Journal on Communications ›› 2020, Vol. 41 ›› Issue (11): 12-21.doi: 10.11959/j.issn.1000-436x.2020237

• Topics:Key Technologies of Intelligent Wireless Communication for 6G • Previous Articles     Next Articles

Electromagnetic signal modulation recognition technology based on lightweight deep neural network

Sicheng ZHANG1,Yun LIN1,Ya TU1,Shiwen Mao2   

  1. 1 College of Information and Communication Engineering,Harbin Engineering University,Harbin 150001,China
    2 Department of Electrical and Computer Engineering,Auburn University,Auburn 36849,USA
  • Revised:2020-11-06 Online:2020-11-25 Published:2020-12-19
  • Supported by:
    The National Natural Science Foundation of China(61771154);The Fundamental Research Funds for the Central Universities(3072020CF0813)


In response to the trend that in the 6th generation wireless (6G) era,mobile communications and artificial intelligence will be closely integrated,and a huge number of edge intelligent signal processing nodes will be deployed,an efficient and intelligent electromagnetic signal recognition model was proposed,which could be deployed on resource-constrained edge devices.The constellation diagram of electromagnetic signal was firstly drawn to visualize electromagnetic signal as a two-dimensional image,and color the constellation diagram according to the normalized point density to achieve feature enhancement.Then,a binary deep neural network was adopted to recognize the colored constellation diagrams.It was shown that the approach can guarantee a high recognition accuracy,which significantly reduced the model storage and calculation costs.For verification,the proposed approach was applied to the problem of electromagnetic signal modulation recognition.The experiment uses eight commonly used digital modulation signals and selects additive white Gaussian noise as the channel environment.The experimental results show that the scheme can achieve a comprehensive recognition rate of 96.1% under the noise condition of -6~6 dB,while the size of the network model is only 166 KB.Further,the execution time,when executed on a Raspberry Pi 4B,is only 290 ms.Compared to a full-precision network of the same scale,the accuracy is increased by 0.6%,the model is reduced to 1 26.16 ,and the running time is reduced to 1 2.37 .

Key words: 6G, edge intelligence, electromagnetic signal modulation recognition, image visualization, binary deep neur-al network

CLC Number: 

No Suggested Reading articles found!