low digital out level=low bits??

[maks]

hi all,
its common knowledge that if recording to digital and the source signal is low you lose bits, does this also apply to outs from a converter into analog?

also does anyone have any experience or commentary on this new gadget? a passive attenuater studio controller?? looks very promising....at 99$

http://www.smproaudio.com/mpatch.htm

thanks

[dns]

low digital out level=low bits
is true, always

first of all:
a D/A has a constant noise floor, if you lower the output, there is less room for the signal

inside the DAW (or whatever) two things can happen:
fixed point:
the number are take from a fixed range, the lower you get, the less dynamic is left (less bits are used)

floating point:
the numbers can be extremly high or low - no problem. Just because the CPUs have problems with very small numbers, a (really little) noise is added. So no endless dynamic - but I don't care about adding 10 db gain at the end of the chain.

Normaly (look at the specs of what you are using) floating point calculation is used inside a unit, but the D/As and A/DS are using (always) fixed point .

greetz

[norman_nomad]

Originally posted by maks
hi all,
its common knowledge that if recording to digital and the source signal is low you lose bits, does this also apply to outs from a converter into analog?

Most modern audio editing programs run a very high internal bit resolution (protools runs 48bits) which allows you to turn the signal up or down without truncation.

[Charlie-O]

Is the this true with pt le also?

[Brad Blackwood]

Originally posted by maks
hi all,
its common knowledge that if recording to digital and the source signal is low you lose bits, does this also apply to outs from a converter into analog?

Sure, but do you understand what the 'loss of bits' means and whether or not it's something to worry about?

[jtienhaara]

Any small dynamic range will lose bits. If you compress the crap out of your signal you'll lose bits, too.

[Brenton]

Originally posted by maks
also does anyone have any experience or commentary on this new gadget? a passive attenuater studio controller?? looks very promising....at 99$

http://www.smproaudio.com/mpatch.htm

thanks

Yes, I recently got one so i could easily switch between 2 sets of speakers and two sources PTLE and CD player. It also has mute and mono which is great. Very happy with it, mind you my system is fairly lowend. I don't notice any difference in sound quality on my system.

[maks]

thanks guys

"Sure, but do you understand what the 'loss of bits' means and whether or not it's something to worry about?"

i assume that low-high harmonics of a source and its respective levels are basically distorted due to the lack of required bits to capture them properly in the first place, resulting in a thin uncomplete fuzzy sound? if one is synthesizing drums sounds or whatever it can work in ones favor, but trying to capture acoustic material with a purist frame of mind, it wouldnt be acceptable?

am i in the ballpark?

so as i understand it, within the DAW it doesnt really matter much what fader level or plug levels are at , but it is the fader(master) feeding the DA(soundcard) that one has to worry about for optimal fidelity?


some day ill buy me a Lavry , and forget about all this..in the meantime im off to make lemonaid with the lemon i got

[jtienhaara]

Bits represent amplitude. So if your signal is squashed to the point that harmonic overtones aren't represented, then yes you're affecting the sound. However long before you get to that stage of squashing, you'll find yourself listening to noise floor, as dns mentioned.

Also sample rate has a heavy effect on the quality of harmonic overtones reproduced. For example, a sample rate of 11,000 Hz would represent a sine wave of 5,500 Hz as either "positive" or "negative" (or, in worst case, 2 zero crossings) -- with no variation in between. A sample rate of 44,100 Hz would give you roughly 8 samples across the cycle of the wave; and so on.

Cheers,

Johann

[pedalboy]

If you recorded in a signal at a low leve, boosting the fader will of course not really net you any more bits. No new information there. ANY DSP process you do on audio loses bits, pretty much. If you bost the fader 1 db you gotta run math on that waveform to change the amplitude. Pretty much you're going to do math on your audio, so get over it (assuming you have a few tracks and maybe some plugins and mixing to do at least). Bounce it out about as hot as you can without clipping. That's what i'd do anyway.

