reducing level using gain effect between prolog and asio

mattcro · Sep 22, 2003, 06:25 AM

I've inserted a gain plugin/effect between prolog line-ins (ac97) and the asio on epilog in an attempt to reduce the line-in noise level a bit. setting the gain to about 25% allows me to get good input level and reduce the noise floor somewhat (using sblive 5.1). But I wonder if i am compromising the bit depth of my input signal? I read that the ac97 adc is 18 bits, and if the full 18 bits are available to the DSP plugs then should i be getting the top 16 bits into asio (with gain at 25%). That's assuming that with no gain/attenuation, the asio inputs receive the low 16 bits from the codec... Anybody have any info on this?

Cheers,

Matt

Max M. · Sep 22, 2003, 02:02 PM

>I've inserted a gain plugin/effect between prolog line-ins (ac97) and the asio on epilog in an attempt to reduce the line-in noise level a bit. setting the gain to about 25%

This is indentical to reducing level via ac97 fader on "ins'n'outs" page (just set it to the same 25%)...

mattcro · Sep 23, 2003, 04:26 AM

heh, thanks Max for pointing out the "ins and outs" page ac97 slider, i was trying to use the ac97 recording slider, but that only works for wdm/mme (or whatever you call it...). I wanted to reduce the record level in both wdm and asio applications, and it's working now.

What I am trying to find out is if reducing the input level (input scaling I guess) like this compromises the bit depth or resolution of recording. If i reduce the level to 25%, the noise floor drops by approx 12dB which is what I would expect. But does that mean I am getting 14bits of effective resolution instead of 16? I don't suppose it's worth worrying about, since the noise floor is still about -70dB (SBlive 5.1 line in, no livedrive).

Cheers,

Matt

TravelRec. · Sep 23, 2003, 08:06 AM

Hi Matt!

Why don´t you use an APS-Expander-Plug to reduce the noise from Line In? If there is only noise, then the Expander shut down the channel and you hear nothing. And if there comes a signal, then the "door" opens an let it passing through. Try it! And visit:

Is this possible??

It may answer some questions...

TravelRec.

kokoon · Sep 23, 2003, 12:27 PM

as far as i know the dsp cant record in 18bit. the 2LSB bits are just gibberish if i understood eugene right.

Max M. · Sep 23, 2003, 11:56 PM

>What I am trying to find out is if reducing the input level (input scaling I guess) like this compromises the bit depth or resolution of recording.
>If i reduce the level to 25%, the noise floor drops by approx 12dB which is what I would expect. But does that mean I am getting 14bits of effective resolution instead of 16?

not "noise level" itself is important but "signal-to-noise ratio" is.

Well, it's a bit complex story...
Simply, you do not improve signal-to-noise ratio at all just by reducing that level.. really... (and yes if the maximum peak of your recorded signal is ~-12dB (e.g. 25%) then you get only 14-bit effective bit-witdh... - and it does not depends on ADC resolution whatever it is (16 or 18 bit) and it does not matter do that extra 2 bits reach DSP or not...)

Actually the right way (with this hardware/software) to get the better signal-to-noise ratio is to get a much as possible (but not clipping/distorted) level at ADC input (of ac97) and leave "AC97" level (the one at mixer pages or the one you tweak via gain plugin) at -0dB...

Reducing that input level could make sence (i mean of course only Live/Audigy hardware scheme) only when it takes place between codec's analog part and its ADC (whatever resolution it has)...
(that codecs analog is one of main noise sources of ac97 (but not the only of course))

(btw. there's some control for that at ac97 codec but unfortunally it is not available through kxmixer (but it could be added there via ome trick))

**Eugene Gavrilov** · Sep 24, 2003, 06:59 PM

>> there's some control for that at ac97 codec but unfortunally it is not available through kxmixer

what control do you mean? master?
if one is recording in 'DOO' mode the controls are already available for each source
while in 'analog' mode such controls aren't used...

more precisely: what AC97 register do you mean?

/E

Max M. · Sep 24, 2003, 07:55 PM

mattcro · Sep 25, 2003, 02:51 PM

travelrec, I can use an expander or noise gate in Cubase if necessary on the input, so I haven't bothered with aps expander.

