Automatic Modulation Classification (AMC) is a critical task in cognitive radio and electronic warfare, enabling the blind identification of a signal's modulation scheme at the receiver. A significant challenge to reliable AMC is the presence of channel-induced distortions, such as carrier frequency offset (CFO) and phase noise, which severely degrade classification accuracy, particularly in low Signal-to-Noise Ratio (SNR) environments. This paper proposes a novel, learnable Distortion Correction Module (CM) based on a deep neural network architecture. The CM is designed to be co-trained end-to-end with a Convolutional Neural Network (CNN) classifier, forming a CM+CNN system. The CM acts as a channel parameter estimator, dynamically correcting the distorted signal before it reaches the classifier. Unlike traditional methods, this approach is entirely data-driven and does not require explicit knowledge of the channel parameters for training, relying only on the modulation scheme label. Through comprehensive evaluation, the proposed CM+CNN system demonstrates a substantial improvement in AMC accuracy across various modulation types and channel conditions, establishing a more robust and reliable solution for non-cooperative communication systems. This work contributes to UN Sustainable Development Goal 9 (Industry, Innovation and Infrastructure) by improving the robustness and efficiency of intelligent wireless communication systems through data-driven distortion correction for reliable modulation recognition in challenging channel conditions.
 

Automatic Modulation Classification (AMC), Deep Learning, Distortion Correction, Cognitive Radio, Convolutional Neural Networks (CNN).