An automated modulation classification approach utilizing progressive neural architecture search

In the past few years, deep learning techniques have gained widespread adoption in signal processing, delivering impressive outcomes. These methods leverage expert-designed neural network models to automatically extract valuable signal features from large datasets. However, manually designing deep neural network architectures continues to be labor-intensive and prone to errors. To tackle this challenge, a novel approach for Automatic Modulation Classification (AMC) based on progressive neural architecture search has been developed. This technique can autonomously generate network architectures tailored to specific modulation classification tasks while identifying the most efficient lightweight deep neural network through a search strategy aimed at optimizing model performance. Experimental results demonstrate that, compared to traditional deep learning-based modulation classification approaches, the proposed method achieves superior classification accuracy without requiring manual design of neural networks. Additionally, it boasts reduced parameter counts and floating-point operations, achieving an average recognition accuracy of 92.75%.