一种激励函数的硬件仿真实现

Hardware Simulation Implementation of Active Function

现在对神经网络的研究以神经网络模型及算法的模拟实现为主,较少考虑硬件实现问题。以模拟器件为主,分析设计了BP神经网络各环节的硬件电路。介绍其中的一种Sigmoid激励函数电路实现,该电路以差分器件为主要部分,通过调整相应的参数可以调节输入电压的范围和改变激励函数的增益,并在EDA环境下仿真验证了电路的有效性。

With the development of information technology, artificial neural network is used widely in many fields. But most researchers give priority to neural network algorithms software implementation and seldom considers hardware implementation. The paper designs the hardware circuits of Backward Propagation（BP） neural network based on analog circuits. The circuit of active function is introduced. The differential structure is mostly used in it. The range of input voltage and function gain can be adjusted by varying correlated parameters. And its effectiveness is demonstrated by EDA simulation.