摘要
针对传统有限脉冲响应(FIR)滤波器设计方法和神经网络设计方法的不足,在改进使用支持向量机(SVM)设计FIR滤波器方法的基础上,提出了SVM设计FIR滤波器的硬件实现方法。使用理想滤波器的幅值响应训练SVM,得到训练参数,据此构建基于SVM的FIR滤波器的嵌入式系统。软件实现FIR滤波器的训练部分,硬件实现FIR滤波器的测试部分。单次判定测试向量的时间约为3 500 ns,滤波准确率可达到98.41%。设计的滤波器具有良好的幅频特性,边界控制精确,逼近理想滤波器。
Aiming at shortcomings of traditional finite impulse response（ FIR） filter design method and neural network design method,a hardware implementation method of FIR filter based on improved support vector machine（ SVM） design method is proposed. Using the amplitude response of the ideal filter to train the SVM,the training parameters are optimized and the embedded system based on the FIR filter of SVM is constructed. Software implements the FIR filter＇s training part and hardware implements the FIR filter＇s testing part. The time of one test is about 3 500 ns,and the filtering accuracy can reach 98. 41 %. The proposed filter has a good amplitudefrequency characteristic; the boundary control is accurate.
作者
计前程
罗小华
JI Qian-cheng, LUO Xiao-hua(School of Electrical Engineering, Zhejiang University, Hangzhou 310007, Chin)
出处
《传感器与微系统》
CSCD
2018年第3期95-98,102,共5页
Transducer and Microsystem Technologies
基金
浙江省自然科学基金资助项目(LY15F040001)
关键词
支持向量机
有限脉冲响应滤波器
核函数
嵌入式系统
support vector machine （ SVM ）
finite impulse response （ FIR ） filter
kernel function
embeddedsystem
