提出一种RBF神经网络算法应用于线性混叠信号的盲分离。所用的RBF神经网络算法是从输入信号的数据中训练出中心值和宽度值,再训练通过用最大熵值的代价函数推导的权值。所用的代价函数保证了网络的输出尽可能独立,使信号能正确地分离。仿真验证了所用的算法能减少分离时间和提高分离效率。对比ME算法,该算法更好。 A RBF neural network algorithm is proposed for blind separation of linear aliasing signals. The RBF neural network algorithm used is to train the center value and the width value from the input signal data and then to train the weights that are derived by the cost function using the maximum entropy value. The cost function used ensures that the output of the network is as independent as possible so that the signal can be correctly separated. Simulation shows that the algorithm used can reduce the separation time and improve the separation efficiency. Compared with ME algorithm, the algorithm is better.