MATLAB: How does MATLAB handle sample size which is not power of two in discrete wavelet transformation

Question

I'd like to know how MATLAB R2012a handles time series which length is not power of two in discrete wavelet transformation (DWT). As I know, in DWT, the sample size N must equal to 2^J (J=decomposition level). But when I load the series with length N=365, and decompose to 5 levels, there were results coming out. So, what is the theory or method used behind?

For your information, I'm using Wavelet Toolbox with graphical interface('wavemenu'), one-dimensional wavelet. Thanks in advance for anyone's help.

Answer 1

MATLAB extends the signal or image based on the DWTMODE setting at the command line. In the UI, only one extension mode is used and that is the default symmetric extension mode

dwtmode('sym')

See the help for DWTMODE for more information. You can also play around with wextend() to see specifically how this extension works.