[norman_nomad]

Originally posted by maks

so as i understand it, within the DAW it doesnt really matter much what fader level or plug levels are at , but it is the fader(master) feeding the DA(soundcard) that one has to worry about for optimal fidelity?

Bits do not = resolution, bits = dynamic range.

In a well designed DAW, moving the individual faders and moving the master faders should yeild the same sonic results.

What is it exactly that you're trying to do / worried about?

[norman_nomad]

Originally posted by pedalboy
If you recorded in a signal at a low leve, boosting the fader will of course not really net you any more bits. No new information there. ..... Bounce it out about as hot as you can without clipping.

If boosting the fader is superfluous in a daw, then why bounce out as hot as one can?

[jtienhaara]

Originally posted by pedalboy
ANY DSP process you do on audio loses bits, pretty much.

Expansion being a notable exception.

[maks]

i guess when Bob Katz mentioned that gain staging is a science of itself he meant it.

[Duardo]

i assume that low-high harmonics of a source and its respective levels are basically distorted due to the lack of required bits to capture them properly in the first place, resulting in a thin uncomplete fuzzy sound?

If you're recording at a low enough level that things are sounding "fuzzy", your problem isn't a lack of harmonics...you're probably hearing quantization noise and/or dither. Bits have nothing to do with harmonics, really...just amplitude. If something's low enough in level it won't be captured, regardless of whether it's a fundamental, harmonic, or whatever. Even with a sixteen-bit system, unless you're recording something with an extremely wide dynamic range, you've got plenty of room to capture most sources. Nice thing about recording at 24-bit resolution is you usually don't have to worry about keeping things as hot as possible, since you've got plenty to work with.

Also sample rate has a heavy effect on the quality of harmonic overtones reproduced. For example, a sample rate of 11,000 Hz would represent a sine wave of 5,500 Hz as either "positive" or "negative" (or, in worst case, 2 zero crossings) -- with no variation in between. A sample rate of 44,100 Hz would give you roughly 8 samples across the cycle of the wave; and so on.

If you're just talking about a sine wave, then there are no overtones. It doesn't really matter how many samples you have across the cycle of a wave...as long as there are more than two, you can capture it accurately and increasing the number of samples does not increast the "accuracy". The 5.5kHz/11kHz example is not a realistic one because you can't sample at exactly twice the sampling rate accurately. If you had a frequency of 4.9 kHz, or 11.1 kHz, you'd be fine (theoretically at least...it does depend on the quality of the filters that are used). That's why we use 44.1 kHz to capture 20 Hz-20 kHz.

-Duardo

[jtienhaara]

Originally posted by Duardo
If you had a frequency of 4.9 kHz, or 11.1 kHz, you'd be fine

Sampling 11.1 kHz signal with an 11 kHz sample rate...? Errrm... Why?

A harmonic overtone is certainly a sine wave. You can break up the most complex waveforms into their sine wave components.

The point being that with higher sample rates you get a much better picture of all the high frequency overtones in a complex waveform.

Whereas an 11kHz sample rate would at best represent a 5.5 kHz signal as a square wave, a 44.1 kHz sample rate would provide some modicum of precision for the same tone. Even if it's just a harmonic overtone on a much lower frequency fundamental.

For capturing overtones, a higher sample rate is always desirable.

Whether capturing overtones is always desirable -- that's a completely different matter.

[Bob Olhsson]

Originally posted by jtienhaara

For capturing overtones, a higher sample rate is always desirable.

ONLY if the frequency in question exceeds half the sample rate.

Higher sample rates require greater precision in all calculations so if you hold all other factors equal there will be some point of degradation at all frequencies from the computational precision not being able to keep up with the sample rate.

It isn't as simple as just cranking the clock up when you go beyond around 60kHz. sample rate. Bob Ludwig once said "never turn your back on digital." He wasn't kidding, you need to listen very carefully before assuming higher sample rates and sometimes even bit depths are going to sound better. Potentially, of course they will but unfortunately a lot of common gear doesn't live up to its potential.

