CACF-Net: A Cross-Attention Cognitive Fusion Network for Robust and Interpretable Signal Recognition

Abstract

With the development of wireless communications, the electromagnetic environment has become increasingly complex. Reliable signal recognition is fundamental for spectrum awareness and intelligent management. However, single-domain features are vulnerable to noise and fading, leading to limited performance. To address this, we propose a Cross-Attention Cognitive Fusion Network (CACF-Net) for robust cross-domain signal recognition. The network integrates time, frequency, and statistical domains. Specifically, conflict modeling is incorporated into self-attention, and adaptive weighting is applied during crossattention to capture inter-domain correlations. Furthermore, evidence theory is introduced for joint decision-making, which reduces uncertainty and provides quantifiable interpretation. Experiments on the RadioML 2018.01A dataset show that CACF-Net outperforms baseline models across all signal-to-noise ratios. Moreover, the network effectively suppresses misleading information from interference domains.