Yes, I have searched the forum.
Yes, maybe I am dumb.

But it seems I cannot find the answer.

Why do we need 24bit/48kHz/96kHz/192kHz if 16bit/44.1kHz is good enough? Are there any situations that 16bit/44.1kHz simply cannot satisfy? In other words, is there any real need for the higher bit depth and sampling rate?

Thanks for answering.

We dont need it. It's just virtual useless number-games to give people the incentive to buy new equipment and then re-buy all our music. There are some *technical* arguments for using 48khz instead of 44khz.... but the actual benefit for normal endusers is zero.

I heard someone saying that increasing the sampling rate improves the SNR. I simply don't get it. It would be grateful if someone can enlighten me on this as well.

Thank you.

Normally, improving sampling rate improves bandwidth, not SNR. SNR is improved by increasing bit depth. Noise shaping allows trading bandwidth for effective SNR.

The question is why one would need a SNR of >96dB or a bandwidth of over 22kHz for end user playback.

This is exactly what I am wondering...

What I learnt is that, in frequency domain, the higher the sampling rate, the larger the bandwidth, and the farther away between the base band and its images..

And I see nothing related to SNR by improving sampling rate.

Would you please give me more information on noise shaping?


Then why DVD audio uses 96kHz? This is something I always wonder.

Thank you.

I totally agree with you all. Increasing sample rate and bit depth is not perceptible, excepting for some animals.

Sony Super Audio Compact Disc (SACD) is the fine high end example of the no-use technology. Sony claims to consumers his high sample rate, bandwidth etc but do not claim that 0.00001% could hear the difference.

Moreover, even if you have a HD Player with 100 kHz bandwidth, you have to get the same characteristics for the whole audio elements (amplifiers, loudspeakers).

"The audio quality of a sound equipment set is equal to the audio quality of the worst device."

Use the search function.

I am sorry. Thanks.

I have read that if you are recording music and performing several layers of edits and signal processing, that higher sampling and/or bit depth may be of actual value to the final mixdown. I don't do that kind of work, so that is hearsay pretty much. For simple audio playback you are correct, 16/44.1 is totally sufficient.

I agree that it makes no difference to end users... though there are technical reasons why this would have its advantages...

most of it comes from the way hardware plays back music.... it has to do with the enforcements of peaks... each sample is essentially enforced as a peak in the reproduced sound wave... but this should NOT be the case... the sample may have been taken along any part of the wave (and most likely not at a peak)...

The affects of this are mostly negligable and hardware also works to minimize the affects... but it shouldn't have to... and higher sampling rates allow the hardware to start with a closer reresentation to the original wave...

I'll look for some articles about this "phenom"... I actually looked them up a year or two ago cause an audiophile engineer I know was telling me about it and I didn't believe him...

in the end the LARGE majority people would never be able to tell... and I think most of the reason CD's sound like s**t to me is due to gawd awful compression they fell they need to put on the CD's

Peace
Ross

This really doesn't make any sense at all.

Exactly. In the digital domain you make rounding errors due to the quantization. If an engineer does all the processing in the 16 bit domain, these rounding errors will build up and therefore decrease the signal to noise ratio. If all the processing is done in the 24 bit domain and the audio is quantized to 16 bits per sample afterwards, these rounding errors will be negligible.

A higher sampling rate is not really necessary.


I also think that's the case. I don't think many people are able to ABX 24 bit versus 16 bit audio, for example. Or 44.1 kHz versus 48 kHz.

edit: I can imagine that a higher sampling rate can be of use if multiple anti-aliasing filters exist in a recording chain. Each anti-aliasing filter will remove high frequencies, so multiple filters may remove too much energy from the high frequency bands. However, in most studios only one A/D converter with anti-aliasing filter is present in the chain (for example in the mixing desk).

I had a thread here: http://www.hydrogenaudio.org/forums/index....showtopic=17118

It is quite an indirect and meandering thread (experimentation, disbelief etc), but amongst all of it are some definite abx results.

