Estimation of the RiIG-distribution parameters using the Artificial Neural Networks

Abstract

In order to improve the estimation of the RiIG (Rician Inverse Gaussian) model parameters, the authors attempt to achieve the parameter estimates using the inverse function of the RiIG CDF (Cumulative Distributed Function) which the latter can not be obtained in a closed form. However, the ANN (Artificial Neural Network) technique is preferred which has the ability to approximate this nonlinear complexity. From recorded sea-clutter data, the regressions of the real CDF are used at the input layer of the ANN estimator. The weights of the network are optimized in real time by means of the genetic algorithm (GA) tool. The mean square error of estimates (MSE) criterion is considered to assess the estimation performance. For almost cases, the experimental results show that adopting the proposed scheme of the ANN estimator turns out the best parameter estimates and also allows a better matching of real CDF and real PDF (Probability density Function) than the standard IMLM (Iterative Maximum Likelihood Method) estimator.