Forums

A better window than hanning

Started by thedude December 20, 2004
Hi

Im working with DSP for a school project.

We have made a sound compression program for wav files.

We uses a hanning window and makes 50% overlap on our sound samples,
before we use out fft.

Is there a window that uses less overlap with the same result?
thedude wrote:

> Im working with DSP for a school project. > > We have made a sound compression program for wav files. > > We uses a hanning window and makes 50% overlap on our sound samples, > before we use out fft. > > Is there a window that uses less overlap with the same result?
So the FFT does compression? What effect does the window have on the compression? Why use a window at all? You'll need to explain a little more clearly what you're doing and what you hope to achieve. Ciao, Peter K.
"thedude" <gonzoz66@hotmail.com> wrote in message 
news:5f5a61f6.0412200611.3da80e5d@posting.google.com...
> Im working with DSP for a school project. > > We have made a sound compression program for wav files. > > We uses a hanning window and makes 50% overlap on our sound samples, > before we use out fft. > > Is there a window that uses less overlap with the same result?
I assume that you're annoyed at having twice as many output samples as input samples. Instead of messing with the windows, you should change the transform. Google for information about the "modified discrete cosine transform", or "lapped orthogonal transforms" in general. -- Matt
"thedude" <gonzoz66@hotmail.com> wrote in message 
news:5f5a61f6.0412200611.3da80e5d@posting.google.com...
> Hi > > Im working with DSP for a school project. > > We have made a sound compression program for wav files. > > We uses a hanning window and makes 50% overlap on our sound samples, > before we use out fft. > > Is there a window that uses less overlap with the same result?
I don't think that overlap and the window selection are related. But, someone somewhere may have shown that it matters the tiniest bit. For what you're doing I don't see how it can matter. You can choose whatever overlap you like - including none. And, you can use a von Hann / hanning window or some other window quite independent of the overlap. It all depends on the algorithms you're using for compression. Fred
In article <wL-dnQUXsZfUhVrcRVn-vA@centurytel.net>, "Fred Marshall" <fmarshallx@remove_the_x.acm.org> wrote:
>
> >I don't think that overlap and the window selection are related. But, >someone somewhere may have shown that it matters the tiniest bit. For what >you're doing I don't see how it can matter. >
It does if you go back to the time domain. A Hann window applied in the time or frequency domain and fifty percent overlap will not have artifacts when converted to the time domain. Fred Harris (fred harris) has a paper on this subject from the 70s. Kaiser-Bessel and 75 percent overlap have the same properties. The only window that has this property and less overlap than 50% is the boxcar window. At least as far as I know.
On Sat, 25 Dec 2004 15:19:37 GMT, george.w.bush@whitehouse.com
(George Bush) wrote:

>>It does if you go back to the time domain. A Hann window >>applied in the time or frequency domain and fifty percent >>overlap will not have artifacts when converted to the time >>domain. Fred Harris (fred harris) has a paper on this >>subject from the 70s. Kaiser-Bessel and 75 percent overlap >>have the same properties.
I was unaware that Kaiser-Bessel had this property. (For which value of Harris' alpha parameter is this true?) Windows that sum-to-constant include: Triangular, with 50% overlap Cosine-Squared (Raised-Cosine; Hann), with 50% overlap Cosine-Fourth, with 75% overlap de la Vall&#2013265929;-Poussin, with 75% overlap
>>The only window that has this property and less overlap than >>50% is the boxcar window. At least as far as I know.
The only other named window that I know of that has the sum-to-constant property with less than 50% overlap is the Tukey. However, one could construct other windows in a manner similar to the Tukey, that also sum-to-constant. Greg Berchin
"George Bush" <george.w.bush@whitehouse.com> wrote in message
news:dyfzd.36958$Ew6.27578@twister.socal.rr.com...
> In article <wL-dnQUXsZfUhVrcRVn-vA@centurytel.net>, "Fred Marshall"
<fmarshallx@remove_the_x.acm.org> wrote:
> > > > >
> Fred Harris (fred harris) has a paper on this > subject from the 70s.
He surely does. See the following citation: ======================== Harris, Fredric J, "On the Use of Windows for Harmonic Analysis with the Discrete Fourier Transform", Proceedings of the IEEE, Vol. 66 No. 1, January 1978 pp 51-83 ======================== On the subject of inverse-transforming to the time-domain, I'm under the impression that any non-rectangular windowing will pretty well destroy the possibility of getting back the original time-domain data. When you IDFT the spectrum, you'll get back to the windowed time-domain data, not the raw time-domain data. Also, if you plan to overlap the time domain samples under a rectangular window, I believe you have to adjust the phase relationships in the frequency domain before you add frequency components. For example, with 50% overlap all the odd sine and cosine components of the overlap piece will have to be negated with respect to the previous piece in order to preserve proper phase relationships.