How many bits are really meaningful?

michael123 · Nov 18, 2011

Hi Bruce

I analyzed some recordings that sound great to my ears, findings only 70db of dynamic range at best
Frequency range rarely goes beyond 25Khz, but what about the sample resolution?

Leaving personal taste aside, what parameters would you recommend to look at while choosing a DAC?
I frequently see the measurements at Miller Audio Research (Hi Fi New), the best DAC out there, like Bricasti Design M1, have about 118db of dynamic range (btw, these measurement differ from stereophile since stereophile tests only on 1Khz, but HFN use full spectrum and after 1KHz the dynamic range is narrower)

What do you think also of new 32/384 DAC like these M2Tech?
The only 24/354Khz recordings (consumer, I mean) I saw at Lindberg 2L site..

--Michael

Orb · Nov 18, 2011

Michael,
seems you are a subscriber to HifiNews like me.
I find their hi-rez download review section of use as it shows both the dbfs-amplitude against frequency for the albums reviewed (using an average track on the album for the graphs).
This analysis is very useful as it provides a view of whether upsampling has caused problems and whether it is a true hirez file, along with the average and peak energy against frequency.

There are a few (one in the December issue as an example) where their mean is around -40dbfs, and have excellent energy through all of the claimed frequency/resolution of the album, this then would require the 24bit capabilities, which as you note on SOTA is usually at 118db when converted, one of the exceptions to this is the Devialet D-Premier and this is where it comes into its own, but from my own experience this needs two of them setup in mono to be exceptional.

The downside is that the average quality they have measured-reviewed so far do not make use or need the full capabilities of 24-bit, nor do they have good extension in the frequency range.
The challenge is finding the exceptional and good ones from the mediocre ones IMO.

To add to your question and funny enough was one I have been mulling over for the last few days.
Bruce, from your experience how do studios manage-master hi-rez (say 24-bit 96khz) music when it comes to the extended frequencies?
What I mean, do they artifically boost these frequencies above a certain frequency in a way similar to the concept of loudness wars so they do not drop off too quickly, or try to keep it as natural to what the mic/arrangement recorded, or a mixture of both.
I appreciate different studios and engineers may also complicate this, would be interesting on your insight how this is best managed and what you feel is a good practice in general.

Thanks
Orb

amirm · Nov 18, 2011

I will add a bit to this puzzle. In my last job, we used to develop lossy audio codecs. To go with the crowd, we supported up to 24-bits and 96 KHz.

A lossy codec uses a human hearing system to decide what to keep, and what not to keep. By throwing out what we can't hear, it saves space. Well, that presents a quandary. By definition, if I played a 30 KHz tone, none of us can hear so anything above 20 Khz should be thrown away. But if you did that, you would be doing no better than CD's 44.1 Khz sampling (22 Khz bandwidth). So I asked my researchers what they were using as far as frequencies kept. They show me a graph that extended above 20 KHz. So I asked them how they knew it was right to keep that much of the ultrasonic content. Answer was that they didn't!

Similar question came up at AES where someone from FHG (guys who invented MP3) was talking about a lossy codec like above. And the presenter got stomped just as well not knowing what to say.

It is a bit like wondering if the fridge light is still on after you close the door.

Ronm1 · Nov 18, 2011

amirm said:
It is a bit like wondering if the fridge light is still on after you close the door.

Its never on if the h/w works as designed/documented.

Bruce B · Nov 18, 2011

As far as bit rate is concerned, short answer, 24-bit is all that is needed. There are currently no hardware converters that can even come close to those figures (-144dB). When you start adding amps, pre-amps, cables and other things, your once dynamic range of >100dB is now down to 60-70dB because of the noise, system and ambient.

We have 2 converters that can do up to 384kHz. Those are the Digital Audio Denmark and the Playback Designs. I feel these sample rates were actually developed so you could edit DSD. We've done quite a few masterings in DXD (352.8kHz) because we felt that not only was it the converters/software sweetspot, but because it could be easily converted to DSD if the client was to produce a SACD as well.

The sound/character of converters has more to do with the analog section and filters that it uses than it does with what chip is used.

Orb said:
Bruce, from your experience how do studios manage-master hi-rez (say 24-bit 96khz) music when it comes to the extended frequencies?
What I mean, do they artifically boost these frequencies above a certain frequency in a way similar to the concept of loudness wars so they do not drop off too quickly, or try to keep it as natural to what the mic/arrangement recorded, or a mixture of both.Orb

When we receive masters > 24/88.2, we try to keep them at that sample rate throughout. You do more damage doing conversions than anything else. If it needs more air/ambience, of course we'll put a shelf on at 18-22k. "Usually" just fixing bass problems solves 80% of the issues we have in the mid/treble region. Studios are getting better now at delivering files greater than 48k. With the Korgs replacing all the Masterlinks, as 2-channel mixdown decks, the music finally has a fighting chance.

