Abstract: This paper proposes a tri-iteration algorithm (TIA) to estimate the two-dimensional(2-D) frequency embedded in additive Gaussian white noise. By exploiting the rotational invariance property of the 2-D data matrix, we established four matrices which possess a diagonal structure. Moreover, we expanded the data matrix in the temporal domain and derived a set of diagonal structure matrices. Consequently the 2-D frequency are estimated by accomplishing the joint diagonalization of the group of data matrices. It is worth mentioning that the estimated 2-D frequency can be paired automatically. The proposed algorithm eliminates the error propagation of the multistage decomposition algorithm because each iteration poses a typical least square problem with a unique closed solution. Hence the estimation accuracy is increased. A comparison of the performance of TIA to that of CRB demonstrates its effectiveness.

Key words: frequency estimation, tri-iteration algorithm(TIA), joint diagonalization, array signal processing