Emu XL-7 & MP-7 User's Group

on 1/13/04 8:34 PM, Aaron Eppolito wrote:

> The entire signal path of the XL-7 is 24 bit.

OK. Does the CS engine (or modern Proteii engine
in general) use integer DSP with some larger
accumulator for multiplies and adds?

> Envision the 16 bit samples LSB justified in
> 24 bit space.  The high 8 bits are headroom.

but why would it be done this way?

I ask because regardless of the bit depth used,
0dBFS should be the same signal peak voltage on the
analogue output...24-bit would just give you the
ability to resolve dynamic range below -96dBFS...

> When you add 128 of these samples, you get a
> 23 bit number.  When you add in gain of up to
> +10dB, you can get 25 bits.  When you add in
> resonance you can get even more bits.  (This
> is why the filters can clip, by the way)

OK, now I understand why it is be done that way
(LSB justified).

> At no point do you lose the 16 bit precision of
> the original samples. Just because they're down
> at the LSB side instead of the MSB side doesn't
> mean that it's in any way less quality.

in the engine, this is true...but as it hits the
D/A converter, it is no longer the case, as the
chipset used in these machines (along with the
analog path) will not resolve a number of bits on
the LSB side, because they are below the noise
floor of the system.

> That being said, if you record digitally at 16 bit,
> yes, you will lose important data.  You should
> record at 24 bit then normalize.  This
> gives the best of all worlds (maximum headroom with
> the full original sample precision).

OK, understood...but still, it would be nice to have
a param by which we could control the gain scaling
on the voice level, so as it have the best control
of that digital out.

> ...a headroom control would still be useful for
> those using the S/PDIF without benefit of a
> normalizer.

oops, just said that! :)

Aaron, thanks so much for the explanation!


cheers,
aeon