IQGCA: A Lightweight Graph Structure of I/Q Signal for Automatic Modulation Recognition

Automatic modulation recognition (AMR) is a key technique in the Internet of Vehicles (IoV), which can help to communicate efficiently. However, the conflict between computational complexity and high recognition efficiency limits the application of AMR. To reduce computational costs and achieve high recognition accuracy, we propose a method named I/Q signal graph coarsening attention network (IQGCA). In the proposed method, the graph structure is generated by mapping 3D I/Q signal onto different planes to learn the temporal correlation and spatial distribution. Consequently, a graph coarsening attention (GCA) block is introduced, which requires few computational costs to coarsen graph structure and extract features. Experimental results on datasets with different sample lengths, RadioML2016.10a and RadioML2018.01a, demonstrate that the proposed IQGCA is superior to state-of-the-art (SOTA) methods in recognition accuracy and lightweight model design.