Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
faq:why_does_my_output.freq_not_match_my_cfg.foi_when_using_wavelet_formerly_wltconvol_in_ft_freqanalyis [2013/04/17 16:06]
131.174.207.123 [Why does my output.freq not match my cfg.foi when using 'wavelet' (formerly 'wltconvol') in ft_freqanalyis?]
faq:why_does_my_output.freq_not_match_my_cfg.foi_when_using_wavelet_formerly_wltconvol_in_ft_freqanalyis [2017/08/17 11:21] (current)
Line 2: Line 2:
 ===== Why does my output.freq not match my cfg.foi when using '​wavelet'​ (formerly '​wltconvol'​) in ft_freqanalyis?​ =====  ===== Why does my output.freq not match my cfg.foi when using '​wavelet'​ (formerly '​wltconvol'​) in ft_freqanalyis?​ ===== 
  
-This is a result from the new implementation of '​wavelet',​ it is now using the low-level module '​specest'​. 
  
 Conceptually,​ a wavelet analysis is a time domain convolution of a signal with a set of wavelets, each of these being designed to capture some feature in the data. In neuroscience,​ we typically use Morlet-wavelets,​ which are designed to capture sine and cosine waves in the data. This is because a Morlet wavelet consists of a sine/cosine wave, tapered by a gaussian window. When doing a spectral decomposition,​ the goal typically is to assign the fluctuations in the signal to distinct frequency bands. Importantly,​ the (implicitly) required behavior of the spectral transformation is, that the power estimated at X Hz truly comes from signal fluctuations at X Hz., and not from signal fluctuations at Y Hz. (and Z Hz etc). This is the issue of spectral leakage, and a few signal processing tricks are needed to optimally control for this. Conceptually,​ a wavelet analysis is a time domain convolution of a signal with a set of wavelets, each of these being designed to capture some feature in the data. In neuroscience,​ we typically use Morlet-wavelets,​ which are designed to capture sine and cosine waves in the data. This is because a Morlet wavelet consists of a sine/cosine wave, tapered by a gaussian window. When doing a spectral decomposition,​ the goal typically is to assign the fluctuations in the signal to distinct frequency bands. Importantly,​ the (implicitly) required behavior of the spectral transformation is, that the power estimated at X Hz truly comes from signal fluctuations at X Hz., and not from signal fluctuations at Y Hz. (and Z Hz etc). This is the issue of spectral leakage, and a few signal processing tricks are needed to optimally control for this.