A Semantic Embedded Deep Clustering Framework With Adaptive Merging for Radar Signal Sorting

Zhihang Zhang

doi:10.1109/taes.2026.3705357

A Semantic Embedded Deep Clustering Framework With Adaptive Merging for Radar Signal Sorting

2026-06-18

0

PRE

AI

DOI：10.1109/taes.2026.3705357

Original

Original request for help

Save

Abstract

En 中文

Radar signal sorting is a fundamental task in electronic warfare, supporting battlefield awareness, target detection, and electronic reconnaissance. Conventional rule-based approaches suffer from limited pulse semantics, dynamically varying multidimensional parameters, and severe pulse overlap, leading to degraded performance in complex electromagnetic environments. This article proposes a semantic embedded deep clustering framework with adaptive merging for robust radar signal sorting. A semantic embedding module is first introduced to map pulse descriptor word (PDW) parameters into continuous high-dimensional semantic vectors, yielding enriched pulse representations. Positional and contextual information is further integrated to enhance temporal and structural awareness. Subsequently, a channel–variable residual deep clustering module is developed to improve interemitter separability and intraemitter compactness in the embedded feature space. To address batch fragmentation caused by parameter jitter and modulation diversity, a temporal–channel–variable adaptive merging module is designed to model pulse repetition interval evolution and cross-parameter dependencies. Experiments conducted on overlapped PDW data from multiple radar emitters demonstrate that the proposed method consistently outperforms state-of-the-art radar signal sorting methods.

Keywords:

Adaptive merging

deep clustering

radar signal sorting

sequential semantic encoding

A Semantic Embedded Deep Clustering Framework With Adaptive Merging for Radar Signal Sorting

Abstract

Journal

Organization