Classification Method Ofdigital-Modulation-Signals Based On Neuralnetwork

The rapid development of communication field and the new requirements of winning the information war determine that there are many kinds of signal modulation styles, which requires us to explore the classification method of different modulation styles. At present, the identification method of the signal has many kinds,which main based on the log likelihood ratio, the zero crossing method, the digital phase correlation variable identification method,the decision theory recognition method and so on. This paper main studies the recognition method based on neural network.ANN is based on the human body's nervous system, to build a model to solve the problem of information processing, so as to achieve intelligently different functions. According to its internal structure and algorithm, and is divided into a variety of network, of course, different networks may have different advantages and disadvantages, in the realization of different functions also can have.Based on the ANN technology, the paper studies the six kinds of digital modulation signals of 2ASK, 4ASK, 2FSK, 4FSK, BPSK and QPSK. Firstly,several characteristic parameters have been established through the study of modulation techniques for digital modulation signals, and the process of signal classification and recognition is designed.Secondly,according to the characteristics of signal classification and research direction, this paper selects the RBF, and based on MATLAB software platform, the classification of several kinds of digital modulation signals is simulated in RBF network, which mainly include signal generation experiment(adding Gauss white noise), neural network training, single RBF network identification signal experiment.Lastly,based on the core of RBF network center selection, this paper is improved in the algorithm, fuse the recursive orthogonal least squares and RBF neural network, and carries on the simulation experiment, and compares with the results with a single RBF network identification, we find that the improved network has a significant improvement in the convergence rate of network training and the identification rate of the signal.