Fast Polynomial Time Frequency Transform And Its Application On Random Noise Attenuation For Seismic Data

In seismic prospecting process, useful signal are often corrupted by random noise. When these noises'energy is oversize, it will influence the subsequent dynamic correction velocity analysis and the authenticity and reliability of final seismic imaging, then bring adverse impact on explaining work. To some extent, this will lead to wrong actual operation such as sinking wells, drilling, thus greatly increase the geological exploration costs. So it's significant to remove the noise of seismic signal, improve the SNR and resolution.Fast polynomial time-frequency transform (FPTFT) is a new kind of polynomial phase signal parameters estimation method, it bases on the nonlinear least-squares criterion (NLS), can make the output signal and ideal signal approximate optimally. It can recover information of the effective wave by using the signal parameters estimated from noise background. This transform retains the traditional maximum likelihood estimation method's advantage such as applicable to multi-component signal, good statistical property under low SNR, meanwhile, it has discrete polynomial phase transform and High ambiguity function's characteristics of less computation, and it can get all the parameters at one time, won't cause error transfer effect. Because the seismic signals and polynomial phase signals have common characteristics such as instantaneous frequency, instantaneous phase etc. In this paper, FPTFT is used to reconstruct signals from reflection seismic data corrupted by random noise for the first time.Based on the characteristics of seismic signals received by geophones, we get the mathematical model of artificial synthesis records. Regard Ricker wavelet with additive white gaussian noise as second order polynomial phase signal, then use the fast polynomial time-frequency transform to deal with seismic data contain a single and more phase axis respectively, estimate the phase parameters , then structure phase, and combine the input data using nonlinear least-square approach to recover the signal. According to the simulation results, the records of recovering can show original records with phase axis position very well, can also suppress the random noise in Ricker wavelet. Based on the wavelet form before and after processing, compare spectrum charts, and compute the SNR of before and after processing, all of above verify this method can be used for the elimination of random noise in seismic data.