Loudness War, the limit of the 0dB in digital or Analog better than Digital?

[Jean.Francois]

Hello,

When we compare different editions of the same album, we regularly notice that the digital versions are impacted by the Loudness War, i.e. by the use of a Brickwall type limiter to increase the average sound level while reducing the peaks, or less cleanly with clipping. This is due to the fact that you can't go beyond 0 dB in digital!

Today, I propose to measure the impact of clipping and limiting and to compare this impact with what can be obtained with analog formats.

To measure this impact, we will use a reference signal at 1 kHz.


The distortion will come from the DAC, from 0% to 0 dB. With the advantage that the digital copy does not introduce distortion!

To implement the clipping, we add 3 dB to the signal.


We then observe that the distortion is now present with a rate of more than 9% for an increase of 2 dB




We test the limitation with a Brickwall limiter, we add 3 dB:


We then observe that the distortion is also present with a rate of more than 3% for an increase of 3 dB with the limiter, so less important than with clipping:


It is interesting to compare these rates with analog sources like vinyl:


Distortion of more than 1 % at 0 dB

Or the cassette :


Distortion of 1.6% at 0 dB.

In conclusion, the digital is indeed very good in term of fidelity of the signal (I do not speak about listening) and superior to the analog formats, but, with only one condition, it is not to seek to have

clipping or the limitation of dynamics of the war of the volume, in which case, the analog can have less distortion!
For more information and comparison, you can find more curves and comparison with tape recorders here.

Enjoy listening

 

[LTig]

You cannot compare like this. Analog clipping depends on the type of analog device. A tape drive for example will clip much softer than a solid state preamplifier.

 

[Jean.Francois]

Indeed, this is the advantage of analog, in the complete test I put the example of 2 tape players. But the objective is to show that digital is very sensitive if you play with the limits, and that even the brickwall limiter strongly damages the signal.

 

[abdo123]

By the way it feels like you’re under the impression that analogue releases in 2023 are getting their own mixes and / or masters but I don’t think that’s the case at all.

 

[AnalogSteph]

Brickwall limiters take advantage of the fact that music doesn't have the crest factor of a sine wave. They're only supposed to act briefly on transients every once in a while. The things are remarkably effective these days, too (audible artifacts were much reduced once using oversampling limiters became widespread practice about a decade ago).

But yes, bunching everything up at the top of the available dynamic range is silly. These days you might just as well just compress as needed, normalize levels so that true peak is about -0.5 to -1 dBFS or so and let playback volume normalization (e.g. ReplayGain and its counterparts in streaming services) take care of the rest. A lot of the dynamic range that would be available even at CD quality goes completely unused in practice... and lossy formats tend to accept 24-bit input just fine.

Mind you, the loudness war never was about being smart... just about being louder than the rest. It's about time to put a stake in it and bury it. Even CD releases (the few that are still being bought, sigh) no longer need this nonsense.

 

[DVDdoug]

The loudness wars actually started in the vinyl days but they didn't have the modern digital "weapons".

There many variations of limiters. For example, the Audacity limiter uses look-ahead so it doesn't change the wave shape. Analog limiters "soft clip".

All "Loudness War" recordings are highly compressed & limited but they aren't all hard-clipped. It may be clipped as the last step but by that time most of the "damage" has already been done.

Most digital recordings and releases are not clipped.

Virtually all commercial recordings (analog and digital) are dynamically compressed. (As you may know, limiting is a kind of dynamic compression.)

When solid state electronics clip they tend "hard clip" similarly to digital. But you can have plenty of headroom in analog electronics (depending on amplifier power, etc.).

Analog tape soft clips, depending on the tape formulation and machine calibration. 0dB is somewhat arbitrary and it can go over 0dB. Analog tape has less dynamic range than digital (the resolution and dynamic range are limited by tape noise) so you generally wanted to record "hot" and maybe occasionally go "into the red".

Tape clipping is further "smoothed" by the NAB tap equalization.

If you go "too loud" with vinyl, the stylus mis-tracks which is worse than clipping.

And the record/playback process on vinyl changes the wave shape, partially "hiding" clipping (without removing the distortion components). Changes in the waveshape also tend to make some peaks higher and some peaks lower (without affecting the sound of the dynamics) so people using the crest factor (peak to RMS ratio) to measure "dynamic range" think the vinyl is less compressed than the digital version. (Sometimes the vinyl is less compressed but not always.)

MP3 compression (file compression unrelated to dynamic compression) behaves similarly, changing the wave shape and making some peaks higher and increasing the crest factor.

Since digital hard clips the main goal (during recording) is to always avoid clipping by leaving plenty of headroom. But since is has so much dynamic range there is no need to record "hot".

And of course with digital you can boost the volume after recording with no quality loss (assuming you avoid clipping). With analog tape there is loss with every generation, whether you adjust the volume or not.

