How much dynamic range is possible in a simple stereo recording?

[Miska]

46 minutes ago, STC said:

Room noise floor is around 40dB. So far the loudest in listening room at insane level the peak hit 113dB. On average 70dB is more than enough. Yes, for playback 16bit is enough but for recording I would not go for anything less than 24bit.

 

You have quite loud room! I've measured around 30 dB!

[Miska]

4 hours ago, esldude said:

With two microphones and a stereo recording how much dynamic range is actually possible? We'll assume using more than two microphones is a treasonous perversion of the stereo medium. 

 

About 18 - 20 bit worth.

 

4 hours ago, esldude said:

Is dithered 16 bit digital all anyone needs for distribution of recordings? If not why not?

 

I think it would be good to preserve original dynamic range of the recordings.

 

4 hours ago, esldude said:

And yes be prepared to defend your opinions.

 

I picked two hires recordings at random. And for example for these cases 16-bit resolution limits dynamic range compared to original.

 

Here's 2L-111 original 24-bit DXD:

Same dithered to 16-bit:

Keeping the sample rate and using noise shaping the 16-bit would preserve all information though.

 

One DSD256 recording, converted to 88.2 kHz 24-bit PCM:

Same dithered to 16-bit:

 

 

It is not very straightforward to say whether 16-bit is enough because it also depends on sampling rate if you use noise shaping. At high enough sampling rate 1-bit is enough.

[Miska]

5 hours ago, esldude said:

What range do you have set for the spectrogram here? As in how many db Range?

 

144 dB, gain 0 dB, 4096 points, Blackman window

[Miska]

1 hour ago, esldude said:

With dither we can get more or less 120 db dynamic range with 16 bit audio.  Do you need more?

 

Well, not dynamic range, but you can resolve signals to -120 dB or lower. If you use noise shaping and high enough sampling rate, you can resolve signals at -200 dB or lower (of course not through real converter, but in digital domain). 16-bit has dynamic range of 96 dB.

 

[Miska]

2 hours ago, fas42 said:

Yes, it's the 'subjective' dynamics that do it for the hearing - so, how does one correlate that with measurements, and number of bits?

 

For that purpose there are couple of calculation standards (dynamic range ratings) and listings like here:

http://dr.loudness-war.info

Also see: https://tech.ebu.ch/loudness/

 

But what we are talking about how quiet and loud signals can be fit together on a recording using single stereo mic pair. It is also related to another similar aspect called signal to noise ratio but these are measured in a different way. SNR (THD+N) contains also distortion products of high level signal.

 

Microphone manufacturers also typically list some figures. Like DPA 4041 has 118 dB dynamic range. And one of the mics most used for hires recordings, DPA 4006 has 124 dB dynamic range. DPA used to have also fancier (better performing) 130V phantom power version which is used by many for hires recordings. Sennheiser and Neumann have similar figures. Best figure I found was 127 dB, which is very close to 21-bit worth (just slightly over 21 bit).

 

[Miska]

4 hours ago, mansr said:

The lowest sound level that can be captured is determined by ambient noise and self-noise of the microphone. The upper limit is likely set by the microphone as going too loud will cause severe distortion and can even damage it. You'll have to check the specs to get accurate numbers, but I doubt you'll find many offering more than 100 dB of usable dynamic range. A good 24-bit ADC will have no trouble recording this.

 

For example Neumann KM183 has A-weighted self-noise of 13 dB and maximum SPL  is 140 dB (0.5% THD). DPA 4006 has A-weighted self-noise of 15 dB and maximum SPL 146 dB. Many of these microphone models have a switchable pad from 10 to 20 dB, so you can roughly shift the dynamic range down by that amount in SPL.

 

But anyway, one way to check is to go through bunch of hires material and see what they have. With my earlier checks and random pick I can say that it is at least more than 16-bit worth.

[Miska]

12 hours ago, Arpiben said:

Noise-shaping has no effect on the noise-floor of the analogue or acoustic signal.

 

Question was about needed digital resolution, so noise-shaping applies in that scope. Noise shaping also affects analog domain if you measure unlimited bandwidth.

