CORDIC vs. Matlab results

[Altera_Forum]

Hello, 

I am using Altera CORDIC for ATan2 and SinCos functions. I am applying the same test vectors to the CORDIC ATan2 and SinCos functions testbench and to the ATan2 and SinCos Matlab Models. I see about 10% of samples don’t match between CORDIC functions and Matlab outputs. For both Matlab and CORDIC I use the same S18,17 inputs for ATan2; S11,8 output for ATan2; S11,8 input for SinCos; S10,8 output for SinCos. Any idea why it happens? Thanks, Mike

[Altera_Forum]

cordic gives approximate result and its accuracy depends on number of internal iterations(rotations).

[Altera_Forum]

--- Quote Start ---  cordic gives approximate result and its accuracy depends on number of internal iterations(rotations).  --- Quote End ---   Kaz, Thanks! Could you please recommend application notes or/and user manuals I could learn more how to increase CORDIC accuracy?

[Altera_Forum]

--- Quote Start ---  Kaz, Thanks! Could you please recommend application notes or/and user manuals I could learn more how to increase CORDIC accuracy?  --- Quote End ---    I understand you are using Altera's cordic but can't find any link to its manual.   Xilinx has cordic ip and its manual at:  http://www.xilinx.com/support/documentation/ip_documentation/cordic/v6_0/pg105-cordic.pdf  This should give you some idea about available settings.  there is also a cordic ip at opencores. It is not difficult to write your own targeting specific trigonometric function.   The following pdf provides a good basis:  http://www.ee.ic.ac.uk/pcheung/teaching/ee3_dsd/crdcsrvy.pdf