中文摘要：

宽频带地震观测数据中有效信号和干扰噪声经常发生混频效应，常规的频率域滤波方法很难将二者分离．地震波信号属于时变非平稳信号，时频分析方法能够同时得到地震波信号随着时间和频率变化的振幅和相位特征，S变换是其中较为高效的时频分析工具之一．本文以S变换为例，提出了基于相位叠加的时频域相位滤波方法．与传统叠加方法相比，相位叠加方法对强振幅不敏感，对波形一致性相当敏感，更加利于有效弱信号信息的检测．时频域相位滤波方法滤除与有效信号不相干的背景噪声，保留了相位一致的有效信号成分，显著提高了信噪比．运用理论合成的远震接收函数数据和实际的宽频带地震观测数据检验结果显示该方法较传统的带通滤波方法相比，即使在信噪较低且混频严重条件下，时频域相位滤波方法的滤波效果依然很明显，有助于识别能量较弱的有效信号．

英文摘要：

In broadband seismology, the signals and ambient noises are usually frequency mixed, so it is hard to extract the signals from the ambient noises by the conventional filter algorithms. Due to the frequency mixing of seismic noise and the non-stationarity of the recorded signal, tools that can effectively take into account the frequency-time variation of the seismic recordings are more suitable for fulfilling this task. The S-transform is an invertible time-frequency spectral localization technique that combines elements of wavelet transforms and short-time Fourier transforms. The time-frequency resolution of S-transform is related to the width of Gaussian window, which has high frequency resolution at low frequency section and high time resolution at high frequency section. The obvious advantage of using the S-transform is that it is convenient not only to denoise but also to directly invert from time-frequency domain to the time domain. Taking the S transform as an example, this paper proposes a new method of phase filter based on phase stack. Phase direction of the coherent signals in the complex plane is consistent, rather that of ambient noises is tanglesome. Compared with the conventional stack method of seismic data, the phase stack of seismic data will get statistical phase of those signals, which is more sensitive to waveform similarity, less to strong amplitude. Phase filter of phase stack in the time-frequency domain easily recognizes ambient noises and also extract coherent signals. So, the method can obviously improve S/N ratio of seismic data. The results of the synthetic benchmark showed that the method performs better than conventional band-pass filter, also very well even if the signal has much lower S/N ratio. Of course, as the increasing of noise levels, the reconstructed signal still has serious noise, especially when the signal and the most energetic part of the noise share the same frequencies and phases. In these cases, the signal after phase filter by the method we presented can estimate r