i would like to do trend analysis for long time series data using multi resolution wavelet analysis. i am struck with preparing input data to analyze using discrete wavelet transform. i have tried two ways to do wavelet analysis but i dont know which one is correct.
1. i have transformed my time series data in to logarithmic scale of difference
2. i have not changed anything in the original data .. i just used as it is .
Hi, Manikandan sathianarayanan.
The most common practice is to transform a signal as is (i.e. in the linear scale). Depending on the application, you may apply some filters.
The logarithmic scale is commonly applied to the transformed signal because it allows to detect peaks which are more than the order of magnitude lower than the main peak.
/Edit: Some applications employ the Hilbert transform before the Fourier / wavelet transform (see "analytic signal") and other also utilize the logarithmic transform before the inverse Fourier transform (see "cepstrum"). But I believe that these are not related to your intended application.
Thanks Luboš Smolík.. i would like to apply wavelet transform on precipitation time series data
In the case of 1D- DWT, the input should be in the form of an array, for example (1*1024)
- Badges
- Science topic