Piecewise算法计算超越函数及其FPGA实现
首发时间:2010-03-01
摘要:本文使用分段逼近(piecewise approximation)算法计算超越函数,以神经网络中最常用的双曲正切型(tanh)传输函数为例来分析逼近精度同分段数、有限字长之间的关系,并给出如何最优的选取分段数和字长来达到所需的精度。本文还在FPGA上硬件实现了基于piecewise算法来计算 tanh函数,并且比较了几种piecewise approximation算法以及其它常见硬件算法(如CORDIC,Look-up table)的资源消耗和速度。
关键词: piecewise;tanh;FPGA
For information in English, please click here
Transcendental function calculated by piecewise algorithm and its FPGA implementation
Abstract:Transcendental function is calculated by piecewise approximation algorithm in this paper. We analysed relation between approaching precise with segments and limited word length in the example of most commonly used transferring function “tanh” in neural networks. The method of choosing segments and word length for the needed precise is given in this paper. We also implemented calculating “tanh” based on piecewise algorithm in FPGA. And piecewise approximation algorithm is compared with others fitting for hardware, like CORDIC and Look-up table, in the aspect of resource consumed and speed.
论文图表:
引用
No.4027851762612674****
同行评议
共计0人参与
勘误表
Piecewise算法计算超越函数及其FPGA实现
评论
全部评论0/1000