-
公开(公告)号:KR100636368B1
公开(公告)日:2006-10-19
申请号:KR1020050107046
申请日:2005-11-09
Applicant: 한국전자통신연구원 , 학교법인 포항공과대학교
IPC: G11B20/10
Abstract: A method for separating a multi-path mixing signal by using relative optimization and an apparatus therefor are provided to estimate a separation coefficient for a mixing signal having a random condition by using a gradient method, a Newton's method, and a trust-region method on the basis of the relative optimization to quickly estimate the separation coefficient having stable convergence. A discrete Fourier transform unit(102) performs the discrete Fourier transform of at least one mixing signal inputted through a multi-path from the outside, and transforms a time domain into a frequency domain. An intersection spectra matrix calculating unit(104) calculates a plurality of intersection power spectra for the mixing signal of the frequency domain transformed in the discrete Fourier transform unit(102), and estimates multi decorrelation. A separation coefficient operating unit(106) estimates a separation coefficient for the mixing signal by using a gradient method, a Newton's method, and a trust-region method on the basis of relative optimization to generate the intersection power spectra calculated in the intersection spectra matrix calculating unit(104) to be decorrelated. An inverse discrete Fourier transform unit(108) performs the inverse discrete Fourier transform of the separation coefficient of the frequency domain estimated in the separation coefficient operating unit(106), and transforms the frequency domain into a time domain. A separating unit(110) filters the mixing signal and separates an original signal by using the separation coefficient of the time domain transformed in the inverse discrete Fourier transform unit(108).
Abstract translation: 提供了一种通过使用相对优化来分离多路混合信号的方法及其装置,用于通过使用梯度法,牛顿法和可信区法来估计具有随机条件的混合信号的分离系数 快速估计具有稳定收敛性的分离系数的相对优化的基础。 离散傅里叶变换单元(102)对从外部通过多路输入的至少一个混合信号执行离散傅立叶变换,并且将时域变换到频域。 交点频谱矩阵计算单元(104)计算离散傅里叶变换单元(102)中变换的频域的混合信号的多个交集功率谱,并估计多重去相关。 分离系数运算单元(106)基于相对最优化通过使用梯度法,牛顿法和可信区法来估计用于混合信号的分离系数,以产生在交叉点谱矩阵中计算出的交点功率谱 计算单元(104)去相关。 离散傅里叶逆变换单元(108)对在分离系数运算单元(106)中估计出的频域的分离系数进行离散傅立叶逆变换,并将该频域变换到时域。 分离单元(110)对混合信号进行滤波并且通过使用在逆离散傅里叶变换单元(108)中变换的时域的分离系数来分离原始信号。