Orb · Nov 18, 2011

Thanks Bruce and agree about conversions as this is something that has come out in HifiNews analysis.
So in your experience it would be unusual to raise say the frequencies that are beyond 25khz?

From the analysis of HifiNews download measurement, there is a big difference in frequency energy at the higher frequencies even when they are not upsampled between different albums and studios; I guess this then is more to do with the recording mic/arrangement rather than the mastering?

Cheers
Orb

fas42 · Nov 18, 2011

The interesting thing in the high res recordings, at least the ones I've looked at, is that the over 20kHz content is pure noise: there's no harmonic or structural content in it. To me, it looks as useful as tape hiss ...

Frank

Orb · Nov 18, 2011

Not sure how you are looking at them Frank,
most that HifiNews looks at do show that the signal above 20khz is not noise, what problems they do identify are more to do with conversion/upsampling/not native hirez recordings.
But they are using a very specific measurement/software tool they developed (Based upon the work started by Keith Howard and his investigation articles on hires downloads) as there are currently no tools that do this accurately and in an easy to read way.
Both Paul Miller and Keith Howard are pretty respectable when it comes to analysis-testing-measuring development.

Cheers
Orb

fas42 · Nov 18, 2011

Orb, if you use a tool like Audacity to view the waveform, and zoom in to a high magnification you can see the "wiggle" of the signal at frequencies above 20kHz just nicely meandering back and forth constantly all along the track, in a very meaningless and random way, irrespective of what the music is doing. If you look at the Redbook version of the same track it's absolutely identical to the last 0.001%, except that the RB doesn't have this constant high frequency shifting of the feet that the hi-res has. So, in my book, that's good ol' noise, as a special extra!

Frank

michael123 · Nov 19, 2011

fas42 said:
The interesting thing in the high res recordings, at least the ones I've looked at, is that the over 20kHz content is pure noise: there's no harmonic or structural content in it. To me, it looks as useful as tape hiss ...

Frank

Not really: if you look at the peaks, CD recordings have flat top mostly (due to a filter at around 16-18KHz or whatever).. giving you this 'digital' sound.. Especially you can hear with percussion tracks

Ripped vinyl recordings and high-rez (if they are not upsampled

) do not suffer from this sickness, you will see smooth response up to 25KHz..

Frequency response is something else.. No question for me that at least 88KHz sampling is needed,
my question was about real sample size.. for curiosity..

--Michael

fas42 · Nov 19, 2011

michael123 said:
Not really: if you look at the peaks, CD recordings have flat top mostly (due to a filter at around 16-18KHz or whatever).. giving you this 'digital' sound.. Especially you can hear with percussion tracks

I'm sorry, I don't follow you. I presume you mean a time domain picture of the track, the waveform on a second by second basis. If you see "flat tops" you're seeing the results of sound engineers caught up in the loudness wars, deliberately clipping and distorting the recording. You can't blame the medium for stupidity in the record company's head office and the recording studio. People often criticise digital for mucking up percussion playback, but that is the result of less than optimum playback. CD, etc reproduction takes no prisoners: unlike vinyl, if something is not quite right with digital playback it can be extremely flat or obnoxious to listen to ...

Frank

NorthStar · Nov 19, 2011

Not from what I'm hearing (well that too) but also from what I'm reading (over the years & by the true experts); all that is sufficient in digital audio is 20-bit/88.2kHz.

But a good DAC like the Burr-Brown PCM-1704K is just fine by me.

beek · Nov 20, 2011

bits

amirm said:
I will add a bit to this puzzle. In my last job, we used to develop lossy audio codecs. To go with the crowd, we supported up to 24-bits and 96 KHz.

A lossy codec uses a human hearing system to decide what to keep, and what not to keep. By throwing out what we can't hear, it saves space. Well, that presents a quandary. By definition, if I played a 30 KHz tone, none of us can hear so anything above 20 Khz should be thrown away. But if you did that, you would be doing no better than CD's 44.1 Khz sampling (22 Khz bandwidth). So I asked my researchers what they were using as far as frequencies kept. They show me a graph that extended above 20 KHz. So I asked them how they knew it was right to keep that much of the ultrasonic content. Answer was that they didn't!