[jtienhaara]

Originally posted by Bob Olhsson
ONLY if the frequency in question exceeds half the sample rate.

Not true. A signal that is half the sample rate is on the verge of being completely lost. If you desire to capture high overtones with any degree of precision then you need a high sampling rate. Though decent dithering will help a lot too.

you need to listen very carefully before assuming higher sample rates and sometimes even bit depths are going to sound better.

Like I said in my previous post: Whether capturing overtones is always desirable -- that's a completely different matter. There are many factors to take into consideration.

I've approached this whole thread as a technical discussion. Not a debate about what is subjectively "best". I don't care who likes what sample rates best. I am quite happy working at 44.1. I'm just tired of the half-assed math that gets thrown around in digital audio discussions.

Here is the absolute ideal situation for sampling a sine wave that is 1/2 the frequency of the sample rate. (Worst case scenario you get a flatline!)

EDIT: Aw, crud. Dia crapped out on me and didn't render the sine wave. I'll try to fix it later.

[SteveHalko]

Originally posted by jtienhaara

Here is the absolute ideal situation for sampling a sine wave that is 1/2 the frequency of the sample rate. (Worst case scenario you get a flatline!)

The square wave (11 kHz sample rate) and the stairstep (44kHz sample rate) are what you would get by putting your samples through a sample and hold circuit. Fortunately, this is not what Nyquist specified in his description of how to convert a sampled signal back to its analog counterpart. If you implement Nyquist's reconstruction criteria, you would get back the original 5.5 kHz sine wave exactly - as long as the sample rate was greater than 11 kHz -- 44 kHz would do, but so would 11.0001 kHz

[genericperson]

44.1k/2=22.05khz.

so which sound do you record that has meaningful overtones above 22khz?

fwiw, i think 24/48k strikes the best balance of fidelity and practicality (notwithstanding the problem of usually having to deliver the final product at a less robust spec).

[alphajerk]

lord this thread is so very flawed.... never mind BITS used. number of BITs determine RESOLUTION. more bits used does NOT improve fidelity and only minimally improves what little noise floor digital has [>-100db], which is far below most rooms noise floors, mic noise floors, pre/comps/eq's noise floors... and most definatley tapes noise floor.

[zemlin]

I don't spend a lot of time recording 11 KHz sine waves.

I record stuff with complex waveforms that seem to be reproduced just fine with a 44.1 KHz sampling rate.

[dasbin]

jtienhaara, you need to better understand the filtering and maths behind the DA stage of conversion.

Your illustration is what happens when you draw straight lines between samples. A DA doesn't do that. Or rather, it does, and then an analog lowpass filter smooths it out. You will have the original sine wave resulting in both cases, because there it is only mathematically possible that a single kind of wave can result from samples at those points.

Let me illustrate this. Figure A shows some sampled points taken at 44.1Khz sampling rate. What do you suppose the resulting waveform will be, post-DAC?
Figure B shows you what it will be, after filtering. This is a 22Khz sine wave.

A


B



The reason for all this is because those "stairstep" waves you showed contain tons of information above Nyquist. That is, a square wave is based on a fundamental sine frequency plus an infinite amount of sine overtones waves. Once you subtract those overtones, which is what the Nyquist lowpass filter in a DA does, you get the original waveform.

There are even some audiophiles out there who build DACs without any filtering at all. They rely on the natural properties of their speakers - which can't go much past 22Khz - and their ears - which can't go much past 18Khz in most cases - to do this exact same kind of filtering for them. And it works. You don't hear those square waves up there or the distortions they introduce. You hear the pure sine.

[jtienhaara]

SteveHalko and dasbin: Great posts! You're right, I know nothing about Nyquist filtering. You've got my curiosity piqued now though. How does one build such a filter? Any recommended readings / websites?

