The algorithm processes very few snapshot data opposed to forming sample covariance matrix and computing matrix inversion.

• 它无需构造协方差矩阵及进行矩阵求逆，又避免了复杂度较高的广义特征值分解。