Archived

This topic is now archived and is closed to further replies.

Crispy

FFT round-off error

Recommended Posts

I''ve been eyeing the round-off error for some time now and, while the algorithm is notorious for its lack of round-off error, the numbers I''m getting are a bit worrying. Here''s some statistics: for data length = 1024 the round-off error stays between 1.0e-9 and 1.0e-7, which is okay. However, as the size increases, so does the error, which, at data length = 1024 reaches 1.0e-2 (~0.02). Is this normal? It doesn''t seem to be affected by whether I carry out the FFT in single or double precision (granted, I only tried converting the input data to double precision before the bit reversal and casting it back to single precision after doing the transform (in double precision)). Any comments?

TCP - Transmission by Carrier Pigeons

Share this post


Link to post
Share on other sites