The trouble is, there is no way I can say for certain that i am not simply hearing large errors introduced by my headphones. And since headphones on the market do not seem to be designed for such high resolution there is not much i can do about that.

ok, i'm a newb so don't give me some slack, but wouldn't that mean that CDs are the last audio format we'd need for the foreseeable future? I mean I know from previous posts that DVD Audio is useless, so anything better than CDs with only 2 channels is pretty useless.

heh... I thought the same thing when I read it back... I'm not the best to explain obviously... I didn't have time (and can't recall where) to find the resources we found talking about this... but I did find this one (which is not the one I remember but talks about the same thing)....
http://www.themusicpage.org/articles/SamplingTheory.html

any better?

A paper where the author doesn't seem to correctly understand Nyquist is not an acceptable argument, no.

Well, the advantage of DVDA and SACD is exactly that they are multichannel. And have DRM, which is obviously a big advantage to some people.

where does he go askew .... ?

I have to step out and disagree that there is no audible difference between 16-bit and 24-bit recordings. The decision to use 16/44.1 was arrived at, not because it was audio nirvana, but because it was the best comprimise possible with the digital to analogue conversion technology of the 80's.

To hear the benifits of 24-bit vs. 16-bit, you need a few things:

1) Capable and descriminating ears. The fact is that no everyone can hear the difference, and that's fine. However, even with a capable ear, the individual has to know what he's looking for. We're talking about very subtle points here, the difference is not going to be night and day.

2) Capable audio equipment. Just because you have a DVD-A or SACD player doesn't mean your system is capable of delivering an audible difference. You need very good equipment to make 24-bit listening worth your while.

3) Quality recording. It doesn't matter if the recording is 64-bits - if it was poorly recorded, and poorly produced, it's going to sound poor. Adding another 8 bits is not a magical fix-all.



I disagree with this, especially with respect to SACD. Alot of hardcore audiophiles (the people who would be investing in expensive high bit recordings) feel very strongly that music should only bet two channels.

In SACD multichannel is optional, but 2-channel is required; in other words every SACD will have high-bit two channel audio, but not necessarily multichannel.

OK, so here are some more technical questions I found...

1) The 44.1kHz has its historical reasons, but how about 48kHz, 96kHz, or even 192kHz?
How are these numbers chosen? Any technical and practical advantage over 44.1kHz? And why creative cards resamples to these sampling rates?

2) 16bits, 24bits, 32bits, etc, I think these values are chosen because they form complete bytes (1 byte = 8 bits). Or are there any other reasons for the numbers higher than 16bits? Again, any technical and practical advantages over 16bits, besides keeping accuracy and preventing errors from various quantization before final down-mixing?

What I only see is that, these higher numbers mean more data is sampled and stored with higher accuracy, and thus gives theoretically higher quality than 16bits/44.1kHz. But, oh well, they are only theoretical after all. Practically we may not hear a single difference...

The Nyquist theorem states that frequencies below the Nyquist frequency can be sampled correctly, not at the Nyquist frequency, to begin with.

Practise disagrees with your suppositions. Is there any proof that we can hear beyond 96dB SNR?

Practical experiments have shown most listeners already get into problems hearing to the 13th bit (78db SNR). If you think you (or another person) can do better, please let them take the MAD challenge with whatever equipment and ears you want, come back aftwards, and see if you would still tell me there is a practical reason to have more.

For the record, claiming you hear over 100dB SNR equals claiming you can hear the person next to you breathing while listening to a live rock concert. Still sounds reasonable?

so, on a slightly different note, would the 24/96 recordings be harder to saturate (eg. to master crappily and make it clip?)

if so, it might be preferable with all the 'new' music that comes out

No, it's exactly the same as with CD.

