ZZZiwei Zhang

Ziwei Zhang

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Ziwei Zhang has a total of 22 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Adaptive Noise-Resilient Test-Time Adaptation for RF Signal Recognition

    by Haoran Zha, Hanhong Wang, Ziwei Zhang, Hongtao Zhan, Guan Gui, Yun Lin

    The rapid growth of wireless communication technologies has led to congested Radio Frequency (RF) spectrum and increased signal interference. RF signal recognition plays a vital role in intelligent spectrum management, with two representative tasks being Automatic Modulation Classification (AMC) and Specific Emitter Identification (SEI). However, deep learning models often suffer performance degradation under dynamic noise conditions due to distribution shifts. To address this, we propose ANR (Adaptive Noise-Resilient), a source-free Test-Time Adaptation (TTA) framework that plugs into existing networks. ANR quickly adapts during inference by updating batch normalization parameters in an unsupervised manner. This adaptation is driven by a combination of signal aware augmentation, pseudo label loss, and consistency losses. This allows efficient and source-free adaptation to changing noise conditions. Experiments on a benchmark dataset with multiple noise models and varying signal-to-noise ratios show that ANR significantly enhances robustness and generalization. It achieves accuracy gains of 19.75% for AMC and 51.58% for SEI under challenging noise scenarios. Ablation studies further analyze key hyperparameters, providing guidance for balancing computational efficiency and recognition accuracy in practical applications.


  • Self-Supervised Aligned Data Augmentation Network for Imbalanced Modulation Classification

    by Ziwei Zhang, Yunjie Li, Mengtao Zhu, Shafei Wang

    Automatic modulation classification (AMC) plays a pivotal role in radar and communication systems. Traditional AMC methods assume an equal number of samples for each modulation during training. However, in real-world scenarios, the number of samples collected for different modulations can vary significantly. This disparity leads the model to overfit to majority classes while underrepresenting minority classes in the optimization process, ultimately leading to degraded classification performance. To tackle this issue, this article presents a self-supervised aligned data augmentation network (SADA-Net) for imbalanced AMC. We leverage data augmentation (DA) not only to balance data distribution but also increase data diversity, boosting the model’s robustness in handling minority classes. To amplify the effectiveness of DA, a self-supervised aligned module is incorporated to maintain semantic consistency, preventing the model from focusing on irrelevant variations. Furthermore, an adaptive fusion learning strategy is proposed to dynamically adjust the focus between majority classes and minority classes during training process. This progressive training strategy avoids damaging the learned universal features from majority classes when emphasizing the minority data, ensuring a balanced feature learning process. Comprehensive experiments on simulated, publicly available and real-world datasets demonstrate the effectiveness and generalization of SADA-Net under various class-imbalanced conditions.


  • Cross-Domain Generalization for Specific Emitter Identification With Unseen Signals via Fourier Phase and MMD Features

    by Haoran Zha, Xiulin Shu, Chang Liu, Ziwei Zhang, Jun Chen, Yun Lin

    Specific emitter identification (SEI) in wireless communications enhances security by distinguishing devices through their unique RF signal impairments. Deep learning (DL) techniques have become instrumental and have attracted considerable focus in this domain. Nevertheless, significant challenges arise from pronounced domain distribution disparities and the lack of labeled signal for target domain emitters, complicating cross-domain individual emitter identification. This article proposes an innovative domain generalization framework for SEI, which exploits both intradomain and interdomain invariant features to enhance cross-domain robustness. These features significantly enhance the model’s generalization capabilities: intradomain invariants are captured through the use of Fourier phase information and knowledge distillation, while interdomain invariants are derived via maximum mean discrepancy (MMD) feature alignment. The proposed methodology exhibits strong performance across three domain generalization scenarios, utilizing datasets SUI-1, SUI-3, and SUI-5. Each dataset comprises signals from seven distinct transmitters, recorded over varying fading channels. This approach achieves an approximate 5% improvement over state-of-the-art SEI domain adaptation methods, highlighting its superior generalization capability. The code and datasets for this article can be found at: https://github.com/ZHR-HEU/Cross-Domain-Generalization-for-SEI.


  • APL: Integrated Discriminative Features and Robust Boundary for Modulation Open-Set Recognition

    by Ziwei Zhang, Mengtao Zhu, Yunjie Li, Shafei Wang

    As the electromagnetic environment becomes increasingly complex, traditional Automatic Modulation Recognition (AMR) methods cannot handle unknown modulation types that may arise under real-world conditions. Therefore, Automatic Modulation Open-Set Recognition (AMOSR) has gained significant attention as a technique that could identify unknown classes, playing a crucial role in enhancing the reliability of cognitive radio systems. Existing AMOSR approaches have primarily concentrated on either extracting discriminative features or establishing robust decision boundaries to enhance the AMOSR performance. To overcome existing limitation, we propose an Adversarial Prototype Learning (APL) algorithm to jointly optimize the features and boundaries through iterative refinement of prototype learning and adversarial learning. The prototype learning exploits the semantic similarity with the designed distribution distance-based cross entropy loss, aiming to obtain compact feature distributions while enhancing inter-class separability. The adversarial learning fully leverages the generated counterfactual images to constrain the unknown feature space, thus conducive to robust decision boundaries between known and unknown classes. The joint optimization yields mutual enhancements to boost AMOSR performance, as discriminative features facilitate boundary formulation, while well-defined boundaries further consolidate feature concentration. Comprehensive experiments on simulated and real-world signals demonstrate the effectiveness and robustness of APL compared to state-of-the-art AMOSR methods across various signal conditions.


