基于独立分量分析的非均匀采样弱信号频域检测  

Nonuniform Sampling Weak Signal Detection in Frequency Domain Based on ICA

作　　者：徐春生[1] 王太勇[1] 王正英[1] 

机构地区：[1]天津大学机械工程学院,天津300072

出　　处：《天津大学学报》2007年第12期1463-1469,共7页Journal of Tianjin University(Science and Technology)

基　　金：国家自然科学基金资助项目(50475117);天津市科技发展计划资助项目(06YFGZGX18200);高等学校博士学科点专项科研基金资助项目(20060056016)

摘　　要：非均匀采样可以在对模数转化器件的采样频率要求较低的情况下,提取超出Nyquist采样定理限制的频率.当包含两个幅值相差大于10%信号的混合信号经过非均匀采样后,由于采样的伪随机性会产生频谱噪声,因此从幅值谱中辨别不出幅值很小的弱信号.根据非均匀采样的幅值谱识别出大信号频率,构造模拟源信号,以混合信号作为另一个源信号,根据独立分量分析(ICA)对幅值不敏感的特点,使用FastICA算法从混合信号的频谱中成功分离出幅值为强信号幅值五百万分之一的弱信号.在弱信号检测过程中,提出利用互相关系数进行"扫相"处理,以解决大信号相位匹配的问题.在相位匹配存在0.70%误差的情况下,成功分离出幅值为强信号幅值0.06%的弱信号.Nonuniform sampling can acquire signal of which the frequency exceeds the threshold in Nyquist theorem when the sample rate of analogue-to-digital device is lower than usual. When two signals are sampled by nonuniform sampling, and the amplitude of one signal is above 10% smaller than that of the other, the weak signal cannot be detected from the mixed signal＇s spectrum because of sampling＇s fake-randomicity. Because independent component analysis（ICA） was immune to signal＇s amplitude, the weak signal could be detected from the mixed signal＇s spectrum with two input source signals. One input source signal was the mixed signal itself, and the other was a constructed signal of which the frequency was equal to that of the strong signal. The weak signal, of which the amplitude was five million times smaller than the strong one, was separated from the mixed signal＇s spectnma by the FastICA algorithm. A phase-scanning method using cross correlation was adopted to solve the strong signal＇s phase matching. A weak signal, of which the amplitude was 0.06% smaller than that of the strong signal, was separated from the mixed signal, as the relative error of phase matching was 0.70 %.

关 键 词：非均匀采样 独立分量分析(ICA) 弱信号检测 相位匹配 

分 类 号：TN911.7[电子电信—通信与信息系统]