摘要
经典估计信源个数的Akaike信息论准则(Akaike information criterion,AIC)和最小描述长度(Minimum description length,MDL)方法需要特征分解,运算量较大且需要较多快拍数,为了便于工程实现,在传统的针对数据域Gram-Schmidt(GS)正交投影算法估计信源个数的基础上,提出了一种对协方差矩阵进行GS正交化来估计信源个数新方法,并推导出这种算法的自适应门限。仿真结果表明,该算法和AIC及MDL算法相比,虽然估计性能有所下降,但性能下降幅度不大,且其运算量小、所需快拍数少;与传统GS算法相比,运算量增加不大,但估计性能有较大提高。
Classic source number detection methods, such as Akaike information criterion (AIC) and minimum description length (MDL) need eigenvalue decomposition with more computation and snapshots. For engineering implementation, based on the source number detection method using the traditional Gram-Schmidt (GS) algorithm for estimating original data, a source number detection method using the GS algorithm is proposed, aiming at the covariance matrix and a novel adaptive threshold is obtained. Simulation results show that the method decreases the performance a little, but highly decreases the computation and snapshots compared with AIC & MDL algorithms. Meanwhile, compared with the traditinal GS algorithom aiming at original data, the computation is not increased but the performance is obviously improved.
出处
《数据采集与处理》
CSCD
北大核心
2006年第B12期22-25,共4页
Journal of Data Acquisition and Processing