On Tuesday 08 February 2005 09:23, Lourens Veen wrote: > On Tuesday 08 February 2005 13:02, Lourens Veen wrote: > > Do note that what you're using is not the best possible linear > > approximation. Thinking graphically, you're drawing a straight line > > between the first and last point of the interval, so that you're > > always a little high in the approximation. If you draw that line a > > little lower, then you'll still be high in the middle, but low at > > the ends, and the error is more balanced. The real maximum error is > > thus about half that, so you get one bit more precision. > > Erm, of course not. The interval remains the same, so you don't have > any extra precision at all. It's just that you eliminate the > systematic error.
I thought you were right the first time: reducing the sample points all by a cleverly chosen amount could cut the maximum error due to interpolation in half. However, this component of error is already 64 times smaller than the quantization error (assuming 16 bit samples) so only a handful of sample points would change and it doesn't seem worth the effort. Besides, it might be desirable to have all the errors going in the same direction, i.e., truncate towards zero, it makes it a lot easier to analyze. Regards, Daniel _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
