Subspace approach to direction finding of rapidly moving sources with an uniform linear array

Whenever the bearings of sources are considerably time-varying, methods like MUSIC or ESPRIT for direction-of-arrival estimation are not applicable, as no clear separation between signal and noise subspace is possible. The bearings could be estimated either from single snapshots, or in order to achieve higher integration gain, and if K parameters are sufficient to parametrize the bearing-trajectories of the P sources in time, one could try to solve the KP-dimensional maximum-likelihood parameter estimation problem. To overcome this computational burden, a subspace method is presented which requires only a K-dimensional optimization. For each snapshot the noise subspace of certain subarray-outputs is computed. Then, all noise subspaces enter jointly a K-dimensional MUSIC-type cost-function. In comparison with single-snapshot methods the new approach shows better performance for low signal-to-noise ratio and can easily handle crossing bearings.