  • Open Access

    Disentangled Representation Learning for Robust Radar Inter-Pulse Modulation Feature Extraction and Recognition

    by Luyao Zhang, Mengtao Zhu, Ziwei Zhang, Yunjie Li

    Modern Multi-Function Radars (MFRs) are sophisticated sensors that are capable of flexibly adapting their control parameters in transmitted pulse sequences. In complex electromagnetic environments, efficiently and accurately recognizing the inter-pulse modulations of non-cooperative radar pulse sequences is a key step for modern Electronic Support (ES) systems. Existing recognition methods focus more on algorithmic designs, such as neural network structure designs, to improve recognition performance. However, in open electromagnetic environments with increased flexibility in radar transmission, these methods would suffer performance degradation due to domain shifts between training and testing datasets. To address this issue, this study proposes a robust radar inter-pulse modulation feature extraction and recognition method based on disentangled representation learning. At first, inspired by the Representation Learning Theory (RLT), the received radar pulse sequences can be disentangled into three explanatory factors related to (i) modulation types, (ii) modulation parameters, and (iii) measurement characteristics, such as measurement noise. Then, an explainable radar pulse sequence disentanglement network is proposed based on auto-encoding variational Bayes. The features extracted through the proposed method can effectively represent the key latent factors related to recognition tasks and maintain performance under domain shift conditions. Experiments on both ideal and non-ideal situations demonstrate the effectiveness, robustness, and superiority of the proposed method in comparison with other methods.


  • Class Information-Guided Reconstruction for Automatic Modulation Open-Set Recognition

    by Ziwei Zhang, Mengtao Zhu, Jiabin Liu, Yunjie Li, Shafei Wang

    Automatic Modulation Recognition (AMR) is vital for radar and communication systems. Traditional AMR operates under closed-set scenarios where all modulation types are predefined. However, in practical settings, unknown modulation types may emerge due to technological advancements. Closed-set training poses the risk of misclassifying unknown modulations into existing known classes, leading to serious implications for situation awareness and threat assessment. To tackle this challenge, this paper presents a Class Information guided Reconstruction (CIR) framework that can simultaneously achieve Known Class Classification (KCC) and Unknown Class Identification (UCI). The CIR leverages reconstruction losses to differentiate between known and unknown classes, utilizing Class Conditional Vectors (CCVs) and a Mutual Information (MI) loss function to fully exploit class information. The CCVs offer class-specific guidance for reconstruction process, ensuring accurate reconstruction for known samples while producing subpar results for unknown ones. Moreover, to enhance distinguishability, an MI loss function is introduced to capture class-discriminative semantics in latent space, enabling closer alignment with CCVs during reconstruction. The synergistic relationship between CCVs and MI facilitates optimal UCI performance without compromising KCC accuracy. The CIR is evaluated on simulated, public and real-world datasets, demonstrating its effectiveness and robustness, particularly in low SNR and high unknown class prevalence scenarios.


  • Open Access

    Class Information Guided Reconstruction for Automatic Modulation Open-Set Recognition

    by Ziwei Zhang, Mengtao Zhu, Jiabin Liu, Yunjie Li, Shafei Wang

    Automatic Modulation Recognition (AMR) is a crucial technology in the domains of radar and communications. Traditional AMR approaches assume a closed-set scenario, where unknown samples are forcibly misclassified into known classes, leading to serious consequences for situation awareness and threat assessment. To address this issue, Automatic Modulation Open-set Recognition (AMOSR) defines two tasks as Known Class Classification (KCC) and Unknown Class Identification (UCI). However, AMOSR faces core challenges in terms of inappropriate decision boundaries and sparse feature distributions. To overcome the aforementioned challenges, we propose a Class Information guided Reconstruction (CIR) framework, which leverages reconstruction losses to distinguish known and unknown classes. To enhance distinguishability, we design Class Conditional Vectors (CCVs) to match the latent representations extracted from input samples, achieving perfect reconstruction for known samples while yielding poor results for unknown ones. We also propose a Mutual Information (MI) loss function to ensure reliable matching, with upper and lower bounds of MI derived for tractable optimization and mathematical proofs provided. The mutually beneficial CCVs and MI facilitate the CIR attaining optimal UCI performance without compromising KCC accuracy, especially in scenarios with a higher proportion of unknown classes. Additionally, a denoising module is introduced before reconstruction, enabling the CIR to achieve a significant performance improvement at low SNRs. Experimental results on simulated and measured signals validate the effectiveness and the robustness of the proposed method.

  • JDMR-Net: Joint Detection and Modulation Recognition Networks for LPI Radar Signals

    by Ziwei Zhang, Mengtao Zhu, Yunjie Li, Shafei Wang

    Low probability of intercept (LPI) radars are widely used in modern electromagnetic environments due to their excellent anti-interception performance. However, this inevitably increases the difficulties in detecting and recognizing LPI radar signals for electronic support systems or radar warning receivers. To address this challenge, this paper proposes a multi-task neural network named JDMR-Net for joint detection and modulation recognition of LPI radar signals. The inherent multi-task learning capability obtains an improved performance through leveraging useful information across tasks. The JDMR-Net receives pulse sequence in I/Q format as input and is computational friendly compared to time-frequency image-based methods. The JDMR-Net consists of a local feature extraction module and a global similarity mining module. The local feature extraction module extracts modulation information within single pulse, while the global similarity mining module determines the similarity relationship among sequential pulses. The JDMR-Net can provide accurate time domain localization of detected pulses, and determine corresponding modulation type simultaneously. Through the multi-task framework, the processing steps of traditional processing chain are compressed efficiently and the two modules are highly parallelizable, making the proposed solution promising for on-line application with raw signal inputs. Extensive experiments on simulated and measured LPI signals demonstrate the effectiveness and robustness of the proposed method in terms of lower detectable signal to noise ratio (SNR) and low computational complexity.