基于改进RBF神经网络的数字调制识别

Digital Modulation Recognition Based on an Improved Method of RBF Neural Network

针对数字调制信号自动识别中分类器的设计,通过将决策树的方法应用到RBF中心的确定中,解决了常用算法计算量大、收敛速度慢的问题,提高了网络的学习精度和训练速度,将其应用到常用的7种数字调制信号(2ASK,4ASK,BPSK,QPSK,2FSK,4FSK,16QAM)的自动识别中,取得了好的结果。经仿真表明,使用该方法构造的神经网络,具有易于构造、可理解性好、收敛速度快且构造的网络规模较小的特点,适于工程应用。

A novel algorithm for automatic recognition of digital modulation is proposed to solve such issues as lower accuracy of RBF and slow training speed using normal method. The algorithm based on decision tree of RBF neural network used in the center determination to improve the accuracy and the learning speed of the neural network. The use of the method in the automatic recognition of 7 digital modulation signals obtains better result. The simulation results show that it works well in digital modulation recognition （2ASK,4ASKBPSK, QPSK,2FSK, 4FSK, 16QAM）. The neural network designed by this method is easy constructed, well understood and fast converged, and has small network scale, so it is adaptable to engineering applications.