Mo Facta
Farts of Nature
DISCLAIMER: I DON'T HAVE ALL THE ANSWERS. THIS IS A DISCUSSION THREAD, SO RELAX!
There have been so many threads about this lately, I thought I might start a thread where we OBJECTIVELY discuss proper level practices and dispel the myths.
I hear a lot of catch-phrasey lingo going on and also much confusion about how signal levels relate in the digital vs. analogue domains. 32-bit floating point seems to be a pivotal argument as well but there are actually so many points of potential errors to creep in because, at the end of the day, a "100% digital system" is never really that because analogue components are still required at the front and back end of the system. Also, because there are many points of signal being handed off from one point to another and many correlations and conversions (i.e. voltage to bits, VU to dBu, dBu to dBfs) involved, it all adds up to be this nebulous cloud of confusion with many opinions.
So, what I would like to do is put down my own conclusions in this OP and then we can discuss it, hopefully with some professionalism and dignity. Here we go:
1. The 32-bit floating point argument.
This is the usual argument for, as Bobby put it, "being sloppy with your levels" when mixing ITB. Granted, when you turn down the master fader in a 32-bit FP environment you effectively turn down the OUTPUT of each track being fed into the master bus. What this DOESN'T do is remedy any slammed levels you may have on channel input and within plugins. Now, Ethan Winer has lambasted me for making comments about plugins and their robustness to these levels so we'll leave that argument for now. Let's just say that all plugins handle values over +1/-1 (digital full scale) with equal robustness. But what about emulations? Many emulation plugins model analogue circuitry and therefore have a non-linear response, depending on the level they receive. Given the variety of plugins we use and given that SOME plugins MIGHT NOT be robust to these slammed levels, it just makes sense to me to keep your levels at a conservative. I would way rather have piece of mind than hear distortion later on when I'm balls deep into a mix and have to make adjustments so late in the game.
And then there's issue of what happens to these slammed levels that the 32-bit FP pundits say is OK when it hits the DA converter...
2. Intersample distortion and DA output distortion
So now you've got these slammed levels ITB, your plugins are slamming and everything is over full scale, which, according to some, is OK because all you have to do is turn down the master bus. Barring that, you'll most definitely get clipping and distortion at the DA if you don't. That's an easy one and it's easy to hear when the analogue portion of the DA is distorting so we'll leave that as a given. What's not so easy to detect is intersample distortion because it is a peak level phenomenon. This is one thing that often gets overlooked because many people don't understand it and therefore disregard it. Traditional peak meters in a DAW will not pick up intersample peaks and these can be 6dB higher than the reported level. This has to do with how the reconstruction filter will reproduce an analogue signal from stored samples. It is in fact very common for the reconstructed waveform to have peaks that are higher than what is reported on a sample by sample basis. This will cause distortion at the DA.
My point in all of this is that to maintain a slammed gain structure becomes a game of patching holes where you have to implement all of these contingencies to remedy a problem that could have been avoided by simply keeping your levels in your equipment's optimal range, which, of course, is dictated by the analogue components in the chain, their calibration to digital, and voltage tolerances.
3. Input Levels, unity gain, and equal perceived loudness A/Bing
This, for me, is where it starts. At this point in my career it seems silly to start a project with slammed input levels. My philosophy on this is simply to give my input signals plenty of headroom on input and this usual corresponds to peaks around the -16 to -12dBfs mark. This gives me a median of about 14dB of headroom, which is plenty for most applications. To me, it seems sensible to SET YOUR GAIN STRUCTURE FROM THE BEGINNING. That's how I work. I set my gain structure when recording, and then when I process something it's ALWAYS A/B'ed at THE SAME PERCEIVED LOUDNESS. How else are you going to know what actually sounds better? Sometimes I feel like this level dilemma we have is because people fool themselves into thinking something is better when it's just louder.
