Something wrong with Pow-r 3 dithering? Much noisier than Pow-r 2, Pow-r 1

I'm using SONAR Producer v 8.5.2.222
 
I recently was checking a finalized exported mix by opening the stereo .wav file in Adobe Audition.  I noticed noise in the quiet places (like the ending fade) at around -66dB on the Audition meters.  Wondering if it was something in the recording, I created a new project, and with only one track and no busses, effects, etc. I recorded several seconds of silence at 44.1\24.  I then opened the resulting file in Audition.  You can magnify the amplitutde of the wave, and the recorded file was abolutely quiet up to the maximum magnification (scale begins at -150dB).  I then  exported the recording like I would a finished mix, using File\Export\Audio and made the settings 44.1\16\Pow-r 3.  The resulting output does indeed have consistent noise that hovers between -69 and -63.  At that level it is absolutely audible above the self noise of my monitors.  I tried the same thing again, but I used Pow-r 2 and Pow-r 1 instead.  The resulting files have noise that hovers between -81 and -75.  This is a significant difference!  The noise is barely detectable above the noise floor of my monitors, and rather than being louder, it is just a different color.
 
Can someone please confirm my findings?  Just record a silent track at 44.1\24 and export to 44.1\16 using Pow-r 3 and do another export with Pow-r 2 and compare the results.  I can add plenty of audible noise to my recordings on my own.  I don't need the dithering tool to do it for me :)
 
Thanks,
 
Poco
 
P.S. I am familiar with the manual.  I have searched the forum (much information about dithering there.  Some good some not.  Nothing on this specific issue)...

A while back I did some preliminary mastering on some tracks in Sonar before sending them out & found I didn't care for Pow-r-3 as most seem to.
Actually thought triangular setting had a bit cleaner sound on my stuff.

I'll watch this thread too, I'd like to read any other results or comparisons to Poco's.

From my understanding, the numbers everyone is seeing are correct.  The problem is we're looking at dither in silence.  Silence is the time when dither is useless.  Dither is supposed to help with quiet parts, not silence.  It can allow a 16 bit signal to be perceived, without distortion, below -96dbFS...even through the dither "noise".  It's a bit of trickery.  It's not just any noise, though.  Sure, ones like Triangle look more like constant noise across the whole frequency range, but Pow-r is filtered/shaped.  Pow-r 3 is seen as the "preferred" of the choices because of it's shape that drops around 3k, an area we are most sensitive, and the loudest frequencies are above 10k.  Here is a look at a few(I just did these):

Blue: Triangle     Green: Pow-r 1     Orange: Pow-r 2     Purple: Pow-r 3

Notice the top of the graph is -90db.  I left the stats up for Pow-r 3 so you can see, yes, it does get into the -60s, peak.

Most of this is summed up from the Bob Katz "Mastering Audio" book.  He gets into Pow-r on pg. 59(2nd Edition) and there are pics that mirror what you see above.  Which kind to use still seems to come down to what you think works best for the current project.  There are options out there for dealing with the silence(think gated dither) but not in Sonar.  Well, hopefully that helps some.
 
Too tired to go on...please correct me if I'm wrong.
Forgot to add...I'm using 8.5.3.

Well, hopefully that helps some.

Yes. Thanks. Makes sense, and I'm glad you were able to put this thread to bed early.


Still a little surprising that the higher frequency components are at such a high level, and I kind of wonder what happens when you run it through an MP3 encoder. Is it still better than no dither in terms of the character of the noise that results?

  For those of you that can't hear it, it may be that the noise floor of your monitoring is too high (no offense intended).  -60s is actually pretty noticeable, especially when compare to complete silence.

Also no offense intended, but I think you need to check the rest of your signal chain. I just did a little testing, and confirmed what bil_g's chart shows... that the dither signal level falls off significantly with decreasing frequency. Below 16kHz, the RMS level is down around -95dB (consistent with that one-bit wiggle I was talking about).


I don't know about you, but I know my 49-year-old ears can't hear those levels at those frequencies, even with monitoring levels cranked. The activity in your meters might be disconcerting, but I'd be really surprised if even teenage ears would notice it.

I've never been a fan of dithering inside Sonar. That's not to say there is something wrong with the dithering functions, but I always use a dedicated dithering device and do it outside of Sonar. It's given me the best sound and least amount of audible noise. Something from Waves, PSP Xenon, or the tools in WaveLab 6 are what I use these days.

The levels are cranked pretty loud to hear it, but not above what my JBL 4410's can handle with regular program material at that volume.   If you can't hear it, you may want to check your system's self noise.  It may be masking the dither noise.

