组合径向基函数神经网络的硬件实现  

Hardware implementation of a combined radial basis function neural network

作　　者：梁艳[1] 靳东明[1] 

机构地区：[1]清华大学微电子学研究所,北京100084

出　　处：《清华大学学报（自然科学版）》2009年第10期1692-1695,共4页Journal of Tsinghua University(Science and Technology)

摘　　要：为了提高径向基函数(RBF)神经网络的泛化能力,提出了一种组合径向基函数神经网络,并通过对英文字母的识别进行了仿真验证。基于CMOS电路设计了该组合径向基函数神经网络,所有单元电路均采用HJTC 0.18μm CMOS数模混合工艺设计制造。通过PCB板实现了一个2×3的组合RBF神经网络,并对"一"和"1"的识别问题进行了验证。实验结果表明:该组合RBF神经网络电路结构简单,便于扩展和调节,提高了整个网络的泛化能力,为硬件实现更为复杂的组合径向基函数神经网络提供了可能。A combined radial basis function （RBF） neural network was developed to improve the generalization ability of RBF networks, specifically for character recognition. The combined RBF network was implemented in CMOS analog circuits designed and fabricated in HJTC 0. 18 μm CMOS technology. A 2 × 3 combined RBF network was implemented and used for recognition of the Chinese character and ＂1＂. Tests results show that the circuits for the combined RBF network recognize the characteristics of simple structures with a high operation capacity. They provide an initial basis for hardware implementation of more complex combined RBF network.

关 键 词：CMOS模拟电路 组合径向基函数神经网络 字母 识别 硬件实现 

分 类 号：TN492[电子电信—微电子学与固体电子学]