Sample depths

wingfeather · Sep 10, 2003, 03:13 PM

When a PCM stream (for example, when I play a CD or an mp3) is sent through the DSP (of an Audigy2 OEM card), is it automatically converted up to 24 bits? If not, how can this be accomplished? I ask because I use the Parametric EQ from the ufx pack quite heavily to increase my low bass at around 12dB/octave so that I get a flat response from my subwoofer (I also remove a pair of room modes with it). Once I get down to about 20Hz I have 15dB of gain - this means that to avoid clipping I've had to set the overall gain to -6dB, and if the sample depth is only 16-bits this digital attenuation could be a real loss of information.
Do I have to choose between a flat response and overall detail (I really hate the boomy 80Hz that I get without EQ) or can I upsample the streams to 24-bits prior to EQing, and hence keep the resolution?

I saw the 16to32 effect in the Passthru category but can't quite figure out what it does. The same goes for a few of the other plugins in kX (call me thick!). Could ambiguous or little-known plugins have a comment added into the microcode briefly explaining their function?

TIA

**Eugene Gavrilov** · Sep 11, 2003, 12:00 AM

the 16-bit audio is converted into 32bit internal representation when it gets inside the DSP
it is then processed with mixed 24/32bit accuracy (depending on the effects you are using)
the output resolution for audigy2 is 24 bit (that is, the 24 bit DSP data is passed to 24-bit DACs)

anyway, there might be better ways of achieving your goals

/E

wingfeather · Sep 11, 2003, 12:18 AM

So the precision is fine for what I'm doing, which is nice. Thanks for the clarification.
What are the better ways? Analogue EQs?

Max M. · Sep 11, 2003, 02:38 AM

>it is then processed with mixed 24/32bit accuracy

correction: DSP processing itself is done with 32/64bit accuracy

wingfeather · Sep 11, 2003, 03:26 AM

Hell, 64 bits? I'm impressed with that. That's some pretty accurate processing.

**Eugene Gavrilov** · Sep 12, 2003, 03:38 PM

well, Max, if we wish to be 100% correct, then it is 24(16)/64

(24(16) is external tram resolution, 32 is the DSP GPR resolution and 64 is DSP accumulator resolution)

anyway, '32bit processing' seems to be the 'golden mean'

/E

Max M. · Sep 12, 2003, 05:30 PM

Ok, ok... Let's do not fall into deep discussions here...
If we are speaking about equalizers (as this thread is inventionally of) then their precision is 32/64 (ufx eq and Soeren's filters use accumullator _very_ hardly and do not use TRAM)...

**Eugene Gavrilov** · Sep 12, 2003, 11:18 PM

ok

I doubt we could have fallen into a deeper discussion, since all the precision discussion stuff is now present in this thread

/E

mattcro · Sep 25, 2003, 09:00 PM

OK, if we're going to be really picky, the 10k1 (and presumably 10k2) have a 67-bit accumulator in the FX processor and does 32-bit integer/fixed-point processing... from my reading of the EMU FX8010 architecture document

Matt

Sep 10, 2003, 03:13 PM	#1
wingfeather DriverHeaven Newbie Join Date: Jul 2003 Location: UK Posts: 18 Rep Power: 0	Sample depths When a PCM stream (for example, when I play a CD or an mp3) is sent through the DSP (of an Audigy2 OEM card), is it automatically converted up to 24 bits? If not, how can this be accomplished? I ask because I use the Parametric EQ from the ufx pack quite heavily to increase my low bass at around 12dB/octave so that I get a flat response from my subwoofer (I also remove a pair of room modes with it). Once I get down to about 20Hz I have 15dB of gain - this means that to avoid clipping I've had to set the overall gain to -6dB, and if the sample depth is only 16-bits this digital attenuation could be a real loss of information. Do I have to choose between a flat response and overall detail (I really hate the boomy 80Hz that I get without EQ) or can I upsample the streams to 24-bits prior to EQing, and hence keep the resolution? I saw the 16to32 effect in the Passthru category but can't quite figure out what it does. The same goes for a few of the other plugins in kX (call me thick!). Could ambiguous or little-known plugins have a comment added into the microcode briefly explaining their function? TIA

Sep 11, 2003, 02:38 AM	#4
Max M. h/h member-shmember Join Date: Dec 2002 Location: Evil Empire Posts: 2,639 Rep Power: 69	>it is then processed with mixed 24/32bit accuracy correction: DSP processing itself is done with 32/64bit accuracy __________________

Sep 12, 2003, 05:30 PM	#7
Max M. h/h member-shmember Join Date: Dec 2002 Location: Evil Empire Posts: 2,639 Rep Power: 69	Ok, ok... Let's do not fall into deep discussions here... If we are speaking about equalizers (as this thread is inventionally of) then their precision is 32/64 (ufx eq and Soeren's filters use accumullator _very_ hardly and do not use TRAM)... __________________ Last edited by Max M.; Sep 12, 2003 at 05:54 PM.

Sep 11, 2003, 12:00 AM	#2
Eugene Gavrilov kX Project Lead Programmer and Coordinator Join Date: Dec 2002 Posts: 3,119 Rep Power: 75	the 16-bit audio is converted into 32bit internal representation when it gets inside the DSP it is then processed with mixed 24/32bit accuracy (depending on the effects you are using) the output resolution for audigy2 is 24 bit (that is, the 24 bit DSP data is passed to 24-bit DACs) anyway, there might be better ways of achieving your goals /E

Sep 11, 2003, 12:18 AM	Thread Starter #3
wingfeather DriverHeaven Newbie Join Date: Jul 2003 Location: UK Posts: 18 Rep Power: 0	So the precision is fine for what I'm doing, which is nice. Thanks for the clarification. What are the better ways? Analogue EQs?

Sep 11, 2003, 03:26 AM	Thread Starter #5
wingfeather DriverHeaven Newbie Join Date: Jul 2003 Location: UK Posts: 18 Rep Power: 0	Hell, 64 bits? I'm impressed with that. That's some pretty accurate processing.

Sep 12, 2003, 03:38 PM	#6
Eugene Gavrilov kX Project Lead Programmer and Coordinator Join Date: Dec 2002 Posts: 3,119 Rep Power: 75	well, Max, if we wish to be 100% correct, then it is 24(16)/64 (24(16) is external tram resolution, 32 is the DSP GPR resolution and 64 is DSP accumulator resolution) anyway, '32bit processing' seems to be the 'golden mean' /E

Sep 12, 2003, 11:18 PM	#8
Eugene Gavrilov kX Project Lead Programmer and Coordinator Join Date: Dec 2002 Posts: 3,119 Rep Power: 75	ok I doubt we could have fallen into a deeper discussion, since all the precision discussion stuff is now present in this thread /E

Sep 25, 2003, 09:00 PM	#9
mattcro DriverHeaven Junior Member Join Date: May 2003 Location: Scotland Posts: 29 Rep Power: 0	OK, if we're going to be really picky, the 10k1 (and presumably 10k2) have a 67-bit accumulator in the FX processor and does 32-bit integer/fixed-point processing... from my reading of the EMU FX8010 architecture document Matt