Radio Frequency-Retentive Network for Automatic Modulation Classification

Abstract

As a common classification task in noncooperative communication scenarios, automatic modulation classification (AMC) is mainly used to quickly demodulate signals and is often deployed in different hardware architectures. In this paper, a radio frequency-retentive network (RF-ReNet) model is proposed to improve the retentive network architecture, which can convert between the parallel and serial modes and can conduct training or inference on different hardware architectures. The experimental results show that the classification accuracy of the proposed RF-ReNet is not greatly reduced, and its spatiotemporal complexity is lower than that of the existing deep models.