ABSTRACT
Signal spectral analysis is one of the most important means to examine the characteristics of
signal. Fourier transform can be used to study the quality of the spectrum of the certainty signal.
For general stochastic signal, it is neither a cycle in general, nor in line with the square
integration .Strictly speaking, general stochastic signal cannot be transformed by Fourier
transform. So the power spectrum is generally used for signal spectral analysis.
In the last 30 years Power spectral estimation was rapidly developed. It related to a range of
disciplines such as Signals and systems, stochastic signal analysis, probability and statistics,
stochastic processes and Matrix algebra. And it is widely used in radar, sonar, communications,
geology, exploration, astronomy, biomedical engineering and many other fields.
Actually, the power spectrum of digital signal can only be estimated by finite length data
derived from the limited records, which produced the study area of power spectrum estimation.
Power spectral estimation can be broadly divided into classical power spectral estimation and
modern power spectral estimation. Two main methods of Classical power spectral estimation are
period gram method and auto-correlation method. For the issues such as low resolution and poor
variance performance in Classical spectral estimation, modern spectral estimation is proposed.
Modern Spectral Estimation can be broadly classified into non-parametric spectral estimation
and spectral estimation model. Modeling based on parameter estimation of the power spectrum is
important content of modern power spectral estimation, and its purpose is to improve the
problem of frequency resolution in classical power spectral estimation, which mainly includes
the AR model, MA model, ARMA model. Modern power spectral estimation based on AR
model is the most commonly used methods.
Theoretical analysis and MATLAB simulation results demonstrate that: the power spectrum
approached by the classic spectral estimation has many false peaks, and the frequency resolution
is very low, while the power spectrum approached by the modern spectral estimation methods
to be more true .And in the modern spectral estimation methods there is no significant frequency
deviation and false peak, and have a high frequency resolution, especially the frequency
bandwidth performance significantly improved.
Keywords: power spectrum estimation, AR model, MATLAB, Levinson-Durbin algorithm,
Burg algorithm