 

[kemmler3D]

What limiter did you use for this? I've used quite a few and I can't recall setting the attack/release fast enough to make 1Khz look like that...

e: just tried this with the George Yohng W1 limiter, an emulation of the infamous Waves L1 limiter, and I take it back, I was able to create sidebands by setting it to -3dB threshold and a short release time. However, a longer release time made them go away. Interesting. Might actually do the FFT later I guess.

OP still interested in what limiter you used for this test.

 

[Cbdb2]

Your comparing apples to oranges. The apples are dbfs the oranges 0vu. 0db means nothing. 0vu is the studio standard level that you try to record at. This is the sweet spot between noise and head room before distortion.. Digitals equivalent to 0vu is supposed to be -18dbfs giving it 18db of headroom before clipping. Try driving an analogue tape at +18vu and measure the distortion. Now try that with vinyl and the stylus won't even stay in the groove. The problem is the loudness wars have eaten up that 18db. In the studio you hit your digital recorders with about 0vu (-18dbfs more or less) same as the analog tapes they were replacing. Clipping should not happen.

 

[Cbdb2]

The loudness wars actually started in the vinyl days but they didn't have the modern digital "weapons".

There many variations of limiters. For example, the Audacity limiter uses look-ahead so it doesn't change the wave shape. Analog limiters "soft clip".

Anything that changes level dynamically will change the wave shape. A limiter flattens anything over the threshold, lots of signal, lots of flattening. They were meant to catch the odd peak not flatten the whole waveform the way these "mastering" engineers use them today. The advantage of look ahead is not missing fast peaks on a slower attack setting (if its attack is fast enough look ahead doesn't make much difference so its more usefull on compressors.) Major distortion in comp/limiters comes from fast attack release times where its changing levels faster than the signal voltage not the signal envelope. This is the crux, when the comp is following the signal instead of the envelope you get lots of distortion. A compressor is supposed to looks ahead, (unless its a feedback compressor instead of a feedforward comp) see how far the level is going to be above threshold and turn the level down for a while then turn it back up when signal drops again. I'm sure you can see how this gets complicated on a music signal and why there are so many different compressors. With fast time constants bass gets distorted the most. Bass players will actually use compressors to add distortion.

 

[Anonamemouse]

This is due to the fact that you can't go beyond 0 dB in digital!

It doesn't get more incorrect than this.
Yes, 0 dB is the clipping point. But not only in digital, it's the clipping point with all forms of sound recording. It 0 dB for a reason.

 

[Jean.Francois]

What limiter did you use for this? I've used quite a few and I can't recall setting the attack/release fast enough to make 1Khz look like that...

e: just tried this with the George Yohng W1 limiter, an emulation of the infamous Waves L1 limiter, and I take it back, I was able to create sidebands by setting it to -3dB threshold and a short release time. However, a longer release time made them go away. Interesting. Might actually do the FFT later I guess.

OP still interested in what limiter you used for this test.

The limiter used is LoudMax version 1.4.
The objective is to show the fragility of the digital quality if we try to be as close as possible to 0 dB digital.
We find clipping on some albums as for example in the album Dangerous on the remastered versions.
The brickwall limiter allows to avoid clipping, but introduces a distortion of the signal compared to the original signal. The brickwall limiter is used to raise the average level so that the sound is louder, and often very loudly when you see tracks with a DR3!
This Loundness War degrades the original signal by introducing a distortion that is not negligible, while the digital allows to have a very very low distortion.
The comparison with analog supports is to show that the loudness war can give worse results on the digital support because of the applied treatments and that it would be necessary to stop wanting to sound louder and to respect the work of mixing which is carried out in studio.
Today, there is hope with the Dolby Digital and Sony 360RA formats.

 

[Jean.Francois]

It doesn't get more incorrect than this.
Yes, 0 dB is the clipping point. But not only in digital, it's the clipping point with all forms of sound recording. It 0 dB for a reason.

Indeed, I wanted to say that for digital, there is no possibility to go beyond 0dB by construction with the 16, 24 or 32 bits fixed formats.

 

[Jean.Francois]

Your comparing apples to oranges. The apples are dbfs the oranges 0vu. 0db means nothing. 0vu is the studio standard level that you try to record at. This is the sweet spot between noise and head room before distortion.. Digitals equivalent to 0vu is supposed to be -18dbfs giving it 18db of headroom before clipping. Try driving an analogue tape at +18vu and measure the distortion. Now try that with vinyl and the stylus won't even stay in the groove. The problem is the loudness wars have eaten up that 18db. In the studio you hit your digital recorders with about 0vu (-18dbfs more or less) same as the analog tapes they were replacing. Clipping should not happen.

Indeed, the -18 dbfs is set for the 0 "dB vu" for the passage between digital and analog. But, in practice, in digital, except at the output of the mix where we respect the dynamic range, after mastering the goal to sound loud is to be as close to 0 dbfs with the highest possible LUFS. This 0 "dB seen" is not used for the digital formats, too bad, because it allows to keep the dynamics as you say.

 

[Cbdb2]

It doesn't get more incorrect than this.
Yes, 0 dB is the clipping point. But not only in digital, it's the clipping point with all forms of sound recording. It 0 dB for a reason.

