Talk:Sample rate conversion

From Wikipedia, the free encyclopedia

There is no mention of the different algorithms (linear interpolation, sine, etc..) or dithering.

OK, I added this. Incidentally, you can (should) sign your comments with four tildes. It will be converted to a user name, date and time, like this: LouScheffer 15:46, 16 June 2006 (UTC)

This is nearly useless as a starting point for understanding how sample rate conversion works, if you don't know the basic terms that get thrown around in the article and are never properly explained (try reading the "example" to see what I mean). I don't know what can be done about this, maybe some links or a careful restructuring might help? Lord Lizard 15:21, 7 November 2006 (UTC)

[edit] Clarification: Movie to PAL/NTSC

[Quote>] "[...] conversion of movies (shot at 24 frames per second) to television (50 or 60 frames per second).

Actually, PAL and NTSC standards show video at rates of 25 and 30 fps (respectively); these rates come from the different AC rates used by countries all over the world (eg 50 Hz for Europe, 60 Hz for US). Two current cycles are used to draw a single frame (odd fields first, even fields then). There is no sample rate conversion from movie to PAL, resulting in a slight time and pitch stretch (1 second shorter every 24 seconds of movie, that means that a 80min movie is 3min 20sec shorter in PAL). For NTSC, 5 frames are created every 4 frames of film (see Telecine/Common pulldown patterns).

(This is only a draft; since I'm not a specialist, some parts should be verified or rewritten) - Lion-hearted II 09:51, 1 August 2007 (UTC)

OK, I added this. It's a little more complex than you stated, since there are several alternatives that preserve running length and pitch. LouScheffer 17:30, 1 August 2007 (UTC)

[edit] Polyphase decomposition

To Louscheffer Polyphase decomposition is really different from not calculating the unused samples by the decimating filter at the end of the converter. Actually, the interpolation and the decimation actions can be interchanged using this method, whereby the filtering is run at significantly lower a sample rate. It is therefore incorrect to merge the two ideas. WimdeValk 23:11, 26 October 2007 (UTC)

You are right that a polyphase filter also take advantage of the fact that most of the inputs are 0. But when this is included, that's exactly what polyphase filtering is. In the example shown, you need only compute 1/147 of the output samples. For each tap, only one in 160 has non-zero value. So you really have 160 short filters, each with 1/160 of the coefficients. Which one you use depends on the phase, hence polyphase. LouScheffer 02:03, 27 October 2007 (UTC)

[edit] Incorrect use of the term "interpolate"?

I think the term "interpolate" is used incorrectly in this article. Generally (and in its own WikiPedia article) it means finding some values of a function that are missing from the available samples of that function. It seems to be used in this article to mean just padding with zeros. This has to be confusing to a somewhat knowledgeable reader as the whole process of sample rate conversion is also known as interpolation when the output rate is an integer multiple of the input rate. I think where the term is used with the meaning of padding with zero samples it should be clearly stated that way. If I remember, I'll come back here after a spell and if there is no objection I will do this. --GrahamDavies 23:26, 6 November 2007 (UTC)