Similar question came up at AES where someone from FHG (guys who invented MP3) was talking about a lossy codec like above. And the presenter got stomped just as well not knowing what to say.

It is a bit like wondering if the fridge light is still on after you close the door.

Many years ago the Audio Critic said 16bit isnt enough, seems now he was right.

fas42 · Nov 20, 2011

Sorry, can't let that go: 16 bits is plenty, which is frequently completely wasted. One of the hilarious things is when the typical screen capture of a workstation is shown, of some audio waveform, to "prove" something, is that you are seeing the information detail that can be held in 6 to 7 bits; you still have up to 10 bits up your sleeve to carry all those lovely micro-dynamics people are so keen on ...

Frank

Thomas.Dennehy · Jan 5, 2012

When transferring vinyl recordings to digital format, I keep my original and all my intemediate results in 24-bit/96 KHz (WAV). It doesn't matter whether or not I can "hear the difference" or whether all these bits are musically meaningful on any particular piece.

When it comes to information, more is always better. The more samples you take, and the more bits per sample you have, the more mathematically accurate copy of an analog waveform you create. Your software tools can see the difference.

TGD

NorthStar · Jan 5, 2012

I dunno if I agree with you Frank. I think that 16-bit has some definitive limitations...

And 44.1kHz ain't enough. ...To get rid of the higher harmonics distortion in the audio frequency range.

And 88.2kHz is easier to implement than 96kHz. ...Same for 176.4kHz versus 192kHz.
{Multiples of 44.1kHz.}

Phelonious Ponk · Jan 5, 2012

I can't hear a tone above 14k. Makes me think that anything above 20k is a pretty serious waste of digits for me. Meyer and Moran provided some pretty serious evidence, and a pretty well-designed study, that people, including audiophiles and audio pros can't reliably differentiate between Redbook and high res. I don't know if it's absolutely true or not, but it doesn't matter much, because the overwhelming majority of the music I care about is not available on hi-res. So I hope they're right. That would be nice.

Tim

fas42 · Jan 5, 2012

LordoftheRingsEE said:
I dunno if I agree with you Frank. I think that 16-bit has some definitive limitations...

And 44.1kHz ain't enough. ...To get rid of the higher harmonics distortion in the audio frequency range.

And 88.2kHz is easier to implement than 96kHz. ...Same for 176.4kHz versus 192kHz.
{Multiples of 44.1kHz.}

Rest easy, Bob. And Tim. 16 bits, 44.1kHz, is plenty good enough, unless you're using a digital editing suite, program, to play around with the audio data. I've been through the exercise many times, of looking at what's actually in a normal hi res files as compared to 16 RB version, and it's pure noise. Of absolutely no musical significance whatsoever. It's exactly equivalent to tape hiss; there's no information stored in it. One exception I've seen of late is the file posted by Basspig before Christmas, of a personal recording of classical material. That had real, above 20kHz information in it; I have yet to try and see if I can pick any audible difference with that information being in the file, versus removed. But perhaps someone else has already gone the exercise, on a very high resolution system ...?

Bob, if you can hear a difference between the 44.1kHz and higher sampling, it's because the DAC is working better with material being fed to it at the higher rate. Not because there is more distortion in the RB source material, but because the DAC is distorting trying to do its job. It's faulty playback, no more and no less than that; the engineering of the converter overall is not up to the mark ...

Frank

NorthStar · Jan 6, 2012

Phelonious Ponk said:
I can't hear a tone above 14k. Makes me think that anything above 20k is a pretty serious waste of digits for me. Meyer and Moran provided some pretty serious evidence, and a pretty well-designed study, that people, including audiophiles and audio pros can't reliably differentiate between Redbook and high res. I don't know if it's absolutely true or not, but it doesn't matter much, because the overwhelming majority of the music I care about is not available on hi-res. So I hope they're right. That would be nice.

Tim

Yes Tim, but what you can actually hear down below, in your own hearing range,
is directly affected by what's going on up there.
[It effects the audible listening range.]

That's why it is best to really throw those distorted second harmonics way up there in the 'firmament'. :b

NorthStar · Jan 6, 2012

Frank, what about SACD?

How many bits are really meaningful?

Well-Known Member

New Member

Banned

Member Sponsor

WBF Founding Member, Pro Audio Production Member

New Member

Addicted To Best

New Member

Addicted To Best

Well-Known Member

Addicted To Best

Member

New Member

Addicted To Best

New Member

Member

New Member

Addicted To Best

Member

Member

Similar threads