采用数模混合电路实现前馈神经网络的一种方法

Method of Implementing Feed forward Neural Network Using Digital and Analog Circuits

本文提出了一种用数模混合电路实现小规模前馈神经网络的方案，并用电子线路实现了每层具有１０个节点的３层前馈网，连接权采用固定电阻实现，节点非线性函数用数字电路实现。该前馈网经ＢＰ算法离线训练之后，可以广泛用于各种智能控制。文中给出了权值、阈值以及非线性单元的实现方法，并给出了用该方案实现的神经网络控制器实例及测试方法。

A small scale architecture of feed forward neural network using digital and analog circuits is described. With ten nodes in each layer,a three layer feed forward neural network is implemented. Fixed resistances are used as weights and node nonlinear function is implemented with digital circuits. Having trained with BP algorithm off-line,such a neural network can be widely used in intelligent control. Implementation of weights,threshold and nonlinear function are discussed. An example of application and verification for the neural network hardware are also illustrated.