mike@asgb.UUCP (10/31/84)
In reference to the 8 bit logarithmic representation described by
Richard Schooler (inmet!schooler): he suggested an 8 bit
representation with a sign bit and a seven bit fixed point exponent
with three fraction bits. This gives a dynamic range of
2^(+-(15+7/8)) or 96 dB and "an absolute S/N of 93 dB, and an
instantaneous S/N (precision, roughly) of 32 dB."
This representation does not however, consider waveform fidelity. At
each sampling point, the signal must be quantized to one of the values
represented by the encoding scheme. In the 8 bit scheme, each quantum
level is greater than the previous by a factor of 1.09. A 9% jump
between levels and a maximum quantizing error of 4.5%. This error due
to the lack of precision does not yield a greater S/N, it's DISTORTION.
True, filtering techniques can minimize the distortion, but 4.5% isn't
quite a high fidelity source to begin with.
This encoding does have the interesting characteristic that the
difference between quantizing levels is a constant factor. In the
current CDs 16 bit encoding, the error varies from 50% (!) between 0
and 1 to .00075% between 65534 and 65535. If a 16 bit logarithmic
representation was used, the error would be a constant .024%. I don't
know if this is a significant characteristic, but it at least has a
nice consistency to it.
Mike Rosenlof ...hplabs!sdcrdcf!\
-bmcg!asgb!mike
...allegra!sdcsvax!/
Burroughs Advanced Systems Group Boulder, Colorado