Modulation classification for cognitive radios using stacked denoising autoencoders. (1st December 2016)

• Record Type: Journal Article
• Authors: Zhu, Xu; Fujii, Takeo
• Other Names: Tarchi Daniele guestEditor.; Chatzinotas Symeon guestEditor.; Vanelli‐Coralli Alessandro guestEditor.; Ottersten Bjorn guestEditor.
• Is Part Of: International journal of satellite communications and networking. Volume 35:Number 5(2017:Sep./Oct.)
• Page Start: 517
• Page End: 531
• Publication Date: 2016-12-01
• Subjects: modulation classification -- cognitive radio -- autoencoder -- satcomm; Artificial satellites in telecommunication -- Periodicals; Digital communications -- Periodicals; 621.3825
• DOI: 10.1002/sat.1202
• Languages: English
• ISSNs: 1542-0973

Abstract

Summary: This paper proposes a modulation classification method based on stacked denoising autoencoders (SDAE). This method can extract the modulation features automatically and classify the input signals based on the extracted features. The scenarios of rapid classification and high‐accuracy classification are considered. In a rapid classification scenario, the classification speed has priority over the classification accuracy. Therefore, a long‐symbol sequence is not attainable for this scenario. Moreover, expert features are not necessary for this scenario, simplifying the modulation classification procedure and rendering rapid classification more achievable. In addition, in a high‐accuracy classification scenario, higher cumulants are used as the expert features owing to their advantage over the other features at noise resistance. We use complex symbols rather than pulse shaped complex signals as the network inputs, simplifying the network topology and reducing the calculation overhead. The results of the average classification accuracy, the individual classification accuracy, the execution time and the influence of the signal sampling synchronization are presented, demonstrating significant performance advantages over the other methods. Copyright © 2016 John Wiley & Sons, Ltd. Abstract : This paper proposes a modulation classification method based on a stacked denoising autoencoder. We use complex symbols rather than pulse‐shaped complex signals as the network inputs, simplifying the network topology and reducing the calculation overhead. The results of the average classification accuracy, the individual classification accuracy, the execution time, and the influence of the signal sampling synchronization are presented, demonstrating significant performance advantages over the other methods.