This paper addresses the problem of direction-of-arrival (DOA) estimation of correlated and coherent signals, and two sparsityinducing methods are proposed. In the first method named L1-EVD, the signal-subspace eigenvectors are represented jointly with wellchosen hard thresholds attached to the representation residue of each eigenvector. Then only the eigenvector corresponding to the largest eigenvalue is reserved for DOA estimation via sparse representation, which aims at highly correlated signals, and a method named L1-TEVD (TEVD: Truncated EVD) is proposed. Simulation results show that L1-EVD and L1-TEVD both surpass L1-SVD in DOA estimation performance and computation efficiency for highly correlated and coherent signals.