Research On Automatic Modulation Recognition Of Communicaton Signal

Modulated digital communication signals in the automatic recognition of non-cooperative communication has important significance. This digital communication signals for automatic identification of modulation are analyzed, summarized and further research. The main achievements can be summarized as follows:First, in the Gaussian white noise, the comparison of several time-domain characteristic parameters of the modulation recognition performance, through several time-domain characteristic parameters distinguish 2FSK, BPSK, QPSK,8PSK, OQPSK, and 16QAM signals. Several characteristic parameters were analyzed at different SNR changes, and use the decision tree algorithm obtained the corresponding recognition rate curve.Secondly, in the frequency domain of several common spectral characteristics of digital modulation signals were analyzed. Since different modulation signals tend to produce different spectral characteristics. This paper focuses on the square of the signal spectrum and the fourth power of the spectrum, given the discrete spectrum quantitative detection of defined parameters. Spectral characteristics of the signal in the frequency domain low computational complexity, the modulation parameters have better robustness.Then, based on higher order cumulants of the modulation signal recognition method, we first give the definition of a higher order statistics and the corresponding estimation method. MPSK signals for class identification problem, given a 4-order cumulant-based invariant feature of the recognition. Simulation results show that the characteristic quantities of the MPSK signal has better recognition performance. For amplitude and phase modulated signal, according to the different characteristics of the signal, the first signal is divided into five categories, and then were two signals for a subset of them to identify, to simplify the identification process. The simulation results show the effectiveness of classification algorithms.Finally, the support vector machine based modulation classification and recognition algorithms. Analysis of support vector machine principle of empirical risk minimization and structural risk minimization principle, then introduces the optimal SVM classification surface. In order to solve multi-classification problem, a multisection class SVM's classification. On this basis, select the Classification of the parameters of the SVM classifier are modulated by classification.