IQTimesNet: A Time-Frequency Hybrid Framework for Robust Automatic Modulation Recognition

Automatic modulation recognition (AMR) plays a crucial role in wireless communications. Although deep neural networks have achieved great progress in this field, existing research has not fully utilized the various features present in radio signals. In order to address this issue, a time-frequency hybrid framework is proposed, named IQTimesNet. It uses I/Q signals and Fourier-transformed data as inputs. Neural networks are used to simultaneously extract temporal features and spatial-periodic features. Additionally, the dual-stages encoder block is designed to extract more comprehensively local and global features. Experimental results on RadioML2016.10a and RadioML2016.10b demonstrate that the performance of the proposed IQTimesNet is superior to other state of the art (SOTA) methods.