0db what? Is that dbm, dbu, dbVU, dbfs, dbspl? 0db means nothing. Its actually multiplying something by 1.
Did you read my post above about the difference between analog and digital "dbs"?
A pro analog recorder puts out 1.22 volts at 0VU which is +4dbm. A digital recorder puts out 1.22volts (0Vu) at -18dbfs. (most digital recorders let you set this level, most studios use between -20dbfs to -16dbfs for more or less headroom) .

 

[kemmler3D]

I'm just going to go ahead and say it - I think the loudness war officially ended when the streaming services standardized on -14dB LUFS. I find this post a bit anachronistic and I'm also not sure what Dolby or Sony formats have to do with it. You can put over-compressed music into any digital container you please, breaking it into stems ("objects") doesn't necessarily help, and the last thing (IMO) the world needs is more proprietary audio formats.

 

[fieldcar]

I think the loudness war officially ended when the streaming services standardized on -14dB LUFS.

I was going to mention the same thing. Everyone masters for streaming now. The general consensus is -3.0dBfs, -1.0dBfs, or -0.5dBfs short term/instantaneous peaks with -14dB LUFS and beyond. Some producers and mastering engineers still clip their peaks, but at least they aren't brickwall compressing the crap out of mixes quite as bad anymore.

 

[Jean.Francois]

I'm just going to go ahead and say it - I think the loudness war officially ended when the streaming services standardized on -14dB LUFS. I find this post a bit anachronistic and I'm also not sure what Dolby or Sony formats have to do with it. You can put over-compressed music into any digital container you please, breaking it into stems ("objects") doesn't necessarily help, and the last thing (IMO) the world needs is more proprietary audio formats.

I wish the loudness war was no longer present, because I wouldn't have to notice that the music I listen to on streaming or cd is compressed in dynamics, and I wouldn't have to do tests and articles like this one. But this is not the case!
The streaming services propose a normalization to -14 LUFS, but do not impose anything. So, nothing really changes, the albums are still limited in dynamic range as regularly shown by the reviews I do comparing the different editions of the same album.
It will certainly take many years before this changes, because an album must always be released to sound strong.
Today, Atmos and 360RA mixes start from the stems from the mix that are not compressed in dynamics, which explains the difference between the dynamics in stereo and in Atmos (or 36RA). The limiter is used during the stereo mastering phase to raise the overall level.

 

[kemmler3D]

start from the stems from the mix that are not compressed in dynamics, which explains the difference between the dynamics in stereo and in Atmos (or 36RA). The limiter is used during the stereo mastering phase to raise the overall level.

I would be SHOCKED if the stems all had no dynamic compression, first of all. It's common practice to put compression on almost anything in pop / electronic mixing.

Secondly, "send uncompressed stems to a master limiter to raise the overall level" is the essence of the loudness war in the first place, so I don't see that it's any different in principle. They probably have a milder threshold for the limiter, but it doesn't necessarily help to have the limiter in the listener's home vs. in the studio.

There is no final cure for irresponsible/bad mixing other than commercial failure of irresponsible mixes. At least the streaming platforms and legislators are doing their part.

 

[Jean.Francois]

I would be SHOCKED if the stems all had no dynamic compression, first of all. It's common practice to put compression on almost anything in pop / electronic mixing.

Secondly, "send uncompressed stems to a master limiter to raise the overall level" is the essence of the loudness war in the first place, so I don't see that it's any different in principle. They probably have a milder threshold for the limiter, but it doesn't necessarily help to have the limiter in the listener's home vs. in the studio.

There is no final cure for irresponsible/bad mixing other than commercial failure of irresponsible mixes. At least the streaming platforms and legislators are doing their part.

On the stems, there can be a little compression for the mix depending on the type of instruments and the expected rendering, but it is not there that we increase the level with a limiter of the Brickwall type for the phenomenon of volume war.
The dynamic limitation is done at the mastering stage.
I exchanged with sound engineers who confirmed this way of working by specifying that it is for this reason that Dolby Atmos for the 360RA remains more dynamic because they start again from the stems and that there is no mastering which is made to compress the dynamics on the spatial mixing object, at least for the moment.

The change might take some time, because it doesn't stop people from listening to music, and Apple for example is now focusing on the spatial format.

 

[fieldcar]

You know what @Jean.Francois ... I'm going to do a 180, and agree with you that clipping is still an issue when it really shouldn't be. The mastering engineers just can't help themselves and are probably ignoring the normalization standards and acting like they are mastering for FM radio or CD's to sound louder. The worst sin is that EVERY SINGLE DAW software has moved to FP32 internally for the mixdown to the master, and only at the output of the MASTER chain, does it flip to INT## with a 0dBfs ceiling.

I think the only way to fix this wholly is to move over to a floating point audio format like WavPack with FP32, as there is essentially no such thing as a ceiling.

From your sample: https://magicvinyldigital.net/2022/...-review-vinyl-qobuz-hi-res-tidal-dolby-atmos/

FP32 explained: