基於 Swin-Transformer 語意分割的雷達訊號解交織之研究;Radar Signal Deinterleaving Based On Swin-Transformer Segmentation

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Communication Engineering > Electronic Thesis & Dissertation > Item 987654321/95332

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/95332

Title:	基於 Swin-Transformer 語意分割的雷達訊號解交織之研究;Radar Signal Deinterleaving Based On Swin-Transformer Segmentation
Authors:	王濰翊;Wang, Wei-Yi
Contributors:	通訊工程學系
Keywords:	解交織;脈衝流;PRI調變;Transformer;語意分割;Deinterleaving;pulse streams;PRI modulation;transformer;semantic segmentation
Date:	2024-07-22
Issue Date:	2024-10-09 16:40:37 (UTC+8)
Publisher:	國立中央大學
Abstract:	在電子戰（EW）不斷發展的背景下，先進的電子支援措施（ESM）系統已變得不可或缺。傳統的解交織技術，在 ESM 中透過脈衝描述字（PDWs）進行發射源分類是基礎，但現在由於訊號參數的增加複雜性、變化以及在電子戰場景中訊號攔截幾率的降低，已不再適用。本文提出了一種新的解交織演算法，主要利用到達時間（TOA）值來克服當前技術的局限性，這些技術在沒有到達方向（DOA）資料的情況下過度依賴 PDWs。為了處理複雜的脈衝重複間隔（PRI）調變和遺失脈衝情況，我們將影像分割應用於雷達解交織中進行創新。我們的方法使用 Swin Transformer 模型，將 TOA 資料轉換成二維影像格式以進行更精細的分析。我們將這項技術稱為像素級 PRI 影像分類，它允許在複雜的 PRI 情境中準確識別脈衝序列。與最先進的基於深度學習的方法相比，我們的方法在有效解交織所需的脈衝數上顯著減少。模擬結果也表明，尤其是在遺失或虛假脈衝的場景中，我們的模型優於現有基於直方圖的和先進學習的方法。我們的雷達序列資料集可以在 https://github.com/ICAN-Lab/PRI-Frequency-Image-Dataset 上公開下載。;In the evolving landscape of electronic warfare (EW), advanced electronic support measure (ESM) systems have become imperative. Traditional deinterleaving techniques, which are fundamental in ESM for emitter categorization through pulse description words (PDWs), are now inadequate due to the increased complexity, variation in signal parameters, and lower chances of signal interception in EW scenarios. This paper presents a new deinterleaving algorithm that primarily utilizes time of arrival (TOA) values to overcome the limitations of current techniques that depend heavily on PDWs, particularly in situations where direction of arrival (DOA) data is not available. To handle complex pulse repetition interval (PRI) modulations and missing pulse situations, we innovate by applying image segmentation to radar deinterleaving. Using the Swin Transformer, our method transforms TOA data into a two-dimensional image format for enhanced analysis. We refer to this technique as \textit{pixelwise PRI image classification}, which allows for accurate identification of pulse sequences in complex PRI situations. Compared to the state-of-the-art deep learning based method, our method achieves a significant decrease in the number of pulses needed for effective deinterleaving. Simulation results also show that our model outperforms existing histogram-based and advanced learning-based methods, especially in scenarios with lost or spurious pulses. Our radar sequence dataset can be publicly downloaded at https://github.com/ICAN-Lab/PRI-Frequency-Image-Dataset.
Appears in Collections:	[Graduate Institute of Communication Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	18	View/Open

社群 sharing

Loading...