Understanding Audio Dynamic Range / SNR (Part 1)

[MRC01]

The Loudness War DB uses DR14 ratings, which are based on the difference or ratio between the peak amplitude (usually 0 dB or just below) and the average/RMS amplitude.

To oversimplify: when listening to high DR14 rated music (say, DR14 > 12), the average levels are a lot quieter than peak. With low DR14 ratings (say, DR14 < 8), all the music is a big loud "wall of sound" that never gets quiet. That said, a low DR rating doesn't necessarily mean the music was dynamically compressed. Some kinds of music, like solo guitar or harpsichord, doesn't have much dynamic range. Other kinds of music, like Orff's Carmina Burana, has huge dynamic range.

 

[Francis Vaughan]

Does anyone understand what the dynamic range database is trying to portray?

Its very interesting. The numbers are basically arbitrary figures of merit. The derivation of the values is complex and a bit arbitrary. Values are generated by the DROffline program. The nicely written and thoughtful manual is here.

The program is intended to detect the application of compression in the recording chain - not to create a value inherent to the audio describing SNR or dynamic range as we are discussing here. It isn't a surrogate for signal to noise. It is intended as a tool in prosecuting the battle in the loudness wars.
There is no perfect DR value. It is context dependent, but a low value is generally a good indication that the recording has been slammed during post processing and won't be good. But low changes from genre to genre.

From the manual:

We can ́t point that out often enough: The aim of the DR toolset is to generate an easy to understand, whole number value, which describes the degree of dynamic reduction vs. the amount of inherent dynamics. It does this by focusing on the top 20% of loudness events, and counting the average of those 20% against peak amplitude. Designed as a motivation to back away from loudness war–driven mastering decisions, the DR measurement system is best suited to all modern and popular main-stream music genres.

The DR algorithm has been designed and crafted to deliver easy to digest measurement of hyper–compressed main stream music releases. The aim is to bring back, for general music re-leases, more dynamic contrast and listening pleasure with less fatigue. It was not meant to measure the dynamics of an a ca-pella Gregorian choir. It wouldn’t make sense as, due to the absence of transients, the DR meter would show misleading lower results. Frankly, we haven ́t found a satisfying solution to solve that issue yet, though there isn’t a huge demand.

 

[hestejoe]

How important is the SNR in a practical setting? As I understand, having a pair of kali LP-6 V2 active speaker set which I rarely listen to at a higher volume than 85 dB, everything that is balanced and above 95 SNR would extremely likely be more than enough? Or should I still aim for 106 given that I listen to CD material?

Edit: Also, if there ever was a follow-up video, it has eluded me.

 

[Cbdb2]

When did the definitions is S/N and dynamic range change? This is what they used to be, the way I learned them:

Moulton Laboratories :: About Acoustic and Microphone Levels: Levels Management I

Dynamic range is S/N plus headroom. The signal is at reference level (say -14dB below full scale) in a studio so you have 14db of headroom. So the dynamic range is 14db more than S/N. Which dosnt matter in home systems but is a big deal when your recording.

 

[Cbdb2]

How important is the SNR in a practical setting? As I understand, having a pair of kali LP-6 V2 active speaker set which I rarely listen to at a higher volume than 85 dB, everything that is balanced and above 95 SNR would extremely likely be more than enough? Or should I still aim for 106 given that I listen to CD material?

Edit: Also, if there ever was a follow-up video, it has eluded me.

A large reason for the loudness wars is the CAR. A large percentage of the population listens to more music while driving than anywhere else. In that noisey environment music needs to be compressed or you won't hear half of it. And of course producers want there songs loudest.

 

[RigorDude]

If there is no problem you can't fix it


Can you even plug Australian plug in the other way around?

We have this stupid plug:
View attachment 115663

So there is no "right" or "wrong polarity" in most of Europe:
Equipment is build so it don't matter if phase and neutral is swapped.

But sometimes there can be "compensating currents" Massive is very relative.
maybe 0,5mA is typical.

This is an other excellent read about the topic:
http://web.mit.edu/jhawk/tmp/p/EST016_Ground_Loops_handout.pdf

The link at the bottom of your post looks interesting — especially as the piece is from MIT! Alas, the link goes to a 404 page. Perhaps the document has been renamed?

