF signal is defined as having the same input information source, encoding, and modulation schemes as the received signal, but without any influences from hardware devices or wireless transmission channels. Our approach begins by analyzing the effects of transmission channels, such as attenuation, dispersion, multi-path propagation, and noise, treating them as a series of processes. We represent both the received and ideal signals in a designed multi-domain space.We then develop a Siamese-based neural network to extract features from the multi-domain representations of paired signals simultaneously. A difference block is used to eliminate common features and retain the expected generalized RFF (GRFF). Identification is achieved by treating each device as a specific class, utilizing a softmax classifier. Besides, we introduce a two-stage strategy involving nearest neighbor and particle swarm optimization to discriminate the out-of-distributionsignals. Experiments are carried out using a public ADS-B signal dataset, and the trained network achieves a classification accuracy of 99.13% on a test dataset with 7442 samples from 58 classes in total and out-of-distributiondetection accuracy of 95.45% under a class confidence 0.96 on 5560 samples from additional 206 classes.

With the rapid development of the internet of things (IoT), hundreds of millions of IoT devices, such as smart home appliances, intelligent-connected vehicles, and wearable devices, have been connected to the network. The open nature of IoT makes it vulnerable to cybersecurity threats. Traditional cryptography-based encryption methods are not suitable for IoT due to their complexity and high communication overhead requirements. By contrast, RF-fingerprint-based recognition is promising because it is rooted in the inherent non-reproducible hardware defects of the transmitter. However, it still faces the challenges of low inter-class variation and large intra-class variation among RF fingerprints. Inspired by fine-grained recognition in computer vision, we propose a fine-grained RF fingerprint recognition network (FGRFNet) in this article. The network consists of a top-down feature pathway hierarchy to generate pyramidal features, attention modules to locate discriminative regions, and a fusion module to adaptively integrate features from different scales. Experiments demonstrate that the proposed FGRFNet achieves recognition accuracies of 89.8% on 100 ADS-B devices, 99.5% on 54 Zigbee devices, and 83.0% on 25 LoRa devices.