Modulation Classification Algorithms Of Digital Communication Signals

Automatic modulation classification (AMC) is a technology to recognize the modulation type of a received signal which transmits through a channel usually corrupted with noise and multi-path fading. There is an emerging need for the intelligent system to correctly and quickly recognize the modulation type of a received signal which is known little so that arguments can be provided to process the signal in later step. AMC has become one of the key measures to ensure licit communications, and plays a key role in various civilian and military applications. With rapid development of communication technologies, communication system becomes complicated and various modulation types are adopted, which determines AMC is still a challenging subject. In this thesis, feature-based automatic modulation classification algorithms are researched and improved.Clustering algorithm is an important method in data mining, which has powerful capability of pattern recognition and can tackle complicated nonlinear problems. In this thesis, a new modulation classification algorithm is proposed based on the combination of clustering and neural network. In order to recognize modulation types based on the constellation diagram such as phase shift keying (PSK) and quadrature amplitude modulation (QAM), Fuzzy C-means (FCM) clustering is adopted for recovering the constellation under different number of clusters. Then cluster validity measure is applied to extract key features which discriminate between different modulation types. The features are sent to neural network so that modulation types can be recognized. In order to conquer the disadvantages of standard back propagation (BP) neural network, conjugate gradient learning algorithm of Polak-Ribiere update is employed to improve the speed of convergence and the performance of modulation recognition. Simulation results show that classification rates of the algorithm proposed in this paper are much higher than those of clustering algorithm.Support vector machine (SVM) is a pattern recognition method which is developed based on statistic learning theory and realizes optimal recognition theoretically. In this thesis, combining clustering and SVM method for automatic modulation classification is proposed. To recognize signals modulated based on constellation diagram, K-means clustering is adopted for recovering constellation under different number of clusters. Silhouette index is employed as a cluster validity measure to extract key features that discriminate between different modulation types. Then hierarchical SVM classifier is designed to recognize modulation types according to the key features extracted. Simulation results show that the classification rates of the algorithm proposed in this paper are also much higher than those of clustering algorithm.With the development of wireless sensor network (WSN), the advantages of distributed detection, estimation as well as classification algorithms are drawing more and more attention. In this thesis, a distributed cooperative recognition method is introduced to recognize different digital modulation types with multiple sensors in WSN. In order to enhance the successful recognition rate when signal noise ratio (SNR) is low and realize correct recognition of several classic modulation types such as MASK, MFSK, BPSK, QPSK and OFDM, effective cooperative methods are designed according to SNR of received signal and based on the principle of lowest sensor overhead. A new combination of features is extracted accordingly by several collaborated sensors to improve the performance of the modulation recognition system. Then the features are sent to the Radial Basis Function (RBF) neural network so that modulation types can be recognized. Further more, different cooperative methods are introduced in this paper to adapt to the condition of the sensor networks. To measure the performance of the proposed methods, simulations are carried out to classify different types of modulated signals which transmit through fading channels. The simulation results show that the proposed distributed cooperative algorithm has higher recognition rates with better system reliability compared with that without cooperation.