Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Bit perfect hubbub? (Read 8421 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Bit perfect hubbub?

If we attenuate with a 32bit internal volume control aren't we losing 1bit per every 6db. I fail to see what is really lost given the amount the bits we start with on most recordings. Frequency reproduction isn't changing, and amplitude isn't changing relatively between freqs.

So why is bit-perfect touted as superior. I doubt that anyone would notice a 10db digital attenuation applied via something like EMU Patchmix of J. River volume control.

I understand certain things like Surround sound must be bit-perfect to be passed via S/PDIF but.....what's the big deal about bit-perfect. Aren't volume controls 32bit for a reason?

jc

Bit perfect hubbub?

Reply #1 –
An "ideal" digital volume control should have higher internal resolution than the input. When requantizing back to a lower resolution signal, dithering should be applied.

If a volume control was designed as a pure "bitshift right with no extra resolution", then you would truncate the signal, and the chance of audible annoying quantization noise would increase the more you attenuated the signal, especially if you were running a non-optimal gain-chain with subsequent amplification.

For situations where many operations are conducted in series, such as a studio production setting, it makes sense to have a high intermediate resolution between different operations, then scale down once a more stable file is to be archived, broadcast etc.

If you are inserting a digital volume control that has e.g. 24 bits resolution input and 24 bits resolution output, then there is no such thing as a "bit perfect" volume control except for unity gain.

-k

Bit perfect hubbub?

Reply #2 –
Your sound card and its driver have to be bit perfect if you want to conserve the pros introduced by extra precision processing by your software. Bit perfect and processing precision have to be distinguished but they need each other to be relevant.

Bit perfect hubbub?

Reply #3 –
Quote from: Nestor Fish on
Your sound card and its driver have to be bit perfect if you want to conserve the pros introduced by extra precision processing by your software. Bit perfect and processing precision have to be distinguished but they need each other to be relevant.



Okay. Here's a scenario.

Externally lock EMU1212M to Lavry AD-10 at 96Khz. Record at 24/96.

Samplitude monitors the input and sends to ASIO out.

With Patchmix, attenuate the ASIO (which runs out to a Lavry DA-10), by 10db.

EMU Patchmix & Samplitude run at 32bit internally.

Result?

At what point (level of Patchmix attenuation) would quantization noise and a loss of fidelity be introduced?

DC

Bit perfect hubbub?

Reply #4 –
Your original data is stored in 24bit. The introduced noise will be below the same as the source quantisation noise as long as you work in 32bit, and attenuation is under 48dB (headroom between 32bit and 24bit; starting from 48dB, least significant bits will become discarded, 1bit/6moredB).
Then you have to process your 32bit data for 24bit output. Depending on how Patchmix do this, you will:
1) lose the famous 1bit per 6dB of attenuation if it only truncates,
2) or introduce noise through dithering and/or noise shaping during/before truncation to try and maintain information below theoretical quantisation floor.

The intermediate 32bit storing is the way for 2) to succeed. But if Patchmix doesn't dither, it is nearly worthless in case of a single attenuation.

Summing up : You will irreversibly lose information (during process) if you apply more than 48dB of attenuation or if your data is only truncated for output. If you stay below 48dB att., It depends on dithering/noise shaping performance:
-how much information below quantisation it 'maintains' - information needed about this -
-relationship between the noise originally present in your source (Lavry doesn't give noise level, but a 117dB dynamic range for the AD10, which would correspond to the 20th bit, before attenuation), the one introduced by conversion and by DA10.

I hope I'm clear enough. 

Bit perfect hubbub?

Reply #5 –
I'm reading some of Dan Lavry's stuff:

"Clearly, we need more space on the most significant bits. If you “run out” on the most significant bits side, the distortions are horrible. The signal is badly clipped and the music is not even recognizable.

But while more subtle, we also want more processing bits on the least significant side. Example: Say you have a 20 bit material and 24 bit DAW. You have enough “space” to attenuate the signal by 4 bits (24dB). Say you wanted to attenuate by 6 bits (36dB attenuation). That new attenuated track is 2 bits short. So far we are “OK” because those missing bits are way below the level we can hear. They are in fact bits 25 and 26 (representing dynamic range of 156dB). The signal is now weaker by 36dB so the level of the lower bits is way below our hearing capability. "

http://lavryengineering.com/lavry_forum/viewtopic.php?t=24

thanks for your replies
DC

Bit perfect hubbub?

Reply #6 –
You've just made me notice I went wrong: I considered 32bit integers instead of (I guess) 32bit floating point numbers. The latter would remove the attenuation problem (by putting it far far away from standard attenuations). So forget about the 48dB stuff... Sorry. Dithering/noise shaping method is still relevant, as well as masking due to input & output devices noise level.

Bit perfect hubbub?