Another thing is that when you slam your levels on input you might pick up distortion from the preamp or at the analogue input section of a converter. Some preamps do better than others at this but I would say it's a pretty safe bet that if you're near clip point on your preamp and then feed it into a converter that might have lower headroom than the preamp, you'll accumulate distortion from both. Distortion is not always immediately detectable and might be subtle enough to creep through without detection. In fact, you may eventually get into the mix and then find that things sound harsh and it's difficult to know why. The other problem is that once analogue distortion has been captured, there's not much you can do about it. It's there to stay and WILL accumulate if you have made this mistake on multiple tracks. I don't care what anyone says, that's what I have heard myself and I'm sticking to it.
4. Analogue components are everywhere.
There is no system that is 100% digital and therefore the fact that 32-bit floating point precision is robust to slammed levels is irrelevant to me. Analogue gear has a dBu scale and clip point for a reason. It all relates to voltage and there are optimal operating ranges for all analogue components. Also, different pieces of gear have different headroom ratings. You may have a preamp that has a max output level of +27dBu and if you're going into a converter that has a max input level of +18dBu, you're going to pick up distortion if you slam the preamp. Analogue hardware has high headroom for a reason and that is because the designers know that analogue components can saturate and produce distortion way before clip point. Ask any professional mastering engineer about the headroom of their outboard gear. Many times they have massive headroom. And why? To supply clean, distortionless sound at all levels. Here are some typical max output levels of some selected hardware:
Manley Massive Passive: +36dBv
Manley Vari-Mu: +30dBv
Pendulum 6386: +27dBu
Dangerous BAX: +28dBu
We can go on. Problem is, most gear nowadays does not go much above +20dBu and since these circuits may or may not saturate or distort at high levels, it's a hit-or-miss situation whether or not we're going to get optimal performance out of them. See my point? Bob Katz reckons that given this slide in quality we've experienced over the years, it's probably better to use the -10dBm consumer standard to compensate for the lost headroom we no longer enjoy and for better quality audio. I'm just not sold that components these days are giving us similar performance and are impervious to distortion at high levels. On the contrary, companies have actively sought to find cheaper and cheaper components to justify their mass manufacturing purposes and mass marketing hype and there's no fucking way that I buy for one second that it doesn't make a difference. Ultimately, you pay for what you get.
Please feel free to challenge what I'm saying or add constructively to the conversation.
Let's just try and keep this a BIT civil, hey guys?
Cheers
There have been so many threads about this lately, I thought I might start a thread where we OBJECTIVELY discuss proper level practices and dispel the myths.
I hear a lot of catch-phrasey lingo going on and also much confusion about how signal levels relate in the digital vs. analogue domains. 32-bit floating point seems to be a pivotal argument as well but there are actually so many points of potential errors to creep in because, at the end of the day, a "100% digital system" is never really that because analogue components are still required at the front and back end of the system. Also, because there are many points of signal being handed off from one point to another and many correlations and conversions (i.e. voltage to bits, VU to dBu, dBu to dBfs) involved, it all adds up to be this nebulous cloud of confusion with many opinions.
So, what I would like to do is put down my own conclusions in this OP and then we can discuss it, hopefully with some professionalism and dignity. Here we go:
1. The 32-bit floating point argument.
This is the usual argument for, as Bobby put it, "being sloppy with your levels" when mixing ITB. Granted, when you turn down the master fader in a 32-bit FP environment you effectively turn down the OUTPUT of each track being fed into the master bus. What this DOESN'T do is remedy any slammed levels you may have on channel input and within plugins. Now, Ethan Winer has lambasted me for making comments about plugins and their robustness to these levels so we'll leave that argument for now. Let's just say that all plugins handle values over +1/-1 (digital full scale) with equal robustness. But what about emulations? Many emulation plugins model analogue circuitry and therefore have a non-linear response, depending on the level they receive. Given the variety of plugins we use and given that SOME plugins MIGHT NOT be robust to these slammed levels, it just makes sense to me to keep your levels at a conservative. I would way rather have piece of mind than hear distortion later on when I'm balls deep into a mix and have to make adjustments so late in the game.
And then there's issue of what happens to these slammed levels that the 32-bit FP pundits say is OK when it hits the DA converter...
