[M4IF Technotes] AC prediction

[Gary Sullivan]

Hirakawa-san,
You have hit a personal sore spot.  I and some others tried to persuade
the committee to specify this part of the design to be done differently
(as in H.263 Annex I).  But I suppose we were not sufficiently effective
in our efforts at persuasion.
Anyhow, it works.  And when something in a standard works, you don't
necessarily need a good answer when you ask why it is the way it is.
You just do it because that's how you get interoperability.  Every
design has a few quirks.
Note that Simple profile level 0 is specified in a way that avoids this
issue.
Best Regards,
Gary Sullivan
+> -----Original Message-----
+> From: technotes-admin   lists.m4if.org 
+> [mailto:technotes-admin   lists.m4if.org] On Behalf Of Keigo Hirakawa
+> Sent: Thursday, June 12, 2023 6:42 PM
+> To: technotes   lists.m4if.org
+> Subject: [M4IF Technotes] AC prediction
+> 
+> 
+> Hi.  I have a tedious question about AC prediction.
+> 
+> In section 7.4.3.3 of the MPEG4 spec, a formula is given to 
+> compute the
+> quantized AC coefficients.  For the sake of discussion, 
+> let's restrict our
+> attention to the case when block 'A' is selected:
+> 
+> QF_X[0][i] = PQF_X[0][i]+(QF_A[0][i]*QP_A)//QP_X
+> 
+> In this formula, (QF_A[0][i]*QP_A) is effectively an inverse 
+> quantization
+> step.  However, it differs from the inverse quantization 
+> step outlined in
+> section 7.4.4.2.1 (or section 7.4.4.1.2).  So, in order to 
+> decode the MPEG4
+> binary file, the decoder must perform inverse quantization 
+> step twice, using
+> two different techniques.
+> 
+> My question is this: it seems to make more sense to me if 
+> (QF_A[0][i]*QP_A)
+> in section 7.4.3.3 were replaced with the inverse 
+> quantization step outlined
+> in section 7.4.4.2.1.  That way, the decoder would be 
+> simpler.  In addition,
+> I suspect that the formula in 7.4.4.2.1 is better than doing
+> (QF_A[0][i]*QP_A).  So, why was MPEG4 designed this way?  Is there an
+> advantage to the way it is written in MPEG4 spec?
+> 
+> Thank you!
+> Keigo Hirakawa
+> 
+> 
+> _______________________________________________
+> Technotes mailing list
+> Technotes   lists.m4if.org
+> http://lists.m4if.org/mailman/listinfo/technotes
+>