Scholarlink

<p>Frequency-hopping (FH) communication, widely utilized in both military and civilian applications, offers robust anti-jamming capabilities and low probability of interception. However, its non-stationary nature and susceptibility to noise complicate the estimation of signal parameters, especially in low signal-to-noise ratio (SNR) and non-cooperative scenarios. These challenges are further exacerbated by the need to process multimodal data, including time-frequency and noise features, simultaneously. To address these issues, we propose a novel framework for FH signal processing that combines advanced time-frequency analysis with a customized multimodal object detection network. High-resolution, denoised time-frequency representations are generated using the Short-Time Fourier Transform (STFT), time-frequency reconstruction, and Gaussian smoothing techniques. A lightweight detection network, FH-YOLO, is designed with a Universal Inverted Bottleneck (UIB), multi-scale fusion, and Depthwise Separable Coordinate Attention (DSCA) to enhance feature extraction and signal localization across different modalities. The detected bounding boxes are then mapped to time-frequency coordinates to enable accurate FH parameter estimation. Extensive evaluations on a dataset of 3,900 simulated signals, covering three modulation types and 26 SNR levels from <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(-{\textbf {15}}\)</EquationSource> </InlineEquation> dB to <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\({\textbf {10}}\)</EquationSource> </InlineEquation> dB, demonstrate the effectiveness of the method, achieving 97.9% mAP@50, 60.9% mAP@50-95, and real-time processing at 90.9 frames per second (FPS). Additionally, the proposed approach reduces frequency estimation errors by up to 50% compared to traditional methods, highlighting its robustness in complex, multimodal environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhanced FH-YOLO Framework for Robust Frequency Hopping Signal Parameter Estimation in Noisy Environments

  • 期刊论文

  • 发表日期：

    2025年8月1日

查看全文
  • Zhedong Wu,
  • Xiansheng Ge,
  • Zaishang Zhang,
  • Yuhai Du,
  • Kai Yang,
  • Fajun Lin

摘要

Frequency-hopping (FH) communication, widely utilized in both military and civilian applications, offers robust anti-jamming capabilities and low probability of interception. However, its non-stationary nature and susceptibility to noise complicate the estimation of signal parameters, especially in low signal-to-noise ratio (SNR) and non-cooperative scenarios. These challenges are further exacerbated by the need to process multimodal data, including time-frequency and noise features, simultaneously. To address these issues, we propose a novel framework for FH signal processing that combines advanced time-frequency analysis with a customized multimodal object detection network. High-resolution, denoised time-frequency representations are generated using the Short-Time Fourier Transform (STFT), time-frequency reconstruction, and Gaussian smoothing techniques. A lightweight detection network, FH-YOLO, is designed with a Universal Inverted Bottleneck (UIB), multi-scale fusion, and Depthwise Separable Coordinate Attention (DSCA) to enhance feature extraction and signal localization across different modalities. The detected bounding boxes are then mapped to time-frequency coordinates to enable accurate FH parameter estimation. Extensive evaluations on a dataset of 3,900 simulated signals, covering three modulation types and 26 SNR levels from \(-{\textbf {15}}\) dB to \({\textbf {10}}\) dB, demonstrate the effectiveness of the method, achieving 97.9% mAP@50, 60.9% mAP@50-95, and real-time processing at 90.9 frames per second (FPS). Additionally, the proposed approach reduces frequency estimation errors by up to 50% compared to traditional methods, highlighting its robustness in complex, multimodal environments.