Reply #7 –
I'm trying to find out if:

Patchmix uses 32bit internal processing. If I set Samplitude 10 to use 24bit to record a 24bit/96khz input (from Lavry AD-10) using an ASIO Send, AND monitor that recording via ASIO OUT 1&2...how do I best maintain the integrity of the monitored signal (playback)?

Assuming I don't use any digital attenuation on the Patchmix ASIO strip:

1. Use an a SEND S/PDIF output in Master
2. Use an a SEND S/PDIF output in the specific ASIO 1&2 coming from Samplitude.

Currently I use a Send S/PDIF output in the Master because this allows both Wave and ASIO output to make it to the same device (Benchmark DAC1). Currently, video playback on most HTPCs utilize wave output for DirectShow). Anyway, I'm not so concerned about bit-perfect, but if I needed to have bit-[perfect, is the ASIO S/PDIF output preferable in the Master or specific ASIO out strip. Is dither applied as in the Master or is it only if processing is engaged (i.e. digital attenuation)? When does 32bit internal "kick-in" (with trim pots I assume but what about faders). All my output strips are pre-fader.

I want to keep the 24bit output "clean" and I am not so much worried about losing a few bits (if I decide to use digital attenuation) on an ASIO out strip rather than having dither applied when I'm not doing any processing post-Samplitude. I just want Patchmix to output to an external DAC w/o adding dither if I 'm using the Master strip (with send to s/pdif) bit not "doing anything" else to the signal.

dc

Bit perfect hubbub?

Reply #8 –
Joncat,

While I was "composing" this, you posted from Dan Lavry's site.  I was trying to say the same thing, but I'm going to click Add Reply anyway....

Quote
If we attenuate with a 32bit internal volume control aren't we losing 1bit per every 6db. I fail to see what is really lost given the amount the bits we start with on most recordings. Frequency reproduction isn't changing, and amplitude isn't changing relatively between freqs.
  When you attenuate, all of the lost data is at the "quiet end".    So, even if you could keep the lost details/data, these details would be be attenuated below the "old" 1-bit level.      (In fact, you can preserve the details if you use floating-point, but you're still not going to hear anything different.) 

Something similar happens with an analog volume control.  The "lost details" fall-into the noise or below the threshold of audibility.

So, you can digitally attenuate a signal 20 or 30 dB (or as much as you want), and you don't hear any "distortion".  You can attenuate it 'till you're down to 1-bit remaining, which would be really lousy audio, but you won't hear anything anyway!

However, if you re-amplify the signal (digitally or by turning-up the volume control on an analog amplifer) you might hear the effects, but you're probably not going to hear any ill-effects with normal adjustments (i.e. 10dB) at 24-bits. 

Quote
With Patchmix, attenuate the ASIO (which runs out to a Lavry DA-10), by 10db...

...At what point (level of Patchmix attenuation) would quantization noise and a loss of fidelity be introduced?
  It shouldn't be a problem under "normal conditions".  At 24-bits, 10dB is "nothing" (assuming normal levels).  This is the advantage of 24-bits.  You've got plenty of dynamic range for headroom, mixing, and processing.    I don't work in pro audio, but it's my understanding that the "meters" are often calibrated to read 0dB at -18dBFS.  This leaves 18dB of "wasted" headroom, and still gives you plenty of resolution at the low end. 

So, I don't think it's a question of attenuation, it s a question of headroom and dynamic range.  If you've got a 24-bit, 0dBFS signal, and you attenuate it by 10dB, you've still got plenty of dynamic range.  But, if you've got a -60dB signal and you attenuate it by 10dB (and amplify it to the point where you can hear it), that might be a problem.

Doug.

Bit perfect hubbub?

Reply #9 –
Thanks all for your insight!

I've been swamped with school and daddy-dom lately, so sorry for the delay in responding.

EMu sent men this reponse:
Quote
"We think that you are confusing converter bits with processing bits. For the details of each and the differences, we suggest you do a web search as this info is widely available.

There is no dithering done by the E-MU hardware or software. As long as you are not attenuating or adding gain, there should be no signal degradation using ASIO, WDM or S/PDIF."


I am now using "the 24bit/96kHz setting on the AD-10 feeding the "24bit driver" of the EMU hardware, with Patchmix passing the ASIO direct send to Samplitude which records at 32bit float: Samplitude sends the signal after realtime effects (digital RIAA, crackle removal, rumble filter) back out to the EMU Patchmix which goes ASIO to Lavry DA-10 for monitoring on the hi-fi. It sounds good. No attenuation in the Patchmix or elsewhere; the flat gain pre is giving me really good results in terms of SNR and dynamic range. I don't think I will even do any normalization in the digital domain with -10db average and -2db peaks (for a Maceo 180g re-issue at least).

I plan to dither to 24bit rather than leave at 32bit after any editing and archive them. From what I have read on the Lavry forum dither has truly audible negative consequences if not done.

