| Magnetotelluric(MT) sounding is a kind of electromagnetic exploration method which uses natural alternating electromagnetic field as a field source. While MT signal is a typical non-linear and non-stationary signal because it is weak, bandwidth and highly vulnerable to be disturbed. The traditional methods of analysis determine the spectra using variations of the Fourier Transform (FT) and must assume that the signals under analysis are stationary over the record length. Because the FT is based on the theory of stationary signal and is ambivalent to the characteristic of MT signal. Therefore, analysis method of MT signal using Fourier transform has obvious flaws. In recent years, the Hilbert-Huang Transform (HHT) has been regarded as a powerful tool for adaptive analysis of non-linear and non-stationary signals and has received much attention in the signal processing community. This paper proposes for the first time the adoption of a new method of analysis for MT data, and focuses on noise suppressing, time-frequency analysis, power spectrum estimation and impedance estimation that are facilitated by applying the HHT.In this paper, starting with the principle of the HHT transform, the empirical mode decomposition (EMD) and the completeness and local orthogonality of EMD are studied. The flow chart of EMD is given. And combined with the characteristic of MT signal, some noduses that exist on the EMD, such as fluctuations, end effects and sifting-stopping criteria, are studied and given the method to resolve the relevant problems. Taking a simulative signal as an example, the basic flow of signal processing based on HHT is given and the superiority of Hilbert-Huang transform to analyze the MT signals is verified.The innovation of the HHT method brings a new way to filter data and reduce noise. The EMD filters signal with different time scales and is adaptive to signal. The intrinsic mode function (IMF) is result of multi-band filtering and can be completely reconstructed. Using the structural characteristics of the IMF, some new space-time filters and a hard (soft) threshold method can be realized to de-noise. This paper studies the characteristics of MT signal and its noise. The principle and step of de-noising method based on EMD are given. Some noises, such as impulse jamming, rectangle disturbing and sine wave noise, are analyzed and processed for the actual MT data. The results show that the de-noising method is effective and the quality of MT data is improved greatly. The EMD method can achieves stationary and reliable parameter estimations.Time-frequency analysis is a powerful tool for non-stationary signal processing. It will express signal as a joint function of time and frequency and reveal the time-varying characteristics of the signal. HHT method has overcome the shortcomings of other ways, completes abolition of the role of the window function and is not restricted with nuclear function and Heisenberg principle. Hilbert-Huang spectrum can precisely describe the nonlinear and non-stationary characteristics of MT signals. This paper compares the different time-frequency spectrums that are from several commonly method such as short time Fourier transform, Wigner-Ville distribution and Wavelet transform. Based on these characteristics, the time-frequency characteristics of HHT spectrum of MT signal are studied. Then Hilbert time-frequency spectrum is used in the MT signal noise recognition, data filtering and sub-smooth. Its application methods and application results are studied in this thesis.Power spectrum estimation takes an important role in MT signal processing. The most traditional and the most important power spectrum estimation method is the classical periodogram method which bases on the Fourier transform. The Fourier transform contradicts the non-linear, non-stationary nature of MT signal, which will inevitably lead to spectral estimation error. A new spectrum estimation method based on Hilbert-Huang transform is proposed in this paper. Firstly the marginal spectrum method based on HHT is presented and the linear property of marginal spectrum is demonstrated. Then, compared with the FT method, the physical signification and the preponderance of marginal spectrum for MT signal are further discussed. And some methods to improve the precision of spectrum estimation are proposed in detail. Finally the simulative data and the practical MT data are analyzed. The simulation and results of real data indicate that the HHT method suits the nature of MT signal much more.The characteristics of apparent resistivity and phase curves are mainly basis for analyzing the structure of the earth and distribute of conduction in MT sounding. Therefore analyzing, calculating and processing of impedance are one of the most basis and importance works. The non-stationary characteristics of MT signal cause that impedance estimation results have time-varying characteristics. Therefore, using time-frequency analysis methold to estimate parameters from the instantaneous spectrum are more conducive to robust estimation comparing with Fourier methold. This paper proposes a kind of instantaneous spectrum method to estimate impedance tensor based on the HHT marginal spectrum. The flow chart and mathematics model are given. Some simulative signal and a place of measured MT signals is processed. The results shows that the impedance tensor obtained from the HHT transient spectrum minimizes the estimation warp brought about by the non-stationary characteristics of MT data, and the estimated parameter is more stable and reliable than that from conventional methods. The parameter curves are less bars and more smooth, reasonable. The interpretability of data has been significantly improved.Through the above studies, the results show that the Hilbert-Huang transform is an effective method to deal with the non-stationary signals and provides a new way to process MT signal. HHT method has a wide application on MT signal noise suppressing, time-frequency analysis, power spectrum estimation and impedance estimation.
|
PDF Full Text Request