摘要
传感器智能化已经成为一种趋势。为了用现场可编程门阵列(FPGA)芯片与传感器相结合的方法,将人工神经网络应用于传感信号的智能处理,首先要解决应用最广泛的Sigmoid激活函数的FPGA实现问题。据此阐述了以流水线的方式实现激活函数的方法,并论述了这个过程中的2个关键点:一是用协调旋转数字计算机(CORDIC)算法经过多次迭代来逼近指数函数;二是实现了一种改进型的高效除法器,并对仿真结果进行了分析。该方案在实现精度和速度上均满足了神经网络的应用要求,可以应用到传感器智能化中。
The intellectualization of sensors is becoming general trend. In order to apply artificial neutral networks in intelligence processing of sensing signal using combination of FPGA chip and sensors, Sigmoid activated function should be implemented on FPGA. Accordingly, an approach is proposed to implement the Sigmoid function activating method using pipeline mode, and two key points in this process are discussed in detail： first is to approximate exponential function by CORDIC algorithm after several iteration and second is implementation of an improved high-efficient divider,and simulation results are analyzed. To some extent, this scheme is able to meet the requirements of neural networks and can be used in intellectualization of sensors.
出处
《传感器与微系统》
CSCD
北大核心
2014年第1期46-48,共3页
Transducer and Microsystem Technologies
基金
福建省产学科技重大项目(2013H61010023)
