Convolutional neural networks have been widely used in intrusion detection and proactive network defense strategies such as network security situation prediction (NSSP). The interaction between cross-channel features and the dependencies between elements in the input data are essential factors that affect the prediction model’s performance. However, existing works have ignored these, resulting in performance that needs to be improved. To this end, we propose a GResNeSt model that combines the advantages of the global context block and ResNeSt to improve the NSSP performance. The GResNeSt model strengthens traditional convolutional neural networks in two ways: it effectively captures cross-feature interactions and obtains long-range dependencies of the input data. This enhances its performance in capturing associations among different elements, making it more effective in extracting critical information from data to identify network attacks. We used the Salp swarm algorithm to select optimal hyperparameters for improving the model’s performance. Furthermore, based on the attack impact, we calculated network security situation values of two public network datasets. Finally, comprehensive experiments on the datasets verified our model design and demonstrated that our scheme is superior to other models in terms of NSSP ability.

In resource-constrained Internet of Things (IoT) environments, lightweight deep learning is crucial for radio frequency fingerprint identification (RFFI). However, existing lightweight designs primarily rely on group convolution with a “hard split” topology, which strictly isolates channels and blocks intergroup information sharing, impairing feature extraction. To address this, we propose overlap convolution, a novel operator employing a “soft coverage” mechanism to facilitate intergroup interaction. By enabling tunable channel overlapping, this method unifies and generalizes standard and group convolutions, restoring intergroup interaction without requiring additional mixing. We then analyze the approach by introducing structural constraint entropy as an interpretive framework to investigate the information flow capacity. Theoretical analysis demonstrates that overlap convolution offers expanded structural flexibility to achieve enhanced learning ability. Furthermore, we establish the OverlapConv framework for systematic integration of the operator into networks. Within this framework, we develop an automatic parameter acquisition strategy based on a differentiable-to-discrete transition mechanism to efficiently narrow down the search space for optimal settings. Extensive evaluations across diverse datasets (LoRa and UAV RFFI) and multiple backbones (MobileNet, ShuffleNet, and EfficientNet) confirm the method’s effectiveness as a universal plug-and-play module. Notably, our approach achieves an accuracy gain of 29.22% on the FireBlock, and improves the accuracy (averaging 2.1% and 2.47% on the two datasets) of FasterNet, EfficientNet, and LMSCNet with reduced FLOPs.

Physical layer identification is an emerging technique that exploits physical layer features to identify wireless devices. The identification accuracy and the device quantity that can be identified at most are significant for the identification scheme. Existing works primarily focus on the feature correlation analysis for multifeature selection without investigating the least upper bound (supremum) of the performance of a single feature. The supremum indicates the limit of the performance, which is another sight for evaluating the quality of features and improving the performance of the identification scheme. Therefore, this paper first investigates the supremum of the performance of the most commonly used physical layer feature, i.e., carrier frequency offset (CFO). Specifically, we offer a rigorous mathematical analysis and derive the closed‐form expression of the supremum of identification accuracy based on the max‐min distance analysis (MMDA) criterion. And then, the supremum of the number of distinguishable devices is also analyzed. Finally, we conducted a simulation study to verify the theoretical analysis result.