PlanetZ forums - scopeusers.com

players: Scope 2448 mixer, VDAT and EnergyXT as a pure input-as-output ASIO 2/24bit host.

playground: a short track of a drumbox and a Minimax stereo pair (each)

the game: after recording to VDAT (32 bit) the tracks were played back through parallel channels of the 2448 mixer, the Asio track phase inverted, basic latency compensation by a stereo delay, fine compensation by sample delay of the mixer channel.

result: perfect phase extinction to a faint white noise (level full up)
this suggests you will NOT loose any quality for the fact alone you passed it to the Asio driver.

nevertheless there will be an advantage (to be expected) on complex mixes by an increased headroom of the 32bit data, but that ASIO for itself spoils the sound is a myth imho.

comments welcome, in particular if someone sees a flaw in the setup - it ran a bit too straight forward

cheers, Tom

Hi Astro,

I don't think there is a problem with ASIO ... but in cubase processing.
Look, recently I loaded a new patche in halion. Originally all samples are in mono in this patch. It was loaded on a stereo bus internally in Halion and I observed that it sounded dull! I finally corrected it by changing the bus from stereo to mono and it sounded OK.

If I am correct a faked stereo signal is:
A being the mono signal => Stereo = K(A+A) with K being the panning law.

That phenomemom seems to be linked to a micro delay between both halves of the stereo bus (kind of phase cancellation).

If this kind of problem appear in cubase mixer, it could explain the difference in sound (particularly in the upper frequency)

Cheers
Michel

I think you haven't reached the full 32 bit dynamic range and this is why the audio material seems to be equal after the 24 bit asio processing.

On 2006-01-24 13:55, astroman wrote:
players: Scope 2448 mixer, VDAT and EnergyXT as a pure input-as-output ASIO 2/24bit host.

playground: a short track of a drumbox and a Minimax stereo pair (each)

the game: after recording to VDAT (32 bit) the tracks were played back through parallel channels of the 2448 mixer, the Asio track phase inverted, basic latency compensation by a stereo delay, fine compensation by sample delay of the mixer channel.

result: perfect phase extinction to a faint white noise (level full up)
this suggests you will NOT loose any quality for the fact alone you passed it to the Asio driver.

nevertheless there will be an advantage (to be expected) on complex mixes by an increased headroom of the 32bit data, but that ASIO for itself spoils the sound is a myth imho.

comments welcome, in particular if someone sees a flaw in the setup - it ran a bit too straight forward

cheers, Tom

well, people have expressed their concerns several times that the different math model of the 'native' driver might already spoil the recording.
This may lead to much too complicated handling of the recordings just to prevent an imaginary loss of quality.
At THIS stage of the process (and only in numeric context) it doesn't seem to be a concern at all.

Yet I'd rather see the result critical as it was more or less what I expected - and it's too easy to get biased in such cases.

There was another another (strange) observation which I didn't verify due to lack of time, that showed up when manually applying the 'sample offset' to extinguish the signal.
On the first pair I needed 75 ticks, on the 2nd it was only 70 - in other words vdat track1/2 had to be delayed 48ms +75, while vdat track 3/4 needed 48ms +70 to extinguish the respective ASIO pair.

according to Michel's 'stereo' observations mentioned above that could have indeed a much more significant influence on the overall sound quality when the Asio channels 'move' in time unexpectedly.

An extended signal dynamic (from 24 to 32bits) is neglectible imho as it's an imaginary thing (there are no 32bit converters) and for processing the 8 bits are close no nothing anyway (due to dsp math).

cheers, Tom

Hi,

32bits floating point have a 22 bits mantissa. This mean that you have about 126 db dynamic range to differentiate 2 consecutive samples (Think about a limiter with a +/- 126db threshold centred around your signal!) I am not sure we are able to ear the difference (taken the adaptative operating mode of our ears!). Now, if you connect a 32 bits signal to a 24 bit ASIO, than yes there will be a difference depending on the 32 to 24 bit reduction.

Now about what I observed, I think the problem do not lie in the operational code of cubase, but in the conversion process from version to version (cubase projects,vsti,...). In my case a 2.1 halion program to a 3.2 version. It could be possible that, when we update/upgrade cubase the loader do not "translate" it correctly and we accumulate "hidden" bugs in our project.

Cheers

yes, the individual numeric values of the samples in their respective audio streams will certainly be different, but the 'identity test' is performed on the mixer after latency compensation.
It will deal with proper (level) adjustment of whatever is fed into it - just as it's supposed to do.

Anyway, I've repeated the test with 4 stereo pairs and (as expected) the necessary time compensation varied (66 to 70 samples).
The pairs 2 and 3 extinguish perfectly, while 1 and 4 left -38db on the left channel each.
Switching the board to external sync modified the time values, but not the way how the channels reacted (still 1 and 4 versus 2 and 3)
There seems to be a range of variance, yet a change in sound character, spectral content or transparancy was not observable (imho).

Since I have 2 generation one cards and the box runs under Win98 it may not even be representative for the majority of setups.
Not to forget that +/- 3 samples shouldn't be overestimated - even more as the mixer has an adjustment for critical setups
[addition] for whatever reason it seems the mixer that's responsible for the timing deviation in the 5 sample range. If I replace the Asio channels by VDAT tracks I get more or less the same results.

cheers, Tom

<font size=-1>[ This Message was edited by: astroman on 2006-01-25 13:58 ]</font>