How many bits, how fast, just how much resolution is enough?

[scan80269]

Practically all contemporary DACs perform upsampling of incoming PCM stream before sending to the D-to-A converter. For example, Auralic Vega upsamples to 1.5MHz 32-bit, NAD M51 resamples to 844KHz PWM. With such DACs we do not hear Redbook 16/44.1 content being played as 16/44.1 (at least that's not what the D-to-A converter chip receives). Therefore the quality of the upsampling/resampling has more impact on the overall SQ of the DAC than the D-to-A converter chip used.

 

As upsampling/resampling technology improves, I've been amazed how well Redbook 16/44.1 content can sound coming through an upsampling DAC, which can be so good that even a high-resolution (e.g. 24/192 or 24/352.8) version of the same content delivers only a rather small incremental SQ improvement. One of my strongest impressions of the Auralic Vega is how well it equalized the SQ between 16/44.1 and hi-res PCM content. This DAC has thoroughly convinced me that Redbook 16/44.1 is itself not lacking in resolution, detail, S/N, etc. Such equipment makes the SQ advantage of hi-res PCM rather difficult to demonstrate or showcase.

 

The main advantage of hi-res PCM, IMHO: it's easier on the reproduction equipment. In other words, it is easier for hi-res content to sound good on a wider range of equipment compared to Redbook. Getting the utmost SQ out of 16/44.1 content is non-trivial, and a heavy burden rests on the upsampling/resampling algorithm, which strongly influences how a DAC sounds. Hi-res content requires less intensive pre-processing and sometimes even gets fed to the DAC without going through internal upsampling. There should be less SQ variation from DAC to DAC with hi-res content.

 

My biggest lament regarding Redbook is how Sony/Philips locked into the 44.1KHz sampling rate (long story), so now the digital audio world has to deal with two core sampling rates: 44.1KHz and 48KHz and their multiples. Many DACs use separate oscillator reference clocks to cover each frequency family, but the design complexity and possibly electrical noise level is increased because of this requirement. These days, when I play hi-res PCM, I tend to favor 88.2KHz, 176.4KHz or even 352.8KHz (DXD) over 96KHz or 192KHz, as the DAC is not forced to switch between 22.579MHz and 24.576MHz reference clocks whenever I switch to playing content based on the other frequency family (e.g. between Redbook 16/44.1 and 24/96). I have several Reference Recordings HRx (24/176.4) titles and used to wonder why RR chose 176.4KHz instead of 192KHz, but now it makes perfect sense to me.

[Ralf11]

and BTW, #3 - if there is an advantage it will most likely be found on transients

 

the whole edifice of digital sound is built on an upper hearing limit of 20 kHz, which was established in the 1930s IIRC and by using sinusoidal waves, NOT music (with say snare drums) - if that is wrong the entire Nyquist NoNo collapses

 

OTOH, don't hold your breath... and the lack of controlled studies to show that a higher (consumer level) bit rate/depth is preferred suggests that the redbook std. is fine

[esldude]

This was an interesting listen. The music is indeed still there, under all the noise.

 

24 bit makes things a lot easier though, no?

 

24 bit does, but had I run this through a DAC and its analog output, or via a power amp before applying that digital gain, it would pick up enough noise the result is virtually identical to the dithered 16 bit. You might gain a little less noise if your gear is really quiet.

[semente]

Well in my case because I took my own advice, and stopped paying attention when I saw the stair case looking graph. Also because you mentioned the more points you have taken the more accurate. That describes a higher vs lower sample rate.

 

OOPS, mansr answered in between and gave a much shorter answer.

 

I wrote points when I should have written levels, otherwise there's no mistake in what I wrote (before that).

PCM digitisation has an amplitude and a time scale; more bit levels are also points.

And, yes, dithering masks quantisation errors but 24 bit still has more resolution.

 

Can you hear that?

Like I said, perhaps. If your room is dead quiet, if you play loud enough, with transparent equipment and the right recording it's impossible not but I find it very difficult...

See post #19.

 

R

[gmgraves]

I've as of yet to find anything in our industry to impugn this presentation on why 16/44.1 is good as it gets for us.

