一种随机相位估计简化EM算法

期望最大化（EM）算法在处理随机相位估计时是一个NP-完全问题,目前主要采用梯度算法来对其求解。但该方法存在计算量大、不易稳定且对相邻时刻估计结果依赖严重等问题。基于随机相位模型EM算法的因子图表示,提出了一种简化EM算法,其思想是只针对当前时刻进行独立的EM迭代计算,然后通过相邻相位偏转之间的关系对结果进行修正。仿真实验说明,该方法在减小计算量的同时,提高了算法性能。

A Simplified EM Algorithm for Random-walk Phase Estimation

Since the implementation of EM algorithm turns into a NP-complete problem in random-walk phase estimation,the gradient method is now exploited for its solution,which,however,is huge in calculation,prone to instability,and seriously dependent on the estimates of adjacent time slots.Based on the model factor graph,a simplified EM algorithm is proposed,with the idea to conduct the current EM iteration first and adjust the final estimate by the interrelations of the adjacent phases thereafter.Simulation shows that the proposed method could achieve both calculation reduction and performance improvement.