Noise Resilient Modulation Recognition for Digitalized Power Grids Using SSA and TCN

In digitalized power grids, particularly high-voltage direct current systems, wireless sensor nodes are deployed to monitor voltage, current, and equipment status in real time. However, electromagnetic interference (EMI), arc discharges, and switching transients under high-voltage conditions often degrade signal quality and reduce the signal-to-noise ratio (SNR), posing significant challenges to automatic modulation recognition (AMR). We propose a hybrid SSA-TCN architecture fusing Singular Spectrum Analysis (SSA) for adaptive noise suppression and a Temporal Convolutional Network (TCN) for robust modulation classification. SSA decomposes the trajectory matrix of the received signal to isolate dominant components from noise, while TCN employs dilated causal convolutions to capture long-range temporal dependencies at the symbol level. Designed for deployment on wireless sensing units installed on high-voltage transmission towers, the proposed method enhances recognition accuracy under low SNR or fading conditions. Experimental results demonstrate that SSA-TCN significantly outperforms conventional AMR techniques, making it suitable for edge-side signal processing in smart grid scenarios such as real-time monitoring, fault detection, and condition-based maintenance.