改善高分辨IMP算法收敛速度的一种新方法

A New Method for Improving Convergence Procedure of IMP Algorithm

提出了一种加速高分辨ＩＭＰ算法收敛过程的新方法。在存在有多个信号源且信号源个数已知或可估计得到的情况下，该方法可显著减少ＩＭＰ算法的收敛时间，从而为高分辨算法的实时实现研究提供了新的可能性。

Real-time implementation of high resolution algorithms is difficult, especially in coherent environment' Starting from my previous work with the IMP (Incremental MultiParameter) algorithm['.s], 1 now propose a new method to speed up its convergence procedure.The original IMP algorithm estimates the source directions iteratively. With the addition of each new source, iterations begin anew, resulting in greater number of iterations.Wax et al[2] and Viberg et al[3] offer ways of determining number of sources. Thus my new method can start on the basis of known number of sources and thus save iterations as compared with original IMP.We take a 3-source case. Fis. 1 shows that in a particular simulation trial the IMP requires 23 iterations and Fig. 2 shows that in the same trial our modified IMP requires 14 iterations. Eqs. (4) and (5) are made use of in obtaining Fig. 2.For the same 3-source case, the results of 100 simulation trials presented in Fis. 3 indicate that arbitary assignment of initial estimates does not affect convergence.Again for the same 3-source case, Fig. 4 presents the results of 100 simulation trails for both our modaled IMP and for the orisinai IMP algorithm. These 100 trials indicate that original IMP needs most probably 20 iterations and our modified IMP algorithm requires most probably only 14 iterations, or only about 70% as compared with orisinal IMP, which means a saing of 30%.For a 4-source case, our work shows a saving of about 40%. As the number of sources increases, a further increase in saving of iterations prcentage-wise may be expected.