Please use this identifier to cite or link to this item:
Full metadata record
DC FieldValueLanguage
dc.contributor.authorHiremath, Shrishail M-
dc.contributor.authorBehura, Sambit-
dc.contributor.authorKedia, Subham-
dc.contributor.authorDeshmukh, Siddharth-
dc.contributor.authorPatra, Sarat Kumar-
dc.identifier.citation25th National Conference on Communication, IISc Bangalore, India,20-23 February,2019en_US
dc.descriptionCopyright of this document belongs to proceedings publisher.en_US
dc.description.abstractDeep learning techniques have recently exhibited unprecedented success in classification problems with ill-defined mathematical models. In this paper, we apply deep learning for RF data analysis and classification. We present a novel method of using I-Q time samples to form images with ‘Time and Discrete Orthonormal Stockwell Transform Domain Channels’ which are used for training a convolutional neural network (CNN) for radio modulation classification. Also, a concept inspired from transfer learning is used in extending the number of output classes of the CNN, which helps the network to estimate the approximate SNR of the input signal as well and further improve the classification accuracy. Such a network trained on Time and Stockwell Channeled Images performs superior to similar networks that are trained on just raw I-Q time series samples or timefrequency images, especially when training samples are less. The network achieved an overall classification accuracy of 97.3% at 8 dB SNR over a class of 10 radio modulation schemes (for both digital and analog systems). The study shows that such a trained network can be well applied to achieve high classification accuracies at low and moderate SNR scenarios.en_US
dc.subjectDeep learningen_US
dc.subjectConvolutional neural networken_US
dc.subjectModulation classificationen_US
dc.subjectStockwell transformen_US
dc.subjectDiscrete orthogonal Stockwell transformen_US
dc.subjectTime and Stockwell domain channeling.en_US
dc.titleDeep Learning-Based Modulation Classification Using Time and Stockwell Domain Channelingen_US
Appears in Collections:Conference Papers

Files in This Item:
File Description SizeFormat 
2019_NCC_SBehura_Deep Learning.pdfPaper3.35 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.