摘要
为了改变人工神经网络的研究仅仅局限于算法,只是在通用的串行或并行计算机上模拟实现的现状,针对函数逼近问题,将BP神经网络的结构分为3个模块,采用VHDL语言完成对各个模块的硬件描述,并使用Altera公司的QuartusII 6.1综合软件进仿真和调试,然后在CycloneII系列FPGA上实现了能够进行片上学习并完成函数逼近的BP神经网络系统。测试结果证明,该系统能够很好地完成在线学习,并能满足一般系统应用的速度和精度的要求,验证了该方法的有效性。
In order to change the situation that the research on the ANN only in algorithm and basing on serial or parallel general computer simulation, this paper focused on the function approximation and divided the BP neural network structure into three modules, it used VHDL to complete the description of hardware and the Altera company＇s QuartusII 6.1 to simulate and debug,then implemented a BP neural network in the CycloneII series devices of FPGA produced by Ahera company,this network coule completed the on-chip learning and the approximation of function. The test results show that the system can learn on-line and meet the general system applications＇ speed and accuracy.The system evaluates this method is effective
出处
《电子设计工程》
2010年第9期151-154,共4页
Electronic Design Engineering
基金
中国高技术研究发展计划863(基金资助项目2008AA121803)
关键词
神经网络
硬件实现
FPGA
片上学习
函数逼近
neural network
hardware implementation
FPGA
on-chip learning
functional approximation
