Noise-space projection: MUSIC without eigenvectors

The MUSIC function for superresolution DF is defined here from an adaptive array point of view and shown to be equal to the square magnitude of the projection of the array steering vector into the noise space of the system covariance matrix. The normal MUSIC algorithm finds this projection in terms of eigenvectors of the matrix. The paper shows that bases for the signal space other than the eigenvector basis can be found and used for the required projection. Some comparisons of the computational loads of the different methods are given, showing a substantial saving by the noneigenvector methods.