Cross-Validated Cross-Channel Self-Attention and Denoising for Automatic Modulation Classification

arXiv cs.LG / 4/14/2026

💬 OpinionSignals & Early TrendsIdeas & Deep AnalysisModels & Research

共有:

Key Points

The paper targets a limitation of deep learning Automatic Modulation Classification (AMC): performance drops sharply at low/noisy SNR because conventional feature extraction can suppress both discriminative signal structure and interference-relevant information.
It proposes an AMC architecture that combines a cross-channel self-attention block (linking in-phase and quadrature components) with dual-path deep residual shrinkage denoising blocks to preserve modulation features while reducing noise.
Experiments on the RML2018.01a dataset use stratified sampling across 24 modulation classes and 26 SNR levels, showing that the denoising depth is a key factor for robustness in low and moderate SNR regimes.
Compared with benchmark models (PET-CGDNN, MCLDNN, DAE), the method reports accuracy gains across -8 dB to +2 dB SNR, including a particularly large improvement over DAE.
Cross-validation results report mean accuracy of 62.6% and macro-F1 of 62.9%, and ablation studies emphasize that feature-preserving denoising plus cross-channel attention are essential for low-to-medium SNR robustness.

Abstract

This study addresses a key limitation in deep learning Automatic Modulation Classification (AMC) models, which perform well at high signal-to-noise ratios (SNRs) but degrade under noisy conditions due to conventional feature extraction suppressing both discriminative structure and interference. The goal was to develop a feature-preserving denoising method that mitigates the loss of modulation class separation. A deep learning AMC model was proposed, incorporating a cross-channel self-attention block to capture dependencies between in-phase and quadrature components, along with dual-path deep residual shrinkage denoising blocks to suppress noise. Experiments using the RML2018.01a dataset employed stratified sampling across 24 modulation types and 26 SNR levels. Results showed that denoising depth strongly influences robustness at low and moderate SNRs. Compared to benchmark models PET-CGDNN, MCLDNN, and DAE, the proposed model achieved notable accuracy improvements across -8 dB to +2 dB SNR, with increases of 3%, 2.3%, and 14%, respectively. Cross-validation confirmed the model's robustness, yielding a mean accuracy of 62.6%, macro precision of 65.8%, macro-recall of 62.6%, and macro-F1 score of 62.9%. The architecture advances interference-aware AMC by formalizing baseband modeling as orthogonal subproblems and introducing cross-channel attention as a generalized complex interaction operator, with ablations confirming the critical role of feature-preserving denoising for robustness at low-to-medium SNR.