[M4IF Technotes] question on the accuracy requirement for FDCT fixed point impleme ntation

[Chen, Kenny]

Hi,
As to IDCT, there is a IEEE paper defining the accuracy very clearly.  But
how about FDCT, as the quantization is right after FDCT, the accuracy
criteria must not be too tight.  But is there any discussion or reference
that will define some threshold such as absolute difference or else?  
Rgds,
Kenny