Automatic digital modulation recognition based on stacked sparse autoencoder

In this paper, a modulation recognition algorithm based on Stacked sparse Auto-Encoder (SSAE) is proposed for the classification of common digitally modulated signals. To this end, a set of eight features including, two instantaneous features and six higher order cumulants features are extracted fro...

Full description

Saved in:

Bibliographic Details
Published in:	Proceedings (International Conference on Communication Technology. Online) pp. 28 - 32
Main Authors:	Bouchou, Mohamed, Wang, Hua, El Hadi Lakhdari, Mohammed
Format:	Conference Proceeding
Language:	English
Published:	IEEE 01.10.2017
Subjects:	Automatic Modulation Recognition (AMR) Classification algorithms Feature extraction Higher Order Cumulants (HOC) Phase shift keying Signal to noise ratio Stacked Sparse AutoEncoder (SSAE) Support vector machines Training
ISSN:	2576-7828
Online Access:	Get full text
Tags:	Add Tag No Tags, Be the first to tag this record!

Description
Summary:	In this paper, a modulation recognition algorithm based on Stacked sparse Auto-Encoder (SSAE) is proposed for the classification of common digitally modulated signals. To this end, a set of eight features including, two instantaneous features and six higher order cumulants features are extracted from the intercepted signal; these features are then fed to the SSAE for classification. Unlike the majority of classifiers used in AMR algorithms, which relies only on the supervised learning scenario, the stacked sparse autoencoder benefits from both, unsupervised and supervised learning approaches. In fact, the main advantage of the SSAE is that it can automatically learn new features to separate the input data during the unsupervised pre-training phase. These new features are used as initialization parameters in the supervised training phase to enhance the convergence of the SSAE to optimal results, as well as improve the noise resistance of the eight features extracted before. Results show that the overall success rate reach 100 % at 5dB SNR. The performance of the proposed algorithm is compared to an SVM-based method, and it is found that the probability of correct classification in our method is considerably improved.
ISSN:	2576-7828
DOI:	10.1109/ICCT.2017.8359478