Max, I increased the audio input to the line-in so my record level can still peak at 0dB, but with the noise floor 12dB lower. I think that means I am increasing the SNR (certainly looks that way), so either the DSP provides gain (+dB) when the AC97 mixer slider is at 100%, or the input to DSP from AC97 is more than 16bits (or perhaps is shifted). I dunno, but what I have now seems to work ok, and the SNR appears to be better

I'm going to try reducing the AC97 input slider further and see what effect it has on recording. I seem to remember that with Creative drivers and mixer applet, reducing the input sliders more than a certain amount made it impossible to get 0dB recorded level (and analog clipping wasn't apparent IIRC).

Matt

PS I don't suppose anyone knows where to get the 10k1 datasheet (or if it's available at all to non-Creative/EMU folks...) I'm just curious, as an electronics engineering student

.

**Eugene Gavrilov** · Sep 25, 2003, 04:14 PM

>> 1c

probably you are right
a 'small trick' (for those who aren't aware of this feature) is to run 'kxctrl -sac97 1c <value>'
read the AC97 specification for details

the ac97 register setting should be saved and automatically restored on reboot
the register will be set to '0' if you click 'reset settings' or 'reset ac97'

/E

Max M. · Sep 25, 2003, 11:23 PM

mattcro

>Max, I increased the audio input to the line-in so my record level can still peak at 0dB, but with the noise floor 12dB lower.

Btw. you have to be right and it have to be my mistake then (and damn, i can't check it) - (E. fixme please)... Those AC97 input seems have 0dB gain exactly at 25% (while 100% is equal to +12dB boost which means bad things taking place after ADC and makes sence only when input signal has not enough level to feed the full range of ADC)...

Anyway all other my statements are correct (e.g. the only question is what level means no cut/boost for ac97 input in dsp - 25% or 100%? - of course we've measured and knew this a while ago but i might forget this since time i get platinum model ;)

well, i'll let you know this for sure as soon as i get my kx-hardware back...

mattcro · Sep 26, 2003, 11:24 AM

OK, I've been doing some checking and it looks like setting the AC97 In/Out level to below 26% (one step up from 25%, each % on the sliders has two 'steps' if you use up/down arrow keys), or the AC97 Record slider to below 26%, prevents the record audio stream (ASIO or WDM/MME) from reaching 0dB. Reducing either slider below 26% causes a proportional reduction in max record amplitude. Zooming in on the recorded waveform in Cooledit shows a sudden, distinct clipping which looks more like a DSP output than analog stage clipping. I haven't checked the audio input level at the line-in socket (I'll drag out the oscilloscope...) but I guess there is enough analog headroom.

My guess is that 25% on the sliders equates to 0dB (i.e. direct signal from ADC) and 100% is 12dB of digital gain (think of digital zoom on a digital/DV camera). At slider settings below 26%, the ADC is probably doing the clipping. Could be wrong though, anyone know a friendly Creative Labs engineer to ask?

Matt.

**Eugene Gavrilov** · Sep 26, 2003, 03:58 PM

>> My guess is that 25% on the sliders equates to 0dB

hmm

then some of our -very-earlier-observations- weren't correct
two years ago we (Max and I) tested AC97 inputs and checked all the required amplifications/attenuations
that's why in the present DSP layout ('prolog') ac97 source is the -only- one that is not pre-processed
it seems that this was wrong

unfortunately, I haven't got any cards with the ac97 codec recently
(and I cannot test this with E-mu APS and Audigy2Platinum Ex)

so, the problem is with DSP amplification
that's why setting ac97 slider to '25%' solves this issue

please double check that and let me know -- I will fix the 'prolog'

anyway, the next driver release will be using different prolog/epilog plugins allowing additional settings
currently, one can only attenuate the inputs & outputs -- I will add amplification feature as well as phase inversion

/Eugene

mattcro · Sep 30, 2003, 02:35 PM

OK, I've tried this on two PCs, both have SBLive 5.1 digital, one using standard Creative drivers and one using kX 3534. The AC97 codecs are Sigmatel STAC9708 (though I suppose all the AC97 codecs are basically the same, maybe different analog routings??) Both systems have the 25% behaviour described before on the AC97 record source.

The only other explanation I can think of is the codecs feed 18bit samples to the DSP, which uses the low (LSB) 16 bits at 100% setting (maximum 'gain', MSBs get truncated) and at 25%, the samples have been scaled down so that the DSP is effectively outputting the 16 MSBs (lower gain, LSBs get truncated). Below 25% the scaling ouputs values less than full scale so 0dB cannot be output, regardless of input. Might be right, might be wrong, I don't know

Matt.

**Eugene Gavrilov** · Sep 30, 2003, 04:09 PM

try 3535b -- I've removed x4 amplification

/E