Time-frequency Peak Filtering With Time-Variant Window Length And Its Application On Noise Attenuation For Seismic Data

Seismic exploration is a method which uses artificial stimulated seismic waveto detect underground geologic structure, and find out the reserve and thedistribution of the oil, coal and natural gas. Among the seismic data collected in thewild, due to the existence of random noise and coherent noise, a number of effectivewaves are drowned in the noise. So improving the SNR of seismic data is theprimary task of seismic data processing. Unlike coherent noise, random noise maybe incoherent in space and time, and is usually caused by a variety of unpredictablefactors. The frequency bandwidth is relatively wide, covering almost the entirefrequency band of effective wave.In2004, Time-frequency peak filtering (TFPF) have been used by B.Boashashto enhance signals. It is an effective algorithm in random noise attenuation. In recentyears, it has been developed and applied to seismic data denoising. TFPF is a signalenhancement algorithm, which is based on the principle of instantaneous frequency(IF) estimation. TFPF is the technique that the noisy signal is encoded as the IF of ananalytic signal by frequency modulated (FM). Then, the signal is recovered bytaking the peak of time-frequency distribution (TFD) of analytic signal. For thenonlinearity of the seismic data, pseudo Wigner-Ville distribution (PWVD) isutilized to execute TFPF, which makes the IF in the window function be linear.The window length of the time-frequency distribution is an important parameterto the TFPF technology. Different frequency signal need different optimal windowlength. The optimal window length could make good performance both in therecovery of the effective signal and the noise attenuation. For the fixed windowlength, it is difficult to achieve the two aspects at the same time. The short windowlength can make the IF in the window approximate linear, but the noise attenuationability in frequency domain will decline. The long window length can attenuate morenoise, but it can’t ensure the IF in the window linear and increases the bias of the IFestimation. Especially at the wave crest and wave trough of the signal, the amplitudechanges greatly and the IF of the analytical signal reached maximum. When it wasdifficult to ensure the IF linear, choosing TFPF with fixed window length can’t achieve ideal results. In this paper, based on the asymptotic formulae of the varianceand bias, we modify the IF estimation based on PWVD and build a new IF estimatorwith a time-variant and data-driven window length. The modified algorithm namedTFPF with time-variant window length which improves the accuracy of the IFestimation.In order to validate the effective and practical of the improved algorithm in thispaper, I do a lot of simulation experiments. Firstly, I apply it to the processing ofdifferent kinds of simulated seismic signal which include single-axis signal,multi-axis signal, multi-frequency signal and fault signal. And then I apply it to theprocessing of real seismic data. The results of experiments show that the modifiedTFPF with time-variant window length provide better performance both in effectivesignal preservation and random noise attenuation.