噪声方差未知时的分布式量化估计融合方法

为了解决观观测噪声和信道噪声概率分布不完全已知时的多传感器分布式量化估计融合问题,提出了一种期望极大化算法(EM算法)的分布式量化估计融合方法。该方法将未知的噪声参数以及局部量化器量化概率建模为EM算法中二元高斯混合模型参数,利用极大似然估计方法的估计不变性得到目标参数的估计融合结果。仿真实验结果表明:该方法在局部传感器观测样本数目大于6000和信噪比大于6 dB时与已有理想信道条件下的估计方法性能相当。本文方法对水下分布式协同探测问题提供了一种简化的估计融合实现途径。 

A method for distributed and quantitative estimation fusion of multi-sensor subject to unknown noise variance

In order to solve the problem of distributed and quantitative estimation fusion using multi-sensor of underwater target detection when the probability distribution of the observation noise and the channel noise is not completely known, we make use of the superiority of expectation maximization (EM algorithm) completely in parameter estimation problem when the observation data is missing, based on EM algorithm, an algorithm of distributed and quantitative estimation fusion is proposed. In this method, the unknown parameters of underwater acoustic channel noise and the quantization probability of local quantizer are modeled as the binary Gaussian mixture model parameters, and then, we use the invariance of the maximum likelihood estimation to get the result of the estimation fusion. Simulation results show that: the estimation performance of the algorithm is comparable to the methods which need ideal channel condition when the number of local sensors samples is larger than 6000 and the signal to noise ratio is highter than 6 dB. This method supplies a simplified path of estimation fusion for underwater distributed and cooperated target detection problem.