Ch (14...) Analyzing Neural Time Series Data

Ch (14...) Analyzing Neural Time Series Data

1. Frequency bands

   • Name: Bandwidth Delta: <4 Theta: 4-7 Alpha: 8-15 Beta: 16-31 Gamma: >32 Mu: 8-12

2. Chapters of Interest

   • LÄS 2,3,7,8,9,10,11,12,13,18,20,21skumma 5, överväg 6

3. Ch 14

   • Hillbert transform
     • Make complex signal from real valued signal.
     • More control over filter than Morlet
       • making the BP, LP, HP, notch filter is easier and its sharpness and width can be chosen freely. FIR or IIR, BP, HP or LP, square or gaussian. FREE CHOICE!
   • Phase angle shows erroneous fft
   • convolution is weighted combination of frequency structure
   • FIR
     • Order
       • high order increase frequency precision, but also calculation time. should be high!
     • Limit
       • Where to start and end the filter. For HP or LP this is just one number. A wide filter is good for temporal precision. A narrow filter is good for frequency precision. Narrow filters also take longer to compute (longer kernels)
     • Shape
       • square, gaussian..... etc. 
   • firls
     • returns filter kernel
       • the kernel is length N+1 where N is the order of the filter kernel.
     • filtfilt
       • uses filter on data
         • input is filter kernel and data and the output is the filtered signal.
       • introduces delay
         • because its a causal filter. Unlike Morlet Wavelet which is not a causal filter.. To remove the delay, run the filter backwards on the filtered data. The the same delay is applied backwards and the time is correct.
     • Butterworth
     • Filter all trials at once!
       • This is better than filtering each trial by itself. Our trials are concatenated as a track by default and: 1. No edge artifacts in trials 2. cleaner code (less for-loops)

4. Ch 15

   • short time fft
   • tapering windows
   • phase angle is not the same as in wavelet