On One Numerical Experiment Involving Calculation of the Spectra of Random Processes

For two models of a stationary random process $\xi_k$, $k=0,\pm 1,\dots$ with a spectral density $f(\lambda)$ that is known in advance the author constructs, using a sample of fixed size, estimates for $f(\lambda)$ that are based on the use of nonnegative and sign-changing weighting functions. Comparison of the estimates thus constructed indicates that, for a specified sample size, estimates based on the use of sign-changing weighting functions result in smaller estimation error in both cases.