In massive multiple input multiple output systems, the nonlinear channel estimation method has been initially proved to completely eliminate pilot contamination, which is based on spectrum separation theory to remove interference through projecting it onto the subspace of the desired signal. However, in view of the existing system parameters design for ensuring exact spectrum separation being not accurate, a hierarchical system parameter design method is proposed in this letter. It clarifies the different functions of the different system parameters, so that the asymptotic spectrum of the covariance matrix of the received signal could be more accurately separated. Furthermore, according to the proposed numerical algorithm, in the presence of pilot interference, a better tradeoff between channel estimation quality and pilot overhead is given. It uses pilots as few as possible to obtain the most of the channel estimation quality gain, which is brought by projecting onto the subspace of the separable asymptotic spectrum described above. All of these make our approach more practical.