People have already noted that many SACD and DVDA recordings are just as clipped and compressed as CD's.

how odd.. since the SNR is that much higher, it can't be that they just master it for sacd/dvda and then use that same mastering for cd, since it would, if i'm not mistaken, be clipped much worse than on the 24/bla version..
i wonder what the point of their advertising with 'higher quality' is it's basically just higher volume out of the box

Clipping has to do with maximum sample values, not with SNR.

The differences between CD and other formats are completely irrelevant for that. The clipping is not a technical problem. It's one of being AS LOUD AS POSSIBLE. Mastering engineers are REDUCING the dynamic range. That's yet another reason "better" formats are useless.

yeah.. forgive me for putting that a bit awkwardly, but as i understand it the things clip because of the 'saturation', which is directly linked to the bit depth, correct? and they clip because of the volume being too high, so in that regard sacd should be harder to saturate than redbook cd's.. correct?

well, and what about the HDCD (20bit) technology? what are the other 4bits used for?

Well, it sounds like these 24bits / whatever kHz are simply marketing gimmicks than really practically useful for the end user.

No, no, no, no....please reread my post. You could have a 2560000 bit format and it would still be able to clip just as easily as a 8 bit wave file, as long as people are trying to make the music as loud as possible.

A CD doesn't store "loudness". The loudness is controlled by the person with the volume know. Given an equal settings, dynamics compression + maximized peak values will give a louder sound. Apparently a lot of people in mastering think this is desirable, so they do just that: maximize the peaks to the maximum the recording format can store, be it 15, 24 or 10000 bits. And very often, they go beyond that (clipping). Adding more bits (more SNR) is helpless against that stupidity. The formats have assloads of headroom but NOBODY USES IT!

HDCD is CD + dither + noiseshaping + a control signal. And it's only 16 bit, which is why it is compatible with CD's.

The dither + noiseshaping will give an effective increase in resolution (marketing claims: 20bit effective). The control signal will prevent clipping in some circumstances (it's a workaround for braindead mastering engineers).

You can do the dithering + noiseshaping on any CD (not just a HDCD) and it will work effectively. But there are few CD's that are properly made like that.

Use the search function if you want to know more.

Sorry for popping in like that, but here's my answer to the main question here...

I do a lot of audio editing (some people call me a "producer"). For my needs, 16/44.1 kHz is far from sufficient - after applying some filters and effects, artifacts are very audible and audio is heavily damaged. This is compensated with higher sampling rates and bit depths - for now I'm content with 32 bit/88.2 kHz audio.

However, this tosses me out from the "end users" basket. Yes, objectively, for the end users 16/44.1 audio is totally sufficient.

Why end users are being presented with "higher-quality" audio now is beyond me, since most of them can't even hear the difference between 192 kbps mp3 and a "common" audio CD.

Cheerz =]

I understand the part about 32 bits per sample (although 24 bit seems sufficient in most cases), but why 88.2 kHz? Do you have many subsequent A/D converters in your recording chain, like I explained above, because I don't think it is of much use if you do all your editing in the digital domain and resample to 44.1 kHz afterwards.

Definitive answer... Artists record at as high as 192khz. It has to be down-sampled to 44.1khz to fit on a media CD. It just gives you (you = audiophiles) a fuzzy feeling knowing that it's "really close" to the original recording.

Most studios know that recording at > 44.1 kHz is pretty useless, unless it is a recording for DVD-A or SACD purposes. All studios I know record simply at 44.1 kHz / 24 bit. There are no artists that I know of that record at 192 kHz to give the end user a fuzzy feeling

Some DSP algorithms may benefit from a high sampling rate, but strictly speaking, they can all be written to work fine at 44.1, or else the algorithm itself will just include its own resampler to do what it needs to do. (A trivially naive example: suppose you want to make a comb filter in less than 10 taps with a starting freq of 12khz. AFAIK, you can't do that with a 44.1 sampling rate, or else you could do it more easily with 96 than you could with 44.1.)