I was trying to find mention of types of dither and the tools to apply them. Some DirectX or VST plugins. Or should I just use whatever stock form is offered in Sampltiude?

best
JC

Quote from: DVDdoug on
Joncat,

While I was "composing" this, you posted from Dan Lavry's site.  I was trying to say the same thing, but I'm going to click Add Reply anyway....

Quote
If we attenuate with a 32bit internal volume control aren't we losing 1bit per every 6db. I fail to see what is really lost given the amount the bits we start with on most recordings. Frequency reproduction isn't changing, and amplitude isn't changing relatively between freqs.
  When you attenuate, all of the lost data is at the "quiet end".    So, even if you could keep the lost details/data, these details would be be attenuated below the "old" 1-bit level.      (In fact, you can preserve the details if you use floating-point, but you're still not going to hear anything different.) 

Something similar happens with an analog volume control.  The "lost details" fall-into the noise or below the threshold of audibility.

So, you can digitally attenuate a signal 20 or 30 dB (or as much as you want), and you don't hear any "distortion".  You can attenuate it 'till you're down to 1-bit remaining, which would be really lousy audio, but you won't hear anything anyway!

However, if you re-amplify the signal (digitally or by turning-up the volume control on an analog amplifer) you might hear the effects, but you're probably not going to hear any ill-effects with normal adjustments (i.e. 10dB) at 24-bits. 

Quote
With Patchmix, attenuate the ASIO (which runs out to a Lavry DA-10), by 10db...

...At what point (level of Patchmix attenuation) would quantization noise and a loss of fidelity be introduced?
  It shouldn't be a problem under "normal conditions".  At 24-bits, 10dB is "nothing" (assuming normal levels).  This is the advantage of 24-bits.  You've got plenty of dynamic range for headroom, mixing, and processing.    I don't work in pro audio, but it's my understanding that the "meters" are often calibrated to read 0dB at -18dBFS.  This leaves 18dB of "wasted" headroom, and still gives you plenty of resolution at the low end. 

So, I don't think it's a question of attenuation, it s a question of headroom and dynamic range.  If you've got a 24-bit, 0dBFS signal, and you attenuate it by 10dB, you've still got plenty of dynamic range.  But, if you've got a -60dB signal and you attenuate it by 10dB (and amplify it to the point where you can hear it), that might be a problem.

Doug.

Bit perfect hubbub?

Reply #10 –
Some stuff on the Waves dither:


"Chapter 4 - Recommended IDR settings
Of course these settings are for when you have access to the full IDR control set, at present just the +L1-
Ultramaximizer, or the L2 hardware.
Remember: The small IDR plug-in is internally set to type1 Normal.
Any combination of dither and noise shaping can be used, but the following settings are particularly recom-
mended for different applications.
• General Purpose high-quality use, including material liable to be edited, EQ’d, and re-dithered: type1
Normal
• Lowest Noise: type2 - Ultra
• Low Noise/Highest quality (final production masters): type1 - Ultra
• Low noise while allowing editing/EQ: type2 - Normal
• High Quality, with lowest risks of spurious noises on edits or cheap CD players: type1 - Moderate
• Low noise, with lowest risk of spurious noises on edits or cheap CD players: type2 - Moderate"

Bit perfect hubbub?

Reply #11 –
Quote from: Joncat on
I was trying to find mention of types of dither and the tools to apply them. Some DirectX or VST plugins. Or should I just use whatever stock form is offered in Sampltiude?

If your version of Samplitude includes POWr, I suggest using that. POWr has three different profiles you can select from, though profile three is described as being the most "universal", i.e. appropriate for various types of audio.

Bit perfect hubbub?

Reply #12 –
Quote from: Ron Jones on
Quote from: Joncat on
I was trying to find mention of types of dither and the tools to apply them. Some DirectX or VST plugins. Or should I just use whatever stock form is offered in Sampltiude?

If your version of Samplitude includes POWr, I suggest using that. POWr has three different profiles you can select from, though profile three is described as being the most "universal", i.e. appropriate for various types of audio.


Thanks for the heads up on POWr.

Some results on it compared to L1&2 here:

http://www.playgroundstudio.com/dithering-in-mastering.htm

DC
 

Bit perfect hubbub?

Reply #13 –
Quote from: Joncat on
So, I don't think it's a question of attenuation, it s a question of headroom and dynamic range.  If you've got a 24-bit, 0dBFS signal, and you attenuate it by 10dB, you've still got plenty of dynamic range.  But, if you've got a -60dB signal and you attenuate it by 10dB (and amplify it to the point where you can hear it), that might be a problem.


Looks to me like this whole discussion assumes fixed point (integer) data and processing.

You do know that the native high-res mode of CoolEdit Pro and Audition has been 32 bit floating point (24 bit mantissa, 8 bit exponent) for as long as the products have existed, right?

An ideal 32 bit floating point process can attenuate or amplify a 24 bit signal by any power of 2 without any loss of precision.

(BTW, I don't think that ever happens in a real world DAW, but the loss of precision is in general, less than with fixed point)