(For all the other folks who want to turn this into a debate about what sounds best / what sounds good enough / whatever: dfegad)

Cheers,

Johann

[dasbin]

Originally posted by jtienhaara
[B]How does one build such a filter?

It's just EQ, really. Plain and simple. Imagine taking a linear-phase digital EQ like Waves LinEQ and applying a very steep lowpass at 22Khz, and you've got the basic ideas. Although, the methodology is a bit different because this is happening at the time of conversion, so they have to think up interesting ways to do this. Oversampling is one.
You can do it in analog, too, but this is much more difficult and costly and usually less effective than an oversampling digital filter.

Any recommended readings / websites?

Yes. http://www.dbtechno.com/documents/Sampling_Theory.pdf

[Zooey]

Originally posted by norman_nomad
If boosting the fader is superfluous in a daw, then why bounce out as hot as one can?

I'm curious as well. I never raise the master fader above 0 dB for a bounce, but I pull it below 0 dB when necessary. Since I broke the habit of tracking hot, it's becoming less and less necessary...

One thing that still drives me nuts: who is telling the producers of loop libraries (especially 24 bit libraries) that we want our loops normalized to 0 dBfs?

[Brad Blackwood]

FWIW, you shouldn't use square waves when discussing sampling for two reasons:
1] nothing anything like a square wave exists naturally, and
2] a square wave is basically a fundamental with tons of harmonics.

Nyquist and Shannon were right. 44.1kHz sampling rate is more than enough to sample a 20kHz sin wave. If you don't know who these people are, do some research.

Bits do not equal resolution - they simply equal dynamic range. More bits = higher peak level available above noise floor.

Everyone should read this book at least once...

[pedalboy]

If boosting the fader is superfluous in a daw, then why bounce out as hot as one can?

There really is a reason, i'm not making this up. The reason is if you are recording a few files at 24 bits, and then you adjust the faders for mixing, your mix bus is actually 48 bits if you are on a protools tdm system. Other systems are similar, i think. So you DO have a little bit more "headroom" so to speak to work with, so take advantage of it. Because the mix bus has a higher bit depth than the output format, you aren't really going to lose anything boosting that fader up.

Oh, and Bob Katz is the MAN for this stuff.

[jtienhaara]

@dasbin: Great link! I'm looking forward to delving into that one. I was a little skeptical at first (it starts off from a marketing standpoint, which usually makes me cringe) but then it gets into math after the introduction so... Looks like a good read.

@Brad Blackwood: If you're referring to the square wave I drew above, that was my idea of sampling. The sampled signal was actually a sine wave, it just didn't show up because "Dia" didn't render it properly. Incidentally I've come across Shannon in Boolean algebra but haven't spent any time with he or Nyquist in the DSP domain. That book is on my Christmas list, thanks!

Cheers,

Johann

[norman_nomad]

Originally posted by pedalboy
So you DO have a little bit more "headroom" so to speak to work with, so take advantage of it. Because the mix bus has a higher bit depth than the output format, you aren't really going to lose anything boosting that fader up.

You also aren't gaining (no pun) anything, as far as I can see.

My understanding is that, although Protools has a 48 INTERNAL mixer, the master channel will always output (to your converters) either a user defined 24 bit/ 16 bit signal. For this reason it’s advisable to use dither on any channel in PT that's going to be bounced down. Even when bouncing 24 bit files during submixing, it's suggested to use a non-noise shaping 24 bit dither (if you have to dither a file more than once, make the last dither noise shaping) to help mask any low level truncation errors. You'll lose a few bits, but the cumulative results are less abrasive.

The 48 bit PT master fader has enough head room and FOOT room to accommodate a wide range of output levels, so as far as I can see, there is really no benefit nor drawback to changing the master faders final level during mixdown... just make sure it doesn't clip internally.

From the digidesign website:

Neat picture


*dasbin* Thanks for the pdf link! I'll be trying to decipher that article for the next couple of weeks!