关键词: 微弱信号盲检测, 直接序列扩频, 多域动态融合, 协同注意力机制, 跨层级融合

Abstract: Direct-sequence spread-spectrum (DSSS) signals are widely used in fields such as non-cooperative space reconnaissance due to their robust anti-interference and concealment capabilities. However, target signals intercepted by reconnaissance satellites are typically extremely weak, causing the performance of traditional detection methods to severely degrade or even fail under extremely low signal-to-noise ratio (SNR) conditions (SNR≤-20dB). To address this challenge, this paper proposes a multi-domain fusion collaborative attention network for the blind detection of weak spread-spectrum signals. First, multi-physical domain features of the signal are constructed, and a cascaded dynamic fusion module is designed to achieve adaptive weighting and correlation of the multi-modal features. Second, a Swin Transformer incorporated with an Efficient Local Attention (ELA) mechanism is adopted as the backbone network to enable the collaborative extraction of global contextual information and fine-grained local features. Furthermore, to overcome the information loss of multi-domain features during the network downsampling process, a multi-branch cross- level fusion strategy is proposed; specifically, at the end of each network stage, deep semantic features are residually connected with the original modal features via the dynamic fusion module to achieve cross-domain synergy. Simulation results demonstrate that under harsh conditions with extremely low SNRs, the proposed network can effectively suppress complex noise interference, and its detection performance for weak spread-spectrum signals significantly outperforms existing baseline methods.

Key words: Blind detection of weak signals, Direct-sequence spread-spectrum (DSSS), Multi-domain dynamic fusion, Collaborative attention mechanism, Cross-level fusion