Hi, I want to implement a Kalman-filter on a DSP56F8323. To do that I need to compute a matrix (9x9). Using Gauss-Jordan algorithmen I need to divide a lot of figures. In my eyes there is a problem using fixed point arithmetics. Its not allowed to divide a smaller figure through a bigger one because the figure range is -1 to 0.999 and it would result in bigger than 1. I am DSP newby so it would be nice when somebody could help me. Best regards, Maik Schormann This message was sent using the Comp.DSP web interface on www.DSPRelated.com
inverse matrix
Started by ●August 31, 2005
Reply by ●August 31, 20052005-08-31
mswu wrote:> Hi, > > I want to implement a Kalman-filter on a DSP56F8323. To do that I need to > compute a matrix (9x9). Using Gauss-Jordan algorithmen I need to divide a > lot of figures. In my eyes there is a problem using fixed point > arithmetics. Its not allowed to divide a smaller figure through a bigger > one because the figure range is -1 to 0.999 and it would result in bigger > than 1. I am DSP newby so it would be nice when somebody could help me. > > Best regards, > Maik Schormann > > > > This message was sent using the Comp.DSP web interface on > www.DSPRelated.comI think your better off using a Cholesky update variation of the Kalman filter.
Reply by ●September 1, 20052005-09-01
>I think your better off using a Cholesky update variation of the Kalman >filter. >Hi, I think I cant use Cholesky because my matrix is not symetric. Maik This message was sent using the Comp.DSP web interface on www.DSPRelated.com
Reply by ●September 12, 20052005-09-12
"mswu" <maik.schormann@partner.bmw.de> wrote in message news:Ms2dnbQEyPvXOojeRVn-2Q@giganews.com...> Hi, > > I want to implement a Kalman-filter on a DSP56F8323. To do that I need to > compute a matrix (9x9). Using Gauss-Jordan algorithmen I need to divide a > lot of figures. In my eyes there is a problem using fixed point > arithmetics. Its not allowed to divide a smaller figure through a bigger > one because the figure range is -1 to 0.999 and it would result in bigger > than 1. I am DSP newby so it would be nice when somebody could help me. > > Best regards, > Maik Schormann > > >As I remember all the inverseing that is done in KFs is to symmetric matrices. eg The innovations covariance matrix will be symmetric. [R+HPH']^-1 Shytot