Fast Adaptive Automatic Modulation Classification Under Non-Gaussian Impulsive Noise

In this letter, we investigate automatic modulation classification (AMC) under non-Gaussian impulsive noise modeled by the \alpha -stable distribution. We propose an RGB Gramian Angular Field (RGB-GAF) representation that transforms complex baseband I/Q signals into three-channel images to enhance feature extraction. Based on this representation, we develop the Feature-wise Transfer Network (FiTNet), a meta-learning framework that integrates pre-training, feature-wise linear modulation (FiLM), and hard-task (HT) mining for parameter-efficient adaptation. Numerical results show that the proposed method achieves fast adaptation and improved classification accuracy across diverse unseen noise environments.