Email: Password: Remember Me | Create Account (Free)

Back to Subject List

Old thread has been locked -- no new posts accepted in this thread
???
10/25/07 23:57
Read: times


 		 #146231 - Adding FP's accurately
Responding to: ???'s previous message
Christoph Franck said:


Do you know the algorithm to "accurately" add more than two floating point numbers ?


Hi Christoph,

My apologies, I misunderstood your question, to keep the FP errors to a minimum, I would try and keep the FP number excursions as small as possible, but I'm guessing here.
To be honest, I don't recall having any accuracy problems using FP but I agree they do exist.
My compiler unfortunately does not support transcandential math functions like sqrt,e^x,ln,sin,cos,..etc, so I wrote my own using the Pade algorithm and Newtons root finder. I tried using Taylors algorithm and also cordics but found the latter two to be less accurate.

Cheers for now.
Jason






List of 64 messages in thread
TopicAuthorDate
ADC last digit stability            01/01/70 00:00      
   DVM            01/01/70 00:00      
      DVM's only 10 bits?            01/01/70 00:00      
         Frequency response            01/01/70 00:00      
            Typically twice or more the sampling frequency            01/01/70 00:00      
               Missing the point            01/01/70 00:00      
   You usually don\'t _want_ a stable last digit.            01/01/70 00:00      
      Result            01/01/70 00:00      
         Hm.            01/01/70 00:00      
      Averaging/DSP/FIR filters            01/01/70 00:00      
         Math lecture ?            01/01/70 00:00      
            Can't use average-too slow            01/01/70 00:00      
               Analysis            01/01/70 00:00      
               already stated            01/01/70 00:00      
               Derivation            01/01/70 00:00      
                  The nitty gritty stuff            01/01/70 00:00      
                     Simulate using a spreadsheet            01/01/70 00:00      
                        Hi Russ and thanks for that            01/01/70 00:00      
                           My spreadsheet vs. your spreadsheet            01/01/70 00:00      
                              Unit test !            01/01/70 00:00      
                     As I said...            01/01/70 00:00      
                        Calibration?            01/01/70 00:00      
                           Horsefeathers            01/01/70 00:00      
                              Longs?            01/01/70 00:00      
                                 A long is more "accurate" than a float.            01/01/70 00:00      
                                    Yes it is, but....            01/01/70 00:00      
                                       _Known_ transfer functions.            01/01/70 00:00      
                                          FP pitfalls            01/01/70 00:00      
                                             The compiler isn't going to take care of that.            01/01/70 00:00      
                                                Mantissa fields            01/01/70 00:00      
                                                   Misunderstandings ?            01/01/70 00:00      
                                                      Adding FP's accurately            01/01/70 00:00      
                                 can you not read ?            01/01/70 00:00      
                           Nonsense ...            01/01/70 00:00      
                              Calculator?            01/01/70 00:00      
                     If you're really curious ...            01/01/70 00:00      
                        Climax time            01/01/70 00:00      
                           OOPS!!....            01/01/70 00:00      
                           Equations            01/01/70 00:00      
                              Quantisation            01/01/70 00:00      
                                 Works in unsigned integer, too:            01/01/70 00:00      
                                    Abstract?            01/01/70 00:00      
                                       Useful abstractions.            01/01/70 00:00      
                                          Nice talking with you, Christoph.            01/01/70 00:00      
                           What are the actual requirements ?            01/01/70 00:00      
                              re filters            01/01/70 00:00      
                              And moving forward            01/01/70 00:00      
               You can not due to Dither            01/01/70 00:00      
         Clueless ?            01/01/70 00:00      
            Just dump the LSB            01/01/70 00:00      
               Don't forget proper rounding.            01/01/70 00:00      
            Speaking of DSP            01/01/70 00:00      
         Fluke stability            01/01/70 00:00      
   Oh yes they do!            01/01/70 00:00      
   my take on this            01/01/70 00:00      
      Walk, don't run            01/01/70 00:00      
         :)            01/01/70 00:00      
      Hello Erik            01/01/70 00:00      
         have fun, Jason            01/01/70 00:00      
            I am having fun            01/01/70 00:00      
               Ok, you are aware of this            01/01/70 00:00      
                  He's learning            01/01/70 00:00      
                  Assumptions            01/01/70 00:00      
                     and I ...            01/01/70 00:00      

Back to Subject List