Hello!

I'm trying to do a real-time effect with the undecimated wavelet transform (á trous algorithm). I programmed faster matlab functions which give the same results as the swt/iswt commands of the matlab wavelet toolbox. It uses periodic extensions for non-expansive convolution. When I manipulate the analysis data before reconstruction I get clicks at the block boundaries ( I tried it with 4096-samples steps). I am using b-spline wavelets which (i think) are symmetric. Can I use symmetric signal extensions instead of periodic extensions to avoid these errors at the boundaries? or do i have to do some kind of windowing, possibly overlapped, to fix it?

Is it the same problem as described here for the FFT?

http://www.hydrogenaudio.org/forums/index.php?showtopic=7687

How would this symmetric extension convolution work?

i'd be thankful for any help

thanx

I don't think you can avoid the need for windowing. It's essentially the same problem, yes.

thanks a lot!

I kept the periodic extension and tried it with a Hann window (50% overlap) and it seems to work fine. I don't get any clicks now. Of course, computation time increased, but I think I just won't bother with other convolution techniques if these results sound alright. If I have the time I will try to implement the dual tree complex wavelet transform as a faster alternative (with less good results I suppose).

tom