 

12 hours ago, Arpiben said:

From what I read that floor is at least 30 dB greater than the digital noise floor of a 16 bit dithered ( 120 db within audio bandwidth).

 

There's a difference on what is detectable level of discrete tone vs  dynamic range. There's no problem detecting single sine at -120 dB in dithered 16-bit data. But if the signal is white noise at -120 dB level, it is undetectable from the noise floor.

 

12 hours ago, Arpiben said:

Typical digital dynamic range of 120 dB (or much more) can be achieved within the audible frequency bandwidth.

 

You can only achieve that with 16-bit if you can move enough of the noise out of audible frequency band, which means you have enough sampling rate to move Nyquist far enough above audible frequency band to make space where to "park" the noise moved away from audible band.

 

For example 16-bit data has fixed total noise level, but it's frequency distribution can be anything you decide. However, if your sample rate is 40 kHz, all your usable bandwidth falls within audible range, so you have no place outside of audible band where to move that quantization noise.

 

[Miska]

1 hour ago, Arpiben said:

And what about sensivity or S/N?

For KM183 S/N = 70 dB (/94dBSPL IEC 60268-1).Noise Floor = 24 dBSPL.

Then what about the noise contribution of amplifiers and other analogue equipment before ADC ? Just curious since I have only a rough idea.

 

As I mentioned earlier, S/N figure is dominated by distortion. -70 dB = 0.03% which is pretty good distortion figure for an acoustic transducer! And A-weighted of -81 = 0.009%!

 

So for KM183, maximum SPL is 140 dB and A-weighted noise is 13 dB yielding 140 - 13 = 127 dB(A) DNR. Or with CCIR peak noise 140 - 24 = 116 dB(CCIR) peak.

 

With those DPA (originally Brüel & Kjær microphone section later split out to a separate company) 130 V phantom microphones the figures are about 8 dB better. Not listed anymore on the DPA site, but still in stock in some shops.

 

AFAIK, for example 2L uses these mic preamps:

https://www.mil-media.com/HV-3C.html

https://www.mil-media.com/HV-3D-8.html

 

And that's it before the ADC.

 

[Miska]

31 minutes ago, semente said:

A/DC
A good 24 bit AD like the Apogee 16 X has an S/N of around 120 dB and this means removing 11 dB from the previous weak link, the mic, at 131 dB
You are now recording at 20 bit (120 / 6 = 20)

 

Quite many use Merging Horus or Hapi these days with their ADC cards:

https://www.merging.com/products/interfaces/specifications#a-d8-d-a-d8-d-p-option-card

People with 130V mics certainly use something like Millennia Media mic pre-amps, and some others use for example completely custom built mic pre's for those.

 

I'm personally using RME ADI-2 Pro:

http://rme-audio.de/en/products/adi_2-pro.php#7

Which is so far among the best with unweighted 120 dB and A-weighted 124 dB SNR.

 

[Miska]

10 hours ago, esldude said:

My understanding is with dither for 44 and 48 khz rates, you shift noise away from where hearing is most sensitive to areas where our thresholds are higher. So the noise is all in the audible band, but still at levels we'll not hear it. While decreasing noise where we might thereby extending the range effectively for human ears.

 

Yes, if you use noise shaping you can shovel the noise around, but it doesn't change total amount of noise within audio band, just it's frequency distribution within the audio band.

 

Standard dither, like in my examples, is TPDF which has flat frequency distribution (so you get flat noise floor).

 

While if you use higher sampling rates you can actually move noise out from the audio band. So for example if you use 352.8 kHz, you have actually some bandwidth outside of frequency range where you have signals from microphones and you can park some of the noise there.

 

[Miska]

I would say that with best recording gear available, dynamic range capabilities of the gear exceeds both practically realizable background noise and maximum tolerable SPL levels at the same time. So the recording gear (with 24-bit resolution) is not the limiting factor. However, this is more than 16-bit audio-bandwidth (Nyquist at or near 20 kHz) encoding can represent, but less than 24-bit.

 

P.S. As a side note, I would like to remind that you can just fine encode for example 18-, 20-, or 21-bit content in FLAC and have corresponding file size savings. And that is completely standard, and gains smaller files than MQA. So if you want, you can optimize bit usage at mastering stage.