 

24/192 Music Downloads are Very Silly Indeed

 

You're overlooking the fact that digital is not very good at either end of the dynamic range spectrum. It doesn't handle peaks very well because they can often exceed 0 Vu which means that the system runs out of bits, and that's a no-no. This potential condition requires that the recordist back-off on overall record level to avoid overmodulating. This means that in a 16-bit system, low level passages are down in the mud where only one or two bits are being "exercised" causing massive distortion + noise and guaranteeing that ambience gets severely truncated. This can be mitigated somewhat by adding small amounts of random noise (called dithering) to the signal, but that's "cheating". Best of add more "bits". With 24 or 32-bit encoding one can avoid overmodulating the system by backing off on the record level without fear that your low-level will be shifted down in the 1 to 2 bit level or lower (where the signal is simply truncated).

[plissken]

then let's see it

 

 

as for monty being the real deal, what are his qualifications? I'm not saying he is wrong; I am just used to seeing things justified by citations to previous work or by the Results section, which is obtained by following the procedures in the Methods section - that is Science 101

 

The qualification is the output on the analog scope. What else do you require?

[plissken]

You're overlooking the fact that digital is not very good at either end of the dynamic range spectrum. It doesn't handle peaks very well because they can often exceed 0 Vu which means that the system runs out of bits, and that's a no-no. This potential condition requires that the recordist back-off on overall record level to avoid overmodulating. This means that in a 16-bit system, low level passages are down in the mud where only one or two bits are being "exercised" causing massive distortion + noise and guaranteeing that ambience gets severely truncated. This can be mitigated somewhat by adding small amounts of random noise (called dithering) to the signal, but that's "cheating". Best of add more "bits". With 24 or 32-bit encoding one can avoid overmodulating the system by backing off on the record level without fear that your low-level will be shifted down in the 1 to 2 bit level or lower (where the signal is simply truncated).

 

That is something that you could comment to Monty on. He seems perfectly capable of answering it.

 

Dithering is not cheating. It's managing the known noise component so you can shift it to the in-audible end of the spectrum and filter it out.

[Ralf11]

don't be evasive plisskit

[gmgraves]

That is something that you could comment to Monty on. He seems perfectly capable of answering it.

 

Dithering is not cheating. It's managing the known noise component so you can shift it to the in-audible end of the spectrum and filter it out.

Uh, that's the reason the word cheating is in quotes. Yes, it is introducing low level noise to linearize low level signals to keep from a situation where only one or two of the LSBs is being manipulated by changing the PCM to virtual PWM. But it's a bandaid when compared to actually adding more bits.

[plissken]

Thx for pointing out the quotes....

[wgscott]

but at what point does the human ear stop hearing any improvement?

 

Between 15 and 20 kHz, depending on your age. So a 24 bit, 96 kHz recording should be greater than a factor of two overkill, which for even a paranoid like me, should be good enough. Also, on recordings exceeding a sampling frequency of 96kHz, I can't even see additional content, so it is most likely wasted space.

[audiventory]

Doesn't higher bit depth lower the noise-floor?

 

Absolutelly.

 

And if the noise-floor is lower, don't we get increased resolution?

 

Lower noise is result of higher resolution. I.e. target of any increasing of resolution is decreasing of the noise.

 

For same decreasing of noise floor, resolution may be increased differently in two domains: time and frequency.

 

For decreasing noise floor 2 times, adding of 1 bit is more effective than increasing of sample rate 2 times.

 

Example:

Original bitstream is 16bit x 44100kHz = 705600 bit per second

 

1. For adding 1 bit bitstream: 17bit x 44100kHz = 749700 bit per second (bitstream increased about 6%)

 

2. For increasing of sample rate 2 times bitstream: 16bit x 44100kHz x 2 = 1411200 bit per second (bitstream increased 100%).

 

First case have 6% increasing of resolution, second - 100%. But noise floor decreased 6 dB for both.

 

I.e. resolution and noise floor are separate physical values.

[yamamoto2002]

On Discrete-time signal processing 3rd ed. by Oppenheim and Schafer, p249: "obtaining a SNR of about 90-96 dB for use in high-quality music recording and playback requires 16-bit quantization, but it should be remembered that such performance is obtained only if the input signal is carefully matched to the full-scale range of the A/D converter"

 

On playback, 16bit 44.1kHz is sufficient. but on recording, 16bit 44.1kHz is not sufficient for some application. 16bit recording is little bit difficult to handle. when recording level is too low, quantization noise can be heard (that is white noise constantly heard). when recording gain is too high, digital clip occurs and it means recording failure. using 24bit bit depth, quantization noise becomes lower and more conservative gain settings can be used than 16bit. Precise recording level matching is possible when source material is repeatable such as tape or automated playback of MIDI device but not possible for live recording of acoustic instruments.

