<p>The secondary radar signals are important for air traffic management. However, the inevitable signal overlapping poses significant challenges for signal decoding, making blind separation a critical issue in radar signal processing. In the previous studies, the rule-based methods perform poorly under low signal-to-noise ratios (SNR) and minor relative time delays between raw signals. While deep learning-based (DL-based) methods have been employed to improve separation accuracy, the deep network structure suffers from feature degradation, resulting in insufficient signal reconstruction performance. To address these issues, this paper proposes a skip connection-enhanced hybrid neural network (Sc-HNN) for effective single-channel blind separation. The overlapping signals are pre-processed first to reduce separation loss during convolution operations. Then, the Sc-HNN decomposes the input into multi-channel features and extracts the separation mask to reconstruct each source signal. The skip connections facilitate the transmission of low-level information across different network hierarchies, thereby improving the reconstruction accuracy of weak sources. Numerical results show that the proposed method outperforms conventional and DL-based approaches regarding decoding success rates, signal-to-distortion ratios, and mean squared errors.</p>