Talk:Window function

From Wikipedia, the free encyclopedia

Contents

[edit] Comparison Table

this article is in dire need of a big comparison table that summarizes the key metrics of the various window functions. the comparison table should be similar to table 1 in the seminal article by harris "on the use of windows", page 5. see article, which is linked in the references.

[edit] Scalloping loss

The paragraph about scalloping loss seems to refer to a figure that isn't there. Sdrpyh 06:23, 14 February 2007 (UTC)

It's just as Bob K wrote it on 18 December 2005, this diff. The figure above is still the same "zoom view" figure in the previous subsection. I find it a bit confusing, too. Dicklyon 06:30, 14 February 2007 (UTC)

[edit] All about frequency analysis?

In this article, it sounds like windowing's only use is in frequency analysis, as for example I am using it for making a windowed-sync FIR filter, and in this case the use of a window function (i chose the Blackman one) is to avoid riple in the frequency response of the FIR filter kernel. --SuperBleda 15:27, 19 August 2005 (UTC)

I still don't know, what is a window function? --anon


Whoever redid this, it's much much better now. 81.105.21.22 13:59, 7 April 2006 (UTC)

I can't claim all the credit, but I did a lot of it. So on behalf of the whole team... thanks! --Bob K 07:18, 10 April 2006 (UTC)


[edit] DFT-even vs. Matlab

I'm a bit puzzled as to why the window functions given were not put in "DFT-even" form ( /N rather than /(N-1)). The reason to use "DFT-even" windows is covered in one of the articles's references. See pg 52 of the following:

Frederic J. Harris, 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.

--SAMerritt 10 May 2006

I had the same question when I first arrived at this article. What I subsequently discovered is that Matlab defines the windows the way it's done here, not the "DFT-even" way. That seemed like a good enough reason to leave it as is. But in spite of that, I was the one who added the classic F.J.Harris reference, for historical reasons, if nothing else. --Bob K 04:03, 10 May 2006 (UTC)

[edit] The Comparison Graph

The comparison graph is great. The logarithmic frequency axis (in units of "DFT bins") is an excellent idea. It would also be helpful to label the axes. --Bob K 00:17, 10 September 2006 (UTC)

SUGGESTION: Please add a similar time-domain comparison of step-response, and maybe a table of statistics from that graph such as %-overshoot, group-delay, settling-time (to 1%, 0.5%, 0.1%). ALSO maybe add a step-response to each of the graphs in the "Window examples" section above. -- citizenDAK —The preceding unsigned comment was added by 70.91.225.169 (talk • contribs).

What does that mean? What is the step response of a window function? Dicklyon 22:29, 3 July 2007 (UTC)
Step response is (in summary) a way to see + compare important characteristics like the TRANSITION-DELAY or SETTLING-TIME and the RINGING or OVERSHOOT of a filter (a Window function *IS* a kind of FIR digital filter, commonly called a "moving average"). (Try a search-engine for "step response DSP window" for more info... One good example I just found is slide 10 of this Intro to Digital Filters PDF-file.) -- citizenDAK
Yes, I know about filters and step responses; but in what sense is a window usefully viewed as a filter? It's not convolved with an input, so it's not being used as a moving-average filter. If it were, its step response would have no overshoot, since windows are non-negative. They would all look very similar, sort of like the windows themselves. Maybe you can compute one of what you have in mind and show us. Dicklyon 19:23, 9 July 2007 (UTC)

[edit] confusing wording

Can someone change this? In the third sentence of the following excerpt, it is unclear what "they" refers to (does it refer to window functions with low dynamic range, or window functions with poor resolution?): "This characteristic is sometimes described as low dynamic range. At the other extreme of dynamic range are the windows with the poorest resolution. And they are also poorest in terms of sensitivity." Lavaka 06:08, 8 January 2007 (UTC)

Yes, very. I believe I have fixed it correctly, but I invite others to check me. Dicklyon 06:30, 8 January 2007 (UTC)

[edit] Bartlett/Triangle Window

Can someone check the equation for the triangle window? I've normally seen it as 1-abs((n-.5N)/.5N), or equivalently 1-abs((2n-N)/N). This page shows 1-abs((2n-N+1)/N)

Kelauder 06:05, 2 April 2007 (UTC)Kevin

It's easy enough to verify by inspection that at the defined end indices of 0 and N-1 the values of the Barlett are 0 and of the triangle are 1/N; that is, the formulas work as expected. Your formula would work if the range was defined from 0 to N for Bartlett, or 1 to N-1 for triangle. But it's not. Dicklyon 06:35, 2 April 2007 (UTC)

[edit] Gauss window

In the provided plots, I was suprised to see that the Gauss window in the Fourier doimain does seem to have sidelobes. This should not be the case though as the FT of a Gauss function is again Gauss. —The preceding unsigned comment was added by 129.27.140.171 (talk) 08:33, 17 April 2007 (UTC).

Read (or re-read) the part about spectral leakage. A Gauss function has infinite duration. A Gauss window does not.
--Bob K 11:24, 17 April 2007 (UTC)

[edit] Tighter format?

What do you think of my "tighter format" edit? I can do the same on the next section, but it might run into problems for narrow browser windows with its longer equations. Any other good suggestions for how to format this without wasting tons of white space? I also like the tight way because it's so much easier to look at several equations at once to compare them. Dicklyon 20:03, 3 June 2007 (UTC)

The tighter format is also what I chose (e.g. http://en.wikipedia.org/w/index.php?title=Window_function&direction=prev&oldid=31779771).
--Bob K 21:30, 3 June 2007 (UTC)

OK, I'll do it. I see the "left" format was contributed by an anon. Dicklyon 23:20, 3 June 2007 (UTC)

[edit] PNG and SVG plots

mmm... how can you say that "The .png files have better resolution?, no bit-map picture can have higher resolution than a vector image! I have noticed that unfortunately I have lost the grid in the SVG generation and I couldn't fix it, but the grid is not necessary in this context. Alessio Damato 21:07, 26 October 2007 (UTC)


[edit] Request for revision

I think that this article is lacking in having a specification for how a window is applied (which would hopefully include an explicit equation). Also a non-technical intro about what windowing is used for, and why. Melcombe (talk) 08:53, 4 April 2008 (UTC)

[edit] Questions

Is there a terminology describing calculating two windows and using the best features of each?

I will be writing a "dual window FFT/IFFT" (basicly two seperate functions ran in sequence on the same data) that will calculate and use both "Low-resolution (high-dynamic-range) windows" and "High-resolution windows" (to use the terminology of this page) in order to provide (the FFT portion) a spectrum with high dynamic range and high resolution, and (the IFFT portion) a reproduction of the input wave with a low loss of residuals.

I've tried searching for any programs or papers describing (what would seem to be) this simple idea but there seems to be no information on this subject.

In the event that there is such a thing a description of it is missing from this page. —Preceding unsigned comment added by 24.85.104.61 (talk) 23:27, 2 June 2008 (UTC)

I'm not sure I would put it here anyway. For instance the Welch method uses window functions, but it is described in a separate article.
Also not sure why you need to "reproduce the input wave". Don't you already have it? And what does that have to do with this article?
--Bob K (talk) 02:45, 3 June 2008 (UTC)