基于K-L变换的网络调制识别方法

针对BP网络进行通信信号调制样式识别的场合,在输入层和输出层节点数确定的条件下,提出在网络训练完成后应用K-L变换方法确定网络隐含层的节点数,对网络结构进行优化。利用K-L变换的正交性,去除了网络结构的冗余,给出了该算法的理论推导和具体应用步骤,通过仿真验证了算法的有效性和顽健性。利用BPSK、QPSK、16QAM等数字调制信号,选取了三个特征量,分别用优化前后的网络进行调制识别,实验表明优化前后网络的识别性能基本一致。

Modulation identification based on neural network optimized by K-L transform

How to determine the topology of BP network is an important problem for the practical application of modulation identification based on neural network.A method called K-L transform is presented to determine the number of nodes in hidden layer with the network training having been finished,provided that the number of the nodes in the input layer and the output layer are fixed.Based on the orthogonality of K-L transform,the redundancy of the network is removed.Both the theory inference and the application steps of the proposed method are presented.The validity and robustness are verified by simulation.The signals including BPSK,QPSK,16QAM et al are used for modulation identification with three features chosen.The performance of the network is almost the same between pre and post the network optimizing.