Imals 2021, 11,eight of
electronicsArticleA Lightweight CNN Architecture for Automatic Modulation ClassificationZhongyong Wang
Imals 2021, 11,8 of
electronicsArticleA Lightweight CNN Architecture for Automatic Modulation ClassificationZhongyong Wang , Dongzhe Sun , Kexian Gong , Wei Wang and Peng Sun School of Information and facts Engineering, Zhengzhou University, Zhengzhou 450001, China; Charybdotoxin Biological Activity [email protected] (Z.W.); [email protected] (D.S.); [email protected] (K.G.); [email protected] (W.W.) Correspondence: [email protected]: Automatic modulation D-Fructose-6-phosphate disodium salt Metabolic Enzyme/Protease classification (AMC) algorithms determined by deep finding out (DL) happen to be broadly studied in the past decade, displaying important overall performance benefit in comparison with classic ones. However, the current DL solutions commonly behave worse in computational complexity. For this, this paper proposes a lightweight convolutional neural network (CNN) for AMC task, where we style a depthwise separable convolution (DSC) residual architecture for function extraction to prevent the vanishing gradient difficulty and lighten the computational burden. In addition to that, in an effort to additional cut down model complexity, international depthwise convolution (GDWConv) is adopted for feature reconstruction soon after the final (non-global) convolutional layer. In comparison with recent operates, the experimental benefits show that the proposed network can save roughly 70 98 model parameters and 30 99 inference time on two well-known benchmarks. Search phrases: automatic modulation classification; convolutional neural network; depthwise separable convolution; function reconstruction; international depthwise convolutionCitation: Wang, Z.; Sun, D.; Gong, K.; Wang, W.; Sun, P. A Lightweight CNN Architecture for Automatic Modulation Classification. Electronics 2021, ten, 2679. https://doi.org/ 10.3390/electronics10212679 Academic Editor: Amir Mosavi Received: 4 October 2021 Accepted: 30 October 2021 Published: 2 November1. Introduction Automatic modulation classification (AMC) is really a very important technologies involving signal detection and demodulation in non-cooperative communication scenarios. AMC implies to non-cooperatively classify the modulation scheme of a received radio signal, which is often regarded as a multi-class selection problem. As the foundation of signal demodulation, the correctness of AMC directly determines whether valid data is often recovered in the received signal. Fast and accurate AMC of wireless signals is widely applied in different civilian and military fields, which include spectrum monitoring, radio fault detection, automatic receiver configuration, and signal interception and jamming [1]. Traditional AMC strategies might be divided into two categories: likelihood primarily based [4] and feature primarily based [5,6]. The likelihood-based procedures calculate likelihood function of candidate modulations and select the modulation mode with maximal likelihood worth. This method treats AMC as a multi-hypothesis test issue, whose implementation is impractical resulting from its high computational complexity. The regular feature-based AMC algorithms is usually realized by two actions: function extraction and classificatory decision. For feature extraction, the common realization strategies include wavelet transform-based attributes, high-order statistical capabilities, cyclic spectrum-based characteristics and so on. For classificatory choice, readily available classifiers include decision tree, help vector machine (SVM), totally connected neural network and so on. The feature-based ones behave effectively in some certain circumstances, whose performances, even so, are restricted by the design of manual characteristics when the systems include challe.
