RBF网络高斯型传输函数的FPGA实现

FPGA Implementation of Gauss function for RBF Neural Network

本文在研究CORDIC算法的基础上,在FPGA上实现了RBF网络的Gauss型传输函数,分析了该实现方法在输入为[-70]时的误差,并得出实现该函数的CORDIC迭代所需资源仅和同样字长的乘法器相当。当把CORDIC算法实现的传输函数用于解决XOR问题的RBF网络时,该网络也表现出良好的性能。

Gauss transfer function for RBF Neural Network was implemented with FPGA based on CORDIC algorithm. The error of CORDIC algorithm with input of [-7 0] was analyzed, and the FPGA resources used by CORDIC iteration was approximately equivalent to that of a single multiplier with the same word length. When the transfer function based on CORDIC algorithm was used to solve a RBF Neural Network of the classic XOR problem, the Network had admirable performance too.