抄録
An efficient procedure based on the asymptotic diagonalization of covariance matrix is proposed to obtain the approximate maximum likelihood estimate (MLE) of parameters of the discrete fractional Gaussian noise. Since this procedure can be executed using the FFT method, it requires an extremely cheaper computational cost than the procedure to acquire the exact MLE which is to some extent efficiently performed using the Levinson algorithm. It is found from numerical experiments that the mean square error (MSE) of the approximate MLE based on the proposed procedure is comparable to the MSE of the exact MLE and nearly attains to the Cramer-Rao lower bound when N (number of data points) ≳10^2.
