Skip to Main Content
This paper addresses a novel approach based on neuro-fuzzy inference system to solve the estimation problem of the K-distributed parameters. The method is based on a network implementation with real weights and the real genetic algorithm (GA) tool is applied for an off-line training of the fuzzy-neural network (FNN) shape parameter estimator. The proposed FNN estimator is based on the arithmetic and geometric means computed from the data in a manner which significantly reduces the computational requirements when compared to Raghavanpsilas and the maximum likelihood methods. The simulation results are presented to demonstrate the validity of the approach as well as the successfulness of the FNN estimator for low variances of parameter estimates when compared with existing MOFM (method of fractional moments) and ML/MOM (maximum-likelihood and method of moments) approaches. In addition, the method yields parameter estimates with lower computational complexity compared with standard techniques.