基于KBANN的调制识别仿真研究  

Simulation of Modulation Recognition Algorithm Based on Knowledge-Based Artificial Neural Network

作　　者：王晓斌[1] 石昭祥[1] 

机构地区：[1]解放军电子工程学院602教研室,安徽合肥230037

出　　处：《计算机仿真》2009年第3期152-155,159,共5页Computer Simulation

摘　　要：BP网络广泛应用于多信号调制样式识别,但普通BP网络存在隐层数目难以确定、收敛速度慢、容易陷入局部最小等缺点。为了克服上述缺点,仿真研究了一种基于知识人工神经网络(KBANN)的信号调制样式识别算法。首先将C4.5算法引入信号特征参数的阈值分割,根据输出的决策树构造出具有决策树特征的拓扑结构,然后使用共轭梯度学习算法提高BP网络的收敛性能。仿真结果表明,与普通BP网络相比,基于知识神经网络的识别算法网络的结构易于实现、能有效改善网络收敛,并提高低信噪比下的正确识别率,为利用神经网络进行调制识别提供了新的思路。A knowledge - based artificial neural network for modulation automatic recognition is presented in this paper. BP network has some common shortcomings, such as falling into the local minimum, designing topology hardly, and converging slowly. A general effective way to enhance the network performance is to build a Knowledge - Based Artificial Neural Network by embedding the prior knowledge into the network. Simulation utilizes the decision tree produced by CA. 5 to construct the topology and initialize the weight of network unit. Combining with conjugated grads learning algorithm, the result shows that the knowledge - based artificial neural network has a better performance in solving the problem of slow convergence. And the accuracy is also improved.

关 键 词：反向传播网络 知识神经网络 决策树 调制识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]