[semente]

Uh, that's the reason the word cheating is in quotes. Yes, it is introducing low level noise to linearize low level signals to keep from a situation where only one or two of the LSBs is being manipulated by changing the PCM to virtual PWM. But it's a bandaid when compared to actually adding more bits.

 

That's it.

Masking through dither, even when used below the audible threshold, is still a crutch.

 

Alternatively one could always turn down the volume or open the windows to increase the noise in the room.

 

R

[r_w]

Exactly mirrors my experience... the VEGA sounds fabulous with Redbook.

 

;-)

 

One of my strongest impressions of the Auralic Vega is how well it equalized the SQ between 16/44.1 and hi-res PCM content. This DAC has thoroughly convinced me that Redbook 16/44.1 is itself not lacking in resolution, detail, S/N, etc. Such equipment makes the SQ advantage of hi-res PCM rather difficult to demonstrate or showcase.

[BlueSkyy]

Don't get confused by the stair step of the wave form. Too many audiophiles think it's distortion rather then sampling points. Even Nelson Pass has incorrectly referenced this. Lollipop plot would be a bit better representative.

 

Either way here a tutorial that shows actual reconstructed signal on an analog scope.

 

 

This is one of the reasons I don't placate or cater to audiophiles that don't understand D/A theory.

 

Once the analog signal is accurately recreated I don't care the PCM or DSD encoding before that.

 

Plissken, thank you for the link! This video really hit it out of the park for me and reset my way of thinking. Mr. Montgomery's knowledge and his ability to put it into layman's terms will save me money! Now, if I can overcome the part of my brain that says high dollar speaker wires are better sounding than appropriately sized (AWG) lamp cord and .999 silver cryogenically treated fuses sound better than their $0.50 cent equivalent, I'll be all set!

[PeterG]

I don't have the technical patience to dispute those advocating that listeners/buyers stop at redbook. But a couple of global thoughts back to the OP's fundamental question of which music to buy:

 

First, I think it's been explained here and elsewhere why a 24-bit recording is superior to 16. It's kind of funny to hear guys who have dropped serious coin on their hifi turn around and suggest the difference between 24 and 16 is "probably" too small to hear. C'mon, man--how about those titanium speaker enclosures and the hand spun interconnects you bought last year?

 

Second, as a person who buys plenty of redbook and plenty of HD Tracks, I have to say that a disproportionate number of my best sounding files are higher resolution and bit depth from HD Tracks--Jazz at the Pawnshop, Kind of Blue, Van Morrison's Astral Weeks, Joni Mitchell's Blue--just to name a few. The sound quality of these files may or may not be resolution or bit depth related, but everything we buy for listening is packaged with features that we may or may not need. I only know they sound great.

 

Trust your ears!

[r_w]

I'm my system the difference is too small to hear... not suggestion... real listening experience.

 

It might be how well the VEGA DAC does its job (or how badly other do theirs).

 

I'm confident enough that 16/44 delivers everything the ear needs for 'music magic' and therefore I'm missing nothing of concern.

 

Mathematically 24 bit is utterly superior to 16... but those extra 8 bits aren't being detected by my ears, so they are wasted.

 

For that reason I don't worry about hires.

 

As always YMMV.

 

;-)

 

 

It's kind of funny to hear guys who have dropped serious coin on their hifi turn around and suggest the difference between 24 and 16 is "probably" too small to hear. C'mon, man--how about those titanium speaker enclosures and the hand spun interconnects you bought last year?

[audiventory]

Mathematically 24 bit is utterly superior to 16... but those extra 8 bits aren't being detected by my ears, so they are wasted.

 

There is physically no way to correctly compare 16 and 24 bit from technical point of view.

[semente]

There is physically no way to correctly compare 16 and 24 bit from technical point of view.

What if we reduce the gain of a recording that maxes at 0dB by say 36dB on both Redbook and 24/44.1?

Won't this give us an idea of 16 bit limitations?

 

Perhaps we could do this with Mario's PlayClassics files...

 

R

[audiventory]

What if we reduce the gain of a recording that maxes at 0dB by 36dB on both Redbook and 24/44.1?

Won't this give us an idea of 16 bit limitations?

 

Perhaps we could do this with Mario's PlayClassics files...

 

Signal noise ratio of 16 bit is worse 24 bit, of course. It is technical fact. Everybody can see it.