 

[RigorDude]

I second this question. All noise is definitely not created equal. To say we want +10db over a bitrate's dynamic range is a great rule for modern equipment, but doesn't really address a few things.
With some equipment, you might have some mains related noise. Turns out that a -80 dBV peak at 60Hz is a heck of a lot better than that same level at 2 kHz

Amir, could you address the advantages/disadvantages of applying A-weighting? Or perhaps in your reviews you could overlay the fletcher munson curve to show relative impact of noise to our (human) ears? Would love to hear some thoughts.

I strongly agree. “Floor” is a misleading word in this context. The “noise floor” is often not a floor so much as the top of a steeple, a specific frequency or handful of frequencies at a level well above the rest of the spectrum. Such noises can be heard past if one is not trying to notice them. Listening to music isn’t the same as vigilantly listening for a snapping twig. On the other hand, a music signal that is “below” (at a lower level than) the noise floor/steeple doesn’t necessarily disappear.

 

[Angsty]

The link at the bottom of your post looks interesting — especially as the piece is from MIT! Alas, the link goes to a 404 page. Perhaps the document has been renamed?

I think this is the same document referenced in the link. Used at MIT, but sourced from Jensen Transformers.

https://www.jensen-transformers.com/wp-content/uploads/2014/08/generic-seminar.pdf

 

[Angsty]

I love this quote from the paper:

"Expensive and exotic cables, even if double or triple shielded, made of 100% pure un-obtainium, and hand-made by a team of virgins, will have NO significant effect on hum and buzz problems!"

The author goes on to cite the applicability of Belden 8241F. "In engineering terms, a high-performance cable for unbalanced audio should have low capacitance and very low shield resistance."

Notably, Blue Jeans LC-1 (custom sourced from Belden) has superior engineering properties with lower capacitance and lower shield resistance.

 

[RigorDude]

I think this is the same document referenced in the link. Used at MIT, but sourced from Jensen Transformers.

https://www.jensen-transformers.com/wp-content/uploads/2014/08/generic-seminar.pdf

Thanks - very nice to have on hand for the next hum event.

 

[tuga]

If there is no problem you can't fix it


Can you even plug Australian plug in the other way around?

We have this stupid plug:
View attachment 115663

So there is no "right" or "wrong polarity" in most of Europe:
Equipment is build so it don't matter if phase and neutral is swapped.

But sometimes there can be "compensating currents" Massive is very relative.
maybe 0,5mA is typical.

This is an other excellent read about the topic:
http://web.mit.edu/jhawk/tmp/p/EST016_Ground_Loops_handout.pdf

If you use double insulated equipment then you will get a different reading from a tester depending on how you plug it.

 

[Lambda]

The link at the bottom of your post looks interesting — especially as the piece is from MIT! Alas, the link goes to a 404 page. Perhaps the document has been renamed?

https://apapes.physics.uoc.gr/docs/articles/EST016_Ground_Loops_handout.pdf

 

[fin]

Hi, I have a question regarding dynamic range but may not worth a new thread to discuss as it seems to be quite basic. I also didn’t find other thread discussing similar stuff. TL;DR: Is the dynamic range enhancer techniques considered cheating?

It all start from the observation that the measured dynamic range of many audio interfaces can’t quite match the number on the spec paper. For example there’s one audio interface claimed to reach 122dbA in their A to D converter but the measurement showed only 113dbA. Look at the data sheet of PCM1840 I get this:

So basically it switched the ADC to low signal mode, and adjusted the op_amp to lower the noise floor, when the input signal level is low. The PCM1840 can thus increase 10db of dynamic range by enabling DRE, jumping from 113db to 123db, which will be reflected in measurements using RMAA but AP might be easy to realize the trick. There’s a clearer explanation in TLV320ADC5140’s manual.

I'm confused by this as I would assume the circuit shouldn’t optimize in low signal mode, as the idea of testing noise floor with -60db signal is to avoid some mute or gate cheating in measuring dynamic range, while an optimization involving using the amp to lower noise floor is kind of crossing the line and at least it can’t change anything regarding SNR. Is my understanding correct? Thank you!

 

[Cbdb2]

"Dynamic range enhancement" is an effects box. Its an expander, the opposite of a compressor. It makes loud parts louder and quiet parts quieter, thus more dynamics in the signal. There supposed to reverse the dynamic compression done in the mix and mastering. They rarely do.

The only time this will work properly is if the ALL the parameters, time constants (attack and decay), treshold and ratio are the same as when compressed. (this is done in compander noise reduction and it works great). Theres also feedforward and feedback compressors that have different output even if all the parameters are set the same. https://en.wikipedia.org/wiki/Dynamic_range_compression
Most things are recorded with each instrument compressed, than the drum sub mix (bg vocal sub mix etc) is compressed, and then the final mix is compressed and than the mastering engineer compresses it again.
Hows does the expander know the parameters of even one of these stages?

