>
> >The question is, what kind are the quantizations lame makes at
> >differend sampling rates?
> >If ISO/IEC 13818-3 "table B.1 Possible quantization per subband, Layer II"
> can give a hint,
> >the largest value is 16383 = 2^14 -1. A milk maid calculation gives
> > 10*log10(2^(2*14)) = 84.3 db
> >Whereas the largest value in ISO/IEC 11172-3 "3-B.2 Layer II bit
> allocation tables"
> >is 65535 = 2^16-1. Above calculation gives
> > 10*log10(2^(2*16)) = 96.3 db
> >
> >So can we conclude, that for MPEG2 layer 2 the dynamic range is 12 db lower
> >than in MPEG1 layer 2, and likewise in layer 3?
>
> The quantization, at least for layer 3 is non-linear. Is the formula above
> correct for non-linear quantization?
>
> /jp (just trying to understand)
>
For the non-linear quantization in layer 3, it is s*x^(3/4), quantized
between 0 ... 8205, where x is the MDCT coefficient and s is a
scalefactor. So the possible quantized values are not a good
indication of the dynamic range. I think the raw pcm input is the
best way to estimate the dynamic range: layer3 works with 16bit singed
integers, or 65535 differnt values. Your ("back of the envelope", as
we'd call it over here) calculution for 65536 = 96.3db which is nice
agreement with the oft quoted 96db dynamic range for a CD.
Mark
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
