Abstract: Projection approximation subspace tracking with deflation (PASTd) algorithm belonging to eigen-decomposition algorithm was widely used in adaptive beam forming for antenna array. However, it has been found that this algorithm can only work satisfactorily when the signal to noise level is low. In order to solve this problem, fast orthonormal PAST (OPAST)algorithm was introduced into PASTd algorithm. The modified algorithm can estimate eigenvector and eigenvalue of useful signals when the dimension of useful signals is unknown. The modified algorithm can insure the orthogonality of the eigenvectors, and the modified algorithm’s astringency is better than PASTd algorithm while the computation complexity almost remains the same. The modified PASTd algorithm combined with multiple signal classification (MUSIC) algorithm applied in large-scale adaptive arrays can restrain the interfaces in the main beam and vicinities of the main beam. The combined algorithm significantly deduces the computation cost of MUSIC and it is also very practical to form deep nulling at interfere direction.