Subject: [linux-audio-dev] [-1,+1], float, 24 bit etc.
From: Paul Barton-Davis (pbd_AT_Op.Net)
Date: Mon Mar 27 2000 - 07:41:12 EEST
OK, I duly accept james' point: all signed 24 bit integer values can
be converted to -1,+1 without loss.
I think my concern/confusion has arisen because with both the RME
hardware and the Creamware Pulsar/Scope, we are presented with what
either is, or is most easily handled as, 32 bit data, and this cannot
be handled in the same way as 24 bit stuff. For the RME, one could
choose to do a slightly tricky conversion to 24 bit data by doing a
sign-extended bit shift into the low 3 bytes. But as I've said for
some time, given that (at least) Analog Devices already makes 32 bit
converters, and the record of the industry, it strikes me as less than
3 years away until we can routinely buy 32 bit converters.
OTOH, using 32 bit floats for 32 bit ints won't work anyway, without
loss of precision, so the [-1,+1] range is a bit of a red herring here
anyway.
Makes me more supportive of the argument in favor of the basic data
type being a double (64 bit float).
--p
This archive was generated by hypermail 2b28 : Mon Mar 27 2000 - 08:20:51 EEST