Can anybody tell me how convergence rounding(fixed point ) works in dsp, with an example. Thanks.
convergence rounding fixed point
Started by ●April 16, 2007
Reply by ●April 16, 20072007-04-16
On 16 Apr., 07:38, "bhargavall...@gmail.com" <bhargavall...@gmail.com> wrote:> Can anybody tell me how convergence rounding(fixed point ) works in > dsp, with an example.Once you tell us how convergence rounding (fixed point) works in general, somebody here might come up with an idea how it works in dsp. Regards, Andor
Reply by ●April 16, 20072007-04-16
Andor wrote:> On 16 Apr., 07:38, "bhargavall...@gmail.com" <bhargavall...@gmail.com> > wrote: >> Can anybody tell me how convergence rounding(fixed point ) works in >> dsp, with an example. > > Once you tell us how convergence rounding (fixed point) works in > general, somebody here might come up with an idea how it works in dsp.Andor: There's evidently a language problem. bhargavalluri: You need to be clearer about what you want to do. "Convergence rounding" doesn't mean anything to me or anyone else I know. Rounding is usually applied to numbers that would otherwise be truncated. For some purposes, a technique that some call "fraction saving" gives better results. Is that what you mean? Jerry -- Engineering is the art of making what you want from things you can get. ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Reply by ●April 16, 20072007-04-16
On Apr 16, 3:37 pm, Jerry Avins <j...@ieee.org> wrote:> bhargavalluri: You need to be clearer about what you want to do. > "Convergence rounding" doesn't mean anything to me or anyone else I know.Jerry, I found the term "Convergent rounding" used in connection with IT++, a C++ library that supports fixed point computation: http://itpp.sourceforge.net/stable/group__fixtypes.html#fix_base_qmode It seems like that they got the term from SystemC. Cheers, Guenter
Reply by ●April 16, 20072007-04-16
Guenter wrote:> On Apr 16, 3:37 pm, Jerry Avins <j...@ieee.org> wrote: > >> bhargavalluri: You need to be clearer about what you want to do. >> "Convergence rounding" doesn't mean anything to me or anyone else I know. > > Jerry, > > I found the term "Convergent rounding" used in connection with IT++, a > C++ library that supports fixed point computation: > > http://itpp.sourceforge.net/stable/group__fixtypes.html#fix_base_qmode > > It seems like that they got the term from SystemC.Well, Guenter, now I know at least one person who understands the term. :-) Would you be good enough to explain it to bhargavalluri and me? Jerry -- Engineering is the art of making what you want from things you can get. ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Reply by ●April 16, 20072007-04-16
On 16 Apr., 16:48, Jerry Avins <j...@ieee.org> wrote:> Guenter wrote: > > On Apr 16, 3:37 pm, Jerry Avins <j...@ieee.org> wrote: > > >> bhargavalluri: You need to be clearer about what you want to do. > >> "Convergence rounding" doesn't mean anything to me or anyone else I know. > > > Jerry, > > > I found the term "Convergent rounding" used in connection with IT++, a > > C++ library that supports fixed point computation: > > >http://itpp.sourceforge.net/stable/group__fixtypes.html#fix_base_qmode > > > It seems like that they got the term from SystemC. > > Well, Guenter, now I know at least one person who understands the term. > :-) Would you be good enough to explain it to bhargavalluri and me?But Jerry, it says it right there in that link: RND_CONV (Convergent rounding with half-way value rounded to even value): If the most significant deleted bit is 1, and either the least significant of the remaining bits or at least one other deleted bit is 1, add 1 to the remaining bits. RND_CONV_ODD (Convergent rounding with half-way value rounded to odd value): If the most significant deleted bit is 1, and either the least significant of the remaining bits is 0 or at least one other deleted bit is 1, add 1 to the remaining bits (not defined in SystemC). :-)
Reply by ●April 16, 20072007-04-16
Andor wrote:> On 16 Apr., 16:48, Jerry Avins <j...@ieee.org> wrote: >> Guenter wrote: >>> On Apr 16, 3:37 pm, Jerry Avins <j...@ieee.org> wrote: >>>> bhargavalluri: You need to be clearer about what you want to do. >>>> "Convergence rounding" doesn't mean anything to me or anyone else I know. >>> Jerry, >>> I found the term "Convergent rounding" used in connection with IT++, a >>> C++ library that supports fixed point computation: >>> http://itpp.sourceforge.net/stable/group__fixtypes.html#fix_base_qmode >>> It seems like that they got the term from SystemC. >> Well, Guenter, now I know at least one person who understands the term. >> :-) Would you be good enough to explain it to bhargavalluri and me? > > But Jerry, it says it right there in that link: > > RND_CONV (Convergent rounding with half-way value rounded to even > value): If the most significant deleted bit is 1, and either the least > significant of the remaining bits or at least one other deleted bit is > 1, add 1 to the remaining bits. > RND_CONV_ODD (Convergent rounding with half-way value rounded to odd > value): If the most significant deleted bit is 1, and either the least > significant of the remaining bits is 0 or at least one other deleted > bit is 1, add 1 to the remaining bits (not defined in SystemC).I thought Guenter might as well have told us since he was at the site. Is "convergent rounding" the same as "convergence rounding"? I suppose so, but it's not assured. Jerry -- Engineering is the art of making what you want from things you can get. ¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯¯
Reply by ●April 16, 20072007-04-16
On Apr 16, 5:40 am, "Andor" <andor.bari...@gmail.com> wrote:> On 16 Apr., 07:38, "bhargavall...@gmail.com" <bhargavall...@gmail.com> > wrote: > > > Can anybody tell me how convergence rounding(fixed point ) works in > > dsp, with an example. > > Once you tell us how convergence rounding (fixed point) works in > general, somebody here might come up with an idea how it works in dsp. > > Regards, > AndorHi all, Thanks for ur replies. I am trying to use a A/D convertor in one of my applications. This A/D convertor gives out digital data in straight binary format. Now I have to perform some operations on it(some multiplications for example). I have to round off the results of my computation. I read somewhere that convergence rounding (rounding to nearest integer) is the best approach. I know how it works out for floating point dsp numbers but dont know how to do it for fixed point numbers? I hope i am a bit clear now. Thanks.
Reply by ●April 16, 20072007-04-16
On Apr 16, 4:48 pm, Jerry Avins <j...@ieee.org> wrote:> > Well, Guenter, now I know at least one person who understands the term. > :-) Would you be good enough to explain it to bhargavalluri and me? >Actually, those two options of convergent rounding used in IT++ are also know as "round to even" and "round to odd". There is a better- than-I-can-do explanation about rounding that covers "round to even" on the wikipedia page: http://en.wikipedia.org/wiki/Rounding Cheers, Guenter
Reply by ●April 16, 20072007-04-16
On Apr 16, 8:11 pm, "bhargavall...@gmail.com" <bhargavall...@gmail.com> wrote: [...]> computation. I read somewhere that convergence rounding (rounding to > nearest integer) is the best approach.When I googled for "convergence rounding" I got 19 results. When I googled for "convergent rounding" I got 750 results. Under the later term it says that it is an unbiased rounding, which is not rounding to the nearest integer. Maybe this link will help: http://en.wikipedia.org/wiki/Rounding Cheers, Guenter
