L_p模反褶积的一种快速算法

A FAST ALGORITHM FOR L_P NORM DECONVOLUTION

中国科学院计算中心

文献[1,2]提出了一种值得人们进一步探讨、发展的反褶积方法——L_p模反褶积方法.文献[1]限制1≤p≤2,文献[2]指出对0<p<1也是有价值的.在最优预报距离n_b能先验估计时,文献[1,2]指出的加权最小二乘迭代方法(WLSI法)计算速度是快的,但当最优n_b不能先验估计时,WLSI法就显得慢了.对此情形,本文将[1,2]中加权最小二乘迭代法改进为矩阵求逆迭代(MII)法,使得主工作波因子长度n_f而言降低了半个数量级,而且算法更加紧凑,这样就使得L_p模反褶积更具有实用性,方法对p<∞的任意值是适用的,但实际应用中仍限制p∈[1,2].人工地震道实例模拟表明,该法是有效的.

In REFERENCE [1,2], an approach of deconvotion, LP NORM De-convolution is presented, which is deemed worthwhile going further into discussion as well as development. The REFERENCE [1] defines 1≤p ≤2 and REFERENCE [2] indicates that it's of use for 0<p<1 too. When the optimum predicted distance nb can be predeterminedly estimated, the calculation using WLSI was considered efficient as remarked in REFERENCE [1,2]. But when the optimum nb failed to be predeterminedly estimated thereupon the WLSI seems not fast any more supposedly. In view of this case, an attempt is made in the paper, i.e. the WLSI specified in REFERENCE [1,2] has been updated to MII so as to decrease the major workload into 1/2 In quantity in terms of the filtering factor length m. Moreover, this way of calculation turns out even more compact. so it is the reason why Lp NORM Deconvolution is so challenging; The recommended approach can suit the arbitrary value p<∞, however it is restricted to p∈[1,2] still in application. The modelling of synthetic trace example shows that it is a methed tried-and-true indeed.