CAJ | 학술논문

针对多分量调频信号源混合相交时频分布盲分离，提出白化‐均匀加权非正交联合对角化（Whitening‐Uni‐formly Weighted Exhaustive Diagonalization using Gauss iteration ，简称W‐UWEDGE）方法估计混合矩阵。白化对相关信号去冗余处理，无需约束源信号概率密度形式，仅限制源之间整个时频面上无完全重合成分，非正交联合对角化则针对复数域。首先将非正交联合对角化可辨识性从时延平面推广至二次型时频平面，然后利用基于白化处理的梯度范数选择自项时频点（auto‐time frequency point ），进而利用均匀加权近似联合对角化方法估计混合矩阵，分析Amari error值随信噪比及时频矩阵个数的变化规律，与针对混合信号时间历程及时频分布的两类分离方法进行性能比较，显示出所提盲分离方法的优越性。最后应用于转子运行状态识别与齿轮复合故障源分离。仿真与实验数据分析表明所提出方法分离多分量调频相关源的有效性。
침대다분량조빈신호원혼합상교시빈분포맹분리，제출백화‐균균가권비정교연합대각화（Whitening‐Uni‐formly Weighted Exhaustive Diagonalization using Gauss iteration ，간칭W‐UWEDGE）방법고계혼합구진。백화대상관신호거용여처리，무수약속원신호개솔밀도형식，부한제원지간정개시빈면상무완전중합성분，비정교연합대각화칙침대복수역。수선장비정교연합대각화가변식성종시연평면추엄지이차형시빈평면，연후이용기우백화처리적제도범수선택자항시빈점（auto‐time frequency point ），진이이용균균가권근사연합대각화방법고계혼합구진，분석Amari error치수신조비급시빈구진개수적변화규률，여침대혼합신호시간역정급시빈분포적량류분리방법진행성능비교，현시출소제맹분리방법적우월성。최후응용우전자운행상태식별여치륜복합고장원분리。방진여실험수거분석표명소제출방법분리다분량조빈상관원적유효성。
Aimed to mixed multi‐components FM (Frequency Modulation ) sources blind separation problem ,a new method named Whitening‐Uniformly Weighted Exhaustive Diagonalization using Gauss iteration (W‐UWEDGE) is proposed ,whitening procedure is used for redundancy compression ,there are no constraints on statistical properties ,only no overlapping compo‐nents on whole TF plane condition is required ,the nonorthogonal joint diagonalization method is used in complex field .Firstly , identifiability from time‐lag plane to quadratic time‐frequency plane is extended .Secondly ,gradient norm method based on whitening is employed to select autosource (Time‐Frequency ) TF points , then estimated mixed matrix is evaluated by UWEDGE method ,meanwhile the Amari error value is calculated with changes of SNR and number of TF matrixes ,numerical performance of several BSS methods aimed to time‐history and TF distribution was compared ,the results show that the pro‐posed method W‐UWEDGE gains better performance .Lastly ,the method is employed in the rotor operating state identification and composite gear fault source separation .Simulation and experiments show that the proposed method can separate multi component FM sources efficiently .