Subject: Re: [linux-audio-dev] pbd's votes on LADSPA
From: Paul Barton-Davis (pbd_AT_Op.Net)
Date: Mon Mar 27 2000 - 02:57:04 EEST
In message <B503AAA3.A10%asynth_AT_io.com>you write:
>on 3/26/00 11:12 AM, Paul Barton-Davis at pbd_AT_Op.Net wrote:
>
>>The [-1,+1] idea is abhorent to me, if
>> only for the reason that it *guarantees* that we can't satisfy in-bits
>> = out-bits.
>
>?
>It shouldn't. Scaling by 2^16 or 2^24 will only change the exponent, not the
>bits in the mantissa.
Eh ? Who says the scaling factor is 2^16 or 2^24 ? Its close, but not
quite. This is particularly true if you have h/w which generates
24-in-32 sample format, where the "reference dB level" is (2^8)+1, not
zero. This type of format is going to get very common, and is already
used by the RME cards, the ICE1712 (Envy24) and others. I made some
huge errors for a long time trying to use a straightforward approach
toward converting this kind of sample value to a -1,+1 version. Its
much easier converting it to dB, but thats not much use for us.
Do you have some magic that can magically convert a sample value in the
range(s):
-(2^24) ... -((2^8)+1) ... GAP ... (2^8)+1 ... (2^24) - 1
to a scaled value without incredible processing penalties ?
--p
This archive was generated by hypermail 2b28 : Mon Mar 27 2000 - 03:24:59 EEST