Please use this identifier to cite or link to this item:
|Title:||Automatic classification of digital communication signal modulations|
|Keywords:||Modulation classification;Channel estimation;Machine learning;Signal processing;Wireless communications|
|Abstract:||Automatic modulation classification detects the modulation type of received communication signals. It has important applications in military scenarios to facilitate jamming, intelligence, surveillance, and threat analysis. The renewed interest from civilian scenes has been fuelled by the development of intelligent communications systems such as cognitive radio and software defined radio. More specifically, it is complementary to adaptive modulation and coding where a modulation can be deployed from a set of candidates according to the channel condition and system specification for improved spectrum efficiency and link reliability. In this research, we started by improving some existing methods for higher classification accuracy but lower complexity. Machine learning techniques such as k-nearest neighbour and support vector machine have been adopted for simplified decision making using known features. Logistic regression, genetic algorithm and genetic programming have been incorporated for improved classification performance through feature selection and combination. We have also developed a new distribution test based classifier which is tailored for modulation classification with the inspiration from Kolmogorov-Smirnov test. The proposed classifier is shown to have improved accuracy and robustness over the standard distribution test. For blind classification in imperfect channels, we developed the combination of minimum distance centroid estimator and non-parametric likelihood function for blind modulation classification without the prior knowledge on channel noise. The centroid estimator provides joint estimation of channel gain and carrier phase o set where both can be compensated in the following nonparametric likelihood function. The non-parametric likelihood function, in the meantime, provide likelihood evaluation without a specifically assumed noise model. The combination has shown to have higher robustness when different noise types are considered. To push modulation classification techniques into a more timely setting, we also developed the principle for blind classification in MIMO systems. The classification is achieved through expectation maximization channel estimation and likelihood based classification. Early results have shown bright prospect for the method while more work is needed to further optimize the method and to provide a more thorough validation.|
|Description:||This thesis was submitted for the degree of Doctor of Philosophy and awarded by Brunel University|
|Appears in Collections:||Electronic and Computer Engineering|
Dept of Electronic and Computer Engineering Theses
Items in BURA are protected by copyright, with all rights reserved, unless otherwise indicated.