 

1. I want to said when we compare 16 and 24 bit we compare not only signal noise ratio but different modes of apparatus.

 

The different mode issue give additional ambiguity that distort picture of signal noise ratio.

 

2. Also there added resolution-conversion distortions.

 

3. But worst case is comparing two separate records produced in different resolution.

 

There is music production/post-production issues may be added to previous items.

[r_w]

Neither it seems can my ears come up with a sonic comparison.

 

I wonder if someone will chime in with something really solid that'll start me wondering if my brain has decided that hires is pointless and expectation bias is confirming my preference for Redbook?

 

;-)

 

... seriously though, if someone gave me my entire library in 24 bit (if that was possible) I'd obviously accept the gift and use it... but I'd never contemplate doing any work myself to get (any of) my library in hires. On my system that would seem a waste of time.

 

 

 

There is physically no way to correctly compare 16 and 24 bit from technical point of view.

[gmgraves]

That's it.

Masking through dither, even when used below the audible threshold, is still a crutch.

Alternatively one could always turn down the volume or open the windows to increase the noise in the room.

R

 

That's what I meant when I said it's "cheating". I.E. it doesn't solve the problem, it merely masks it, like a bandaid on a cut finger.

 

It's interesting how dithering was discovered. Digital mastering engineers kept wondering, in the early days of CD, why CDs mastered from analog tape sounded better than the latest DDD recordings, since both went through the same digital equipment. Turned out that the tape hiss from the analog recoding was acting like an inadvertent dithering signal. The engineers, once having discovered this, experimented with reducing the noise level to find the right balance between adding enough noise to act as a dither, but not so much as to be blatantly audible. Once they found that happy medium, the rest was recording history. It is interesting to note that as long as recording engineers use enough record level to keep at least 4-bits below the level of the quietest passages in one's recording, and 4-bits above the loudest passage (in other words use 16-bits within a 24-bit "envelope") 24 and 32-bit recording don't need dither, until down-converting the file to Red Book for CD mastering.

[Sal1950]

44.1/16, maybe 44.1/24 or 48/24. In extreme cases with exemplary gear and young terrific hearing 48/24 might be needed.

Some of the more knowledgeable people for the most extreme possibilities estimate you might need 65 khz/20 bit. As that isn't an available rate, 96/24 or 88/24 does it for you with room to spare. If one just wishes to absolutely end any possibility of any person in the world hearing something in less than max fidelity due to recording limitations then 96/24 is enough and that.....is.......simply......that.

Long winded version of jjjjusta's reply.

 

That totally wraps it up guys.

All the rest can only be equated to breast and penis enhancement surgery, more is better right?

Have you seen the lips on this generation of actresses lately? They all look like someone gave em a good punch in the mouth. LOL

[esldude]

That's what I meant when I said it's "cheating". I.E. it doesn't solve the problem, it merely masks it, like a bandaid on a cut finger.

 

It's interesting how dithering was discovered. Digital mastering engineers kept wondering, in the early days of CD, why CDs mastered from analog tape sounded better than the latest DDD recordings, since both went through the same digital equipment. Turned out that the tape hiss from the analog recoding was acting like an inadvertent dithering signal. The engineers, once having discovered this, experimented with reducing the noise level to find the right balance between adding enough noise to act as a dither, but not so much as to be blatantly audible. Once they found that happy medium, the rest was recording history. It is interesting to note that as long as recording engineers use enough record level to keep at least 4-bits below the level of the quietest passages in one's recording, and 4-bits above the loudest passage (in other words use 16-bits within a 24-bit "envelope") 24 and 32-bit recording don't need dither, until down-converting the file to Red Book for CD mastering.

 

History of Dither

 

The concept of dithering to reduce quantization patterns was first applied by Lawrence G. Roberts in his 1961 MIT master's thesis and 1962 article though he did not use the term dither. By 1964, dither was being used in the modern sense. To ameliorate the negative effects of quantization, early workers in the field simply combined white noise to the original ADC input signal. Dither is often used in digital audio and video processing, where it is applied to sample-rate conversions and to bit-depth transitions; it is utilized in many different fields where digital processing and analysis are used — especially waveform analysis. These uses include systems using digital signal processing such as digital audio, video, photography, seismology, radar weather forecasting systems and more. Considering the premise is that quantization and re-quantization of digital data yields error, if that error is repeating and correlated to the signal, the error that results is repeating, cyclical, and mathematically determinable.