I don't know. If anything, it's just that my ears won't go there any more, though they're pretty good for my age, I think. I was monitoring direct from the headphone out on my interface, ultimately with the level control cranked to max, and couldn't hear a thing when muting/unmuting the noise sample. I'm pretty sure my E-MU 1820m and BeyerDynamic DT770 headphones should be reproducing those high freqs as well as anything, and the headphone out is extremely quiet.

I played around with it some more, and if I bump the track level up 6dB, I can just start to hear it, but any normal signal would be absolutely deafening at that monitoring level; it was downright scary to have phones on with the knob all the way up.

I still can't imagine anyone would hear this at normal listening levels without knowing what they were looking for. 

and BeyerDynamic DT770 headphones

I love my Beyer 770s :) And I'm trying out the 990s now.

To Poco, the only test you should do to try to hear dither is to play it back at your normal-to-normally loud listening levels. The whole point of noise shaped dither is that they move the energy of the noise from where the ear is most sensitive to where the ear is least sensitive.  And all that energy still has to be there, so the high frequency noise has to be at a much higher raw level. But since your ear isn't as sensitive at that level, you won't hear it under normal conditions.

So, are you listening under normal conditions? Adjust your monitors until they are as loud as you would like to listen. Not as loud as they can go, and not as loud as you can stand. Turn them as loud as you would ever want to listen to your mix--I'm going to guess that it's going to be around 95 dB sound pressure level.  Can you still hear the dither?

if you have to crank up your monitors to hear the dither, then it's fine. That's normal. Try it with any dither and you'll probably get the same results.

  About halfway through one of the tunes, his cell phone went off.  Fortunately, it's in the same key as the song... 

That's funny. I occasionally have the opposite problem: I'm playing piano, and the overtones make me think I hear my phone ringing so I stop playing, and... nothing.

But instead I´ll add a lot of noise to the range that can also be heard, as annoying hiss. It´s a bit stupid?

 
Yeah, you're right, all those PhD audio engineers who developed this stuff are a bunch of dolts.
 
Just kidding. I understand the conundrum, but it's still based on the premise that you can "clearly" hear an "annoying hiss" from dither, which I don't accept.
 
Dither operates at a level, and in a way which is meant to mask quantization noise (which happens at all levels, not just the quiet parts) when rendering from 24 bits down to 16. The idea is that you can't distinctly pick out the noise, but you can perceive the reduction in distortion due to quantization.
 
As I've already confessed, I know very little about the details, but it seems clear that the more sohpisticated noise-generation patterns and distribution of the various Pow-r dither algorithms are intended to enhance the effectiveness of the perceived reduction in distortion (or maybe more aptly the substitution of less obnoxious forms of distortion) without making the dither noise apparent at audible frequencies and levels.
 
If there is any hiss detectable at the tail end of a recording, it's only at extremely high listening levels which would probably render you incapable of hearing hiss at the end, anyway. It's not a reasonable test to crank up the volume to abnormally high levels and then only listen to a sample of dithered silence without listneing to the ear-splitting sound levels that would precede it in practice.
 
 

Poco


I'm using SONAR Producer v 8.5.2.222
 
I recently was checking a finalized exported mix by opening the stereo .wav file in Adobe Audition.  I noticed noise in the quiet places (like the ending fade) at around -66dB on the Audition meters.  Wondering if it was something in the recording, I created a new project, and with only one track and no busses, effects, etc. I recorded several seconds of silence at 44.1\24.  I then opened the resulting file in Audition.  You can magnify the amplitutde of the wave, and the recorded file was abolutely quiet up to the maximum magnification (scale begins at -150dB).  I then  exported the recording like I would a finished mix, using File\Export\Audio and made the settings 44.1\16\Pow-r 3.  The resulting output does indeed have consistent noise that hovers between -69 and -63.  At that level it is absolutely audible above the self noise of my monitors.  I tried the same thing again, but I used Pow-r 2 and Pow-r 1 instead.  The resulting files have noise that hovers between -81 and -75.  This is a significant difference!  The noise is barely detectable above the noise floor of my monitors, and rather than being louder, it is just a different color.
 
Can someone please confirm my findings?  Just record a silent track at 44.1\24 and export to 44.1\16 using Pow-r 3 and do another export with Pow-r 2 and compare the results.  I can add plenty of audible noise to my recordings on my own.  I don't need the dithering tool to do it for me :)
 
Thanks,
 
Poco
 
P.S. I am familiar with the manual.  I have searched the forum (much information about dithering there.  Some good some not.  Nothing on this specific issue)...

Hmm...I've always used Pow-r 3 and have never heard any noise like you describe....let me try to confirm your findings...I have some free time today :)

Ben

All the levels you're stating that you hear noise on are supposed to be inaudible to the human ear. So I'm wondering what you are actually hearing? Recorded tracks with a visible line where there is no sound still have some level of audio unless you delete those sections. 

Maybe he WAS listening to a dot matrix... never know...