摘要
从观测序列中分析隐含的周期信号是天文学和地学领域中许多研究工作的重要步骤之一。现在已有一些很好的数学方法能以很高的频率分辨率给出时间序列的谱结构,在统计上也不难对这些谱峰的显著性作出检验。但是,任何谱分析技术所得的周期都只是数学的结果,它们有时并无直接对应的同周期的物理背景。如果忽视这一点,有时会导出不真实的结论。本文从应用的角度试对这个问题进行讨论。
To detect some implicit periodic physical component from a observable series is an important work in the field of astronomy and geosciences. Algorithms for spectral analysis are usually used for this purpose. Some modern algorithms, such as the AR spectral analysis technique, are perfect in frequency resolution. Using these algorithms, the existence of a known periodic physical variation in a series can be easily confirmed. However, it is usually diff'cult to discover some unknown ones because of too many peaks in the power spectra. Sometimes, few peaks are real and most of them are false. Therefore it is important to differentiate the real periods from the false ones. Generally, the reality of the spectral peaks seems to be suspected in following case, and we prefer to reject them. 1. The peaks in low frequency band, whose time scale are the same order with the length of the series, are not believable. These peaks appear due to the unstability of the series. The low frequency fluctuation in the series should be removed before spectral analysis.
2. A group of peaks which densely distribute symmetry to a centric frequency are usually not independent ones physically. They can be resulted due to from a periodic component with obviously varying amplitude.
3. A group of peaks whose frequencies are multiples of basical one are usually also not independent ones physically. They are Fourier harmonics of the basical periodic component whose wave shape is different far from sine-wave.
4. A group of well-distributed peaks in frequency with only small amplitudes are usually not real physically. Generally they are due to noise of the series. It is necessary to select a suitable order for the AR spectral technique in order to weaken noise spectral peaks. It is the most suitable order of the AR model that the mean square error for fitting the series with the group of peaks obtained by this order is close to the proper noise level of the series.
出处
《中国科学院上海天文台年刊》
1991年第12期19-25,共7页
Annals Shanghai Astronomical Observatory Chinese Academy of Sciences
