Abstract: The co-centered orthogonal loop and dipole (COLD) sensor is one of the most widely used two-component electromagnetic vector sensors. However, the COLD sensor does not make full use of the two component spatial aperture. The uniform linear array, which consists of COLD sensors, is considered. In order to extend the spatial aperture, all loop-components maintain the same, and all dipole-components are sparsely uniformly stretched along the orthogonal direction, thus an L-shaped array is formed. A generalized rotational invariance based dimension reduction multiple signal classification method (GRIDR-MUSIC) is proposed. The proposed algorithm uses the L-shaped array geometry to separate the steering vector into three parts, and uses the generalized rotational invariance matrix to estimate each part separately, thus direction of arrival (DOA) and polarization parameters can be estimated by only one-dimensional spectral peak search. In addition, a dipole is added in the conference point, and the quaternion model is used to resolve the phase cyclic ambiguity caused by sparsely stretching operation. Finally, the numerical simula-tions show the effectiveness of the proposed algorithm.