Blind Source Separation for Wireless Signals Based on GCU-Enhanced Dual-Path Model

Blind source separation (BSS) has emerged as a significant research direction in deep learning, with widespread applications across various domains. With the advancement of wireless communication technology, some studies have explored its application in spectrum sensing to enhance detection accuracy and identify unauthorized signals. However, research on wireless signal separation based on deep learning remains relatively limited. In this paper, we propose a dual-path deep learning model enhanced by a Gated Convolutional Unit (GCU), termed DPGCU, for BSS of complex mixed signals comprising ten classes across five modulation types. The proposed model integrates a GCU into the dual-path architecture to extract global signal features and enhance the feature extraction capability by effectively combining intra-block, inter-block, and global contextual information. Experimental results demonstrate that the proposed model reduces the separation error probability by 1.5 × 10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">-3</sup> under noise-free conditions.