 

[MC_RME]

@Cbdb2 : DRE in this chip is a means to lower the noise floor of the chip itself (!), at lower levels. The original audio level is not touched at all.

 

[Cbdb2]

@Cbdb2 : DRE in this chip is a means to lower the noise floor of the chip itself (!), at lower levels. The original audio level is not touched at all.

"the DRE monitors the incoming signal amplitude and accordingly adjusts the internal DRE AMPLIFIER GAIN automatically." And the last sentence.

 

[fin]

So I also found the related patent by Texas Instruments here

It seem to be a very detailed explaination of dynamic range enhancer. And it's not a new idea as several earlier patents also explored this direction. The motivation is far-field voice pickup, and not to let signal-to-noise ratio (SNR) limit the dynamic range. With a customer-level audio knowledge I could not get how it worked from just a glance.

At least we have the measurements in TLV32ADC5140's manual and here it is:

Notice how noise floor is 12 db lower than when DRE is disabled, and the signal stays at -60db (that might have something to do with the later digital filter part about scale factors, explained in the patent)

If we just consider very low level recording then sure it's 12db of free dynamic range. But signal-to-noise ratio is not changed at all so the use case seems to be very limited. Like @Cbdb2 explained that it works like an expander to expand the DNR.

Look at the explaination provided by AP about why dynamic range is measured using 0 dbFS signal vs. -60 dbFS noise floor:

· In both ADCs and DACs, “idle tones” can be produced within the converter in the absence of applied signal. In the method here, a low-level tone is applied to the converter to avoid production of idle channel noise. The low-level tone is removed by a notch filter before measurement.

· In some DACs, the output of the device is switched off when there is no signal, providing an unrealistically quiet measurement. The low-level tone (again, notched out before measurement) defeats this muting mechanism.

To me DRE looks like a fancy "switch off" mode, although technically we can still use the device for "far field pick-up" but for most recording scenarios there's a 12 db differences between SNR and DNR, which can sometimes confuse customers. Looks it's more of a designed overfitting technique to optimize the score based on AES17 dynamic range measurement standard. It is now also widely used as a marketing tool by manufacturers and people are just crazy about the excellent "123dbA dynamic range" without knowing whether that in most cases that cannot bring more headroom in their recording work.

By the way @MC_RME I appreciate RME for providing only the SNR number and they are not just number on spec but also what people can get in real use cases. Not good for marketing though.

 

[MC_RME]

The way DRE works here is clearly explained in the PCM1980 data sheet. At some level point (threshold) an analog preamp amplifies the analog input signal. At the same time the same amount of level is subtracted from the digital level. So the chips analog input to digital output ratio is 1:1, but the resolution in the lower range is improved. The audio signal is not changed at all (except for unavoidable switching artefacts during transition), this is NOT an expander technique mangling the audio signal with hidden muting or dynamic range extension as known from other effect processors.

 

[pma]

A large reason for the loudness wars is the CAR. A large percentage of the population listens to more music while driving than anywhere else. In that noisey environment music needs to be compressed or you won't hear half of it. And of course producers want there songs loudest.

Yes, car, subway, plane, street. Or PC + desktop. The niche of us (1% of listeners?) suffer from hypercompression. Still some classical music is recorded well, I would recommend

Digital Concert Hall

The Digital Concert Hall is the online concert hall of the Berliner Philharmoniker. Here you can see the orchestra’s concerts live or as recordings in the archive.

www.digitalconcerthall.com

 

[oivavoi]

Yes, car, subway, plane, street. Or PC + desktop. The niche of us (1% of listeners?) suffer from hypercompression. Still some classical music is recorded well, I would recommend

Digital Concert Hall

The Digital Concert Hall is the online concert hall of the Berliner Philharmoniker. Here you can see the orchestra’s concerts live or as recordings in the archive.

www.digitalconcerthall.com

I'm wondering whether the growth in noise-cancelling headphones will lead to a demand for higher dynamic range? Those headphones create artificially a low-noise environment, so it's not as important for the music to cut through all that environmental noise anymore. From what I see around me, noise-cancelling headphones have started to become really common on subways and trains etc.

Music in private cars remains a different story of course (the impact on the recording industry is just another reason why cars are bad!).