基于同步大地电磁时间序列依赖关系的噪声处理

本文从信号与系统的角度讨论了同步大地电磁时间序列信号之间的依赖关系，选取高信噪比的时间序列信号作为先验数据，用最小二乘法估算依赖关系；结合参考道的数据，合成本地道含噪声时段的数据，最后用合成数据替代噪声段数据，组成新数据，从而在时域中去除大地电磁噪声.西藏地区高信噪比实测数据的试算结果表明，无论电场还是磁场，信号之间的依赖关系是相对稳定的，只与先验数据的长度有关，与时间无关；虽然不同参考点之间的依赖关系不同，但都可以精确合成本地点数据，与参考点地下电性结构和参考距离无关.仿真实验显示，去噪后的信号与原始信号基本一致.实测数据处理结果表明，该方法可以有效去除强噪声干扰，抑制中高频段的近场源效应，同时保留了微弱的有效信号，保证了处理结果的正确性.最后针对方差比方法无法识别的方波噪声，提出了一种简单的平移方法，成功去除了持续时间大于窗口长度的方波噪声；将该方法与远参考技术结合，可以有效抑制近场源噪声干扰，获得光滑连续并且可信的测深资料.

Removal of magnetotelluric noise based on synchronous time series relationship

Magnetotelluric (MT) sounding is one of useful electromagnetic (EM) exploration methods by applying propagation rules of natural EM fields. MT has been proved as an important geophysical method to explore fault structures and fluid movements, and for research of plate tectonics and continental dynamics. However, MT signal is typically non-stationary, weak and stochastic mixed with cultural noises which are intense, of wide bandwidth and complex components. Thus conventional methods for noise suppression in the frequency domain, such as robust estimation and remote reference technique, can only eliminate outliers and uncorrelated EM noise. In the case of strong noise, especially correlated ones, general processing methods are not satisfactory and MT sounding curves show obvious near-field effects. In this paper, based on the concepts of the signal and system, we discuss the relationship of MT time-series between the local and reference sites. We use the least square (LS) method to estimate the relationship by selecting some high SNR data as priori information, then calculate synthetic signal via the relationship and reference data, and replace noisy data of local channels with synthetic signal. The results of the test with high-quality field data acquired in Tibet show that the relationship is relatively stable for MT signal, which is independent of the distance and difference of electrical structures between local and reference sites, and only dependent on the priori data length. The simulation test shows that the synthetic data are considerably correlated with original signals with a coherence above 0.9, and the sounding curves are quite similar. Tests on practical MT data show that this technique can effectively remove local noise and suppress the near-field effects. At the same time, the resulting data keep slight natural signal. Finally, in terms of square wave noise which cannot be recognized by the variance ratio method, we propose a shift method. The result indicates that applying this method in the time domain together with remote reference technique in frequency, one can obtain credible data and smoothly varying curves even in a strongly noisy area. Instead of researching the characterization of complicated noises or separating MT signal from contaminated data, we utilize the clean data of the reference site to calculate synthetic signal.