A General Inference Framework for Deep Neural Network of Modulation Recognition

Abstract

Modulation recognition is one of the crucial tasks in intelligent communications. With the development of deep learning, modulation recognition based on deep neural networks has attracted significant attention. Meanwhile, with development of internet of things as well as edge computing, various embedded devices have emerged. Consequently, how to deploy the deep neural network of modulation recognition on embedded devices becomes a research hotspot. Existing inference frameworks for the deep neural network of modulation recognition are highly dependent on the hardware platform, suffer from weak universality, and cannot be widely transplanted into various embedded devices. To solve this problem, this paper proposes a general inference framework for the modulation recognition network. The framework is built with the standard C language library, which is generally supported by embedded devices, to construct all the operators in the deep neural network, so as to ensure that the deployment of the framework is not limited by the hardware platform. Test results show that the inference framework proposed in this paper can run well on various embedded devices and achieve modulation recognition without accuracy loss.