A Study On Modulation Classification Of Digital Modulated Communication Signals

Modulation recognition is a technique to classify the modulation type of a received signal. It's a rapidly evolving signal analysis area and has found a variety of applications both in military and civil communication systems. Modulation recognition can be divided into two categories: inter-class and intra-class. Inter-class means the classification of different modulation types such as ASK, PSK and FSK. Intra-class refers to distinguishing within a single class. It's the classification of modulation level about the same modulation type.In this research, based on statistical pattern recognition method, we use constellation of modulated signals to classify modulation modes both in non-cooperative and cooperative communications.In non-cooperative communications, wavelet transform (WT) is used to extract the constellation of digital modulated signals. WT is an efficient tool for signal singularity detection. We utilize the transient characters of the magnitudes of WT coefficients with multiple synchronizing reference points correcting technique to obtain a more accurate estimation of symbol rate and symbol synchronization than traditional methods. After then, we get the constellation of input signal. To classify the constellation of modulated signals, clustering algorithm is used. Clustering is an unsupervised classification method. It uses either the data similarity or pdf of data set for classification. In this research hierarchical clustering algorithm is used with a newly defined criterion function, and it can cluster data without known cluster number. Combined WT and clustering, we get a new modulation classification algorithm for MPSK/MQAM signals. With this method no a-prior knowledge about modulated signal are requested and a comparable good recognition result can be obtained under low SNR environment.In cooperative communications, a study on modulation recognition of adaptive OFDM system is given. Based on hypothesis testing of constellations, we utilize the error gap of SNR estimation between different hypotheses to classify modulation blindly. This method is of low computational complexity and performs better than ML and minimum-error algorithm. And the simulation shows that even with very short size of signal, it gives reliable detection result.