2. Intersample distortion and DA output distortion
So now you've got these slammed levels ITB, your plugins are slamming and everything is over full scale, which, according to some, is OK because all you have to do is turn down the master bus. Barring that, you'll most definitely get clipping and distortion at the DA if you don't. That's an easy one and it's easy to hear when the analogue portion of the DA is distorting so we'll leave that as a given. What's not so easy to detect is intersample distortion because it is a peak level phenomenon. This is one thing that often gets overlooked because many people don't understand it and therefore disregard it. Traditional peak meters in a DAW will not pick up intersample peaks and these can be 6dB higher than the reported level. This has to do with how the reconstruction filter will reproduce an analogue signal from stored samples. It is in fact very common for the reconstructed waveform to have peaks that are higher than what is reported on a sample by sample basis. This will cause distortion at the DA.
My point in all of this is that to maintain a slammed gain structure becomes a game of patching holes where you have to implement all of these contingencies to remedy a problem that could have been avoided by simply keeping your levels in your equipment's optimal range, which, of course, is dictated by the analogue components in the chain, their calibration to digital, and voltage tolerances.
3. Input Levels, unity gain, and equal perceived loudness A/Bing
This, for me, is where it starts. At this point in my career it seems silly to start a project with slammed input levels. My philosophy on this is simply to give my input signals plenty of headroom on input and this usual corresponds to peaks around the -16 to -12dBfs mark. This gives me a median of about 14dB of headroom, which is plenty for most applications. To me, it seems sensible to SET YOUR GAIN STRUCTURE FROM THE BEGINNING. That's how I work. I set my gain structure when recording, and then when I process something it's ALWAYS A/B'ed at THE SAME PERCEIVED LOUDNESS. How else are you going to know what actually sounds better? Sometimes I feel like this level dilemma we have is because people fool themselves into thinking something is better when it's just louder.
Another thing is that when you slam your levels on input you might pick up distortion from the preamp or at the analogue input section of a converter. Some preamps do better than others at this but I would say it's a pretty safe bet that if you're near clip point on your preamp and then feed it into a converter that might have lower headroom than the preamp, you'll accumulate distortion from both. Distortion is not always immediately detectable and might be subtle enough to creep through without detection. In fact, you may eventually get into the mix and then find that things sound harsh and it's difficult to know why. The other problem is that once analogue distortion has been captured, there's not much you can do about it. It's there to stay and WILL accumulate if you have made this mistake on multiple tracks. I don't care what anyone says, that's what I have heard myself and I'm sticking to it.
4. Analogue components are everywhere.
There is no system that is 100% digital and therefore the fact that 32-bit floating point precision is robust to slammed levels is irrelevant to me. Analogue gear has a dBu scale and clip point for a reason. It all relates to voltage and there are optimal operating ranges for all analogue components. Also, different pieces of gear have different headroom ratings. You may have a preamp that has a max output level of +27dBu and if you're going into a converter that has a max input level of +18dBu, you're going to pick up distortion if you slam the preamp. Analogue hardware has high headroom for a reason and that is because the designers know that analogue components can saturate and produce distortion way before clip point. Ask any professional mastering engineer about the headroom of their outboard gear. Many times they have massive headroom. And why? To supply clean, distortionless sound at all levels. Here are some typical max output levels of some selected hardware:
Manley Massive Passive: +36dBv
Manley Vari-Mu: +30dBv
Pendulum 6386: +27dBu
Dangerous BAX: +28dBu
We can go on. Problem is, most gear nowadays does not go much above +20dBu and since these circuits may or may not saturate or distort at high levels, it's a hit-or-miss situation whether or not we're going to get optimal performance out of them. See my point? Bob Katz reckons that given this slide in quality we've experienced over the years, it's probably better to use the -10dBm consumer standard to compensate for the lost headroom we no longer enjoy and for better quality audio. I'm just not sold that components these days are giving us similar performance and are impervious to distortion at high levels. On the contrary, companies have actively sought to find cheaper and cheaper components to justify their mass manufacturing purposes and mass marketing hype and there's no fucking way that I buy for one second that it doesn't make a difference. Ultimately, you pay for what you get.
Please feel free to challenge what I'm saying or add constructively to the conversation.
Let's just try and keep this a BIT civil, hey guys?
Cheers