基于改进MLE-NR方法的杂波Weibull分布模型参数估计

Parameter estimation for clutter Weibull-distributed model based on improved MLE-NR method

为了解决传统最大似然-牛顿拉夫森(MLE-NR)方法需要谨慎地选择初始值保证迭代过程收敛的问题,该文提出了一种针对杂波韦布尔(Weibull)分布模型的参数估计方法。首先计算迭代点处的海森(Hessian)矩阵,然后根据海森矩阵的值,不断调整迭代过程中的发散点或错误的初始迭代点,使发散的迭代过程重新收敛,从而正确地估计模型参数。针对长度为256、512、1 024、2 048、4 096的随机样本数据,分别进行了N=500的蒙特卡洛(Monte-Carlo)仿真,仿真结果证明了该文方法的收敛性。蒙特卡洛仿真结果和基于实测样本的处理结果说明了该文方法的有效性和鲁棒性。

To solve the problem of the traditional maximum likelihood estimation-Newton Raphson （ MLE-NR ） method that the initial value must be selected carefully to ensure the convergence of iteration,an improved MLE-NR method for parameter estimation of clutter Weibull-distributed model is presented here. The Hessian matrix of the iteration point is calculated, the divergent points in iteration process and wrong initial iteration points are adjusted to a convergence region according to the determinant value of the Hessian matrix,so that the divergent iteration is convergent again and the model parameter can be estimated correctly. Monte-Carlo simulations are proceeded with N=500 and the lengths of the random sample data are 256,512,1 024,2 048,4 096 respectively. The results show that the improved MLE-NR method is convergent for sample data with different length. The Monte-Carlo simulation and processing results based on measured data demonstrate the effectiveness and robustness of this method.