Low integral resamples (88.2 -> 44.1) are quite a bit easier on the math compared to 96->44.1. It shouldn't make an audible difference but the filters need to be much much more complex for the 96 downsample to be legitimate.

The 44.1khz frequency was chosen because the first PCM machines used (drumroll) analog videotape as its media. The math worked out that at 16 bits of resolution you could cram enough samples into a video frame to get 44.1.

The really ironic thing about 192khz is that the SNR numbers are objectively worse compared to 96khz recording (to say nothing of 44.1). Virtually all ADCs are sigma-delta nowadays anyway, and one of their fundamental characteristics is that you can always trade off frequency for resolution. There are a couple hardware vendors who actively speak out against 192khz recording for this reason.

In fact I'd almost argue that 192khz was just as much driven by hardware vendors than audiophiles. They're already trying to cram DxD down studios' necks. 8 bits at 384khz... yuck.

Regarding clipping and samplerate: according to a paper i read recently, a higher samplerate can indeed lower the amount of *analogue clipping*. But this can just as well be achieved on 44khz if its recorded with proper level meters (thus, not overcompressed AND then normalized to the maximum). So, the higher samplerate simply acts as a safeguard against malicious mastering practices. Notice that this will only lower the amount of clipping - dynamics will still suffer from overcompression.

Explanation is as following: even if the digital representation of a signal does not clip, it may still clip when its converted back to analogue. As you know, a digital representation of a waveform can simply be visualized as evenly spaced dots. So, only "snapshots" of the signal are stored, not lines or curves. When this representation goes through the DAC it is reconstructed to the real analogue wave - this however does not simply happen by connecting the "dots" with straight lines - instead it is a spline curve. The implication of this is that even if your signal peaks at 100% in the digital representation, it may still go "over the top" when its reconstructed - because the peak values of the digital snapshots are not equal to peak values of the resulting *analogue* signal. A higher samplingrate can lower this effect - but if proper level meters would be used during mastering, then this wouldn't be necessary at all...... of course, that would mean lowering the loudness a bit.... so.....

>16bit would be necessary if someone wanted to make a recording with dynamics ranging from treshold of hearing to treshold of pain...

While Nyquist states that every signal component below the Nyquist frequency can be preserved realworld filter constraints make the usable bandwidth smaller, CD specs garantee only up to 20Khz. But that is enough. Using higher samplerates though makes implementing the digital reconstruction filter of the DACs less critical, like Axon already mentioned.

On a sidenote; recorders may want higher frequencies if they want the frequencies generated above 22khz (analog) in their production. For instance, I have had interesting results playing back 96khz recordings of cymbals at 44khz. I believe quite some artists use this technique, especially for electronic music.

What do you mean? If you playback anything on 44.1 kHz, you will have no energy content above 22.05 kHz, in fact, due to anti-aliasing filtering the roll-off will even start at a lower frequency.

I've read this also. However, I think that this can only be avoided by using a sampling rate of infinity. Doubling the sampling frequency will reduce the number of clipping samples by a factor of two, or so. But it will still happen.

Second, when a producer records at 96 kHz and sees no clipping on his meters, the clipping effect you explained may occur as soon as he/she resamples back to 44.1 kHz for CD. So, it may be better to record at 44.1 kHz from the beginning in this case.

One more reason to instead simply use proper level meters, and use compressors in a sane way. Actually, just lowering the amount of compression alone would probably make the problem insignificant in practice.

Indeed.

By the way, I thought a little bit longer about this. My question now is: is this clipping effect really a problem in practice? Let's assume the output of a D/A converter is in de range -/+ 1 V. Due to the clipping the output becomes -/+ 1.05 V, for instance. I think no analogue equipment will have a problem with that. Since it is all analogue, the sound will not clip, as it would in the digital domain.

Huu, no. It will be +-1V and not more. The clipping will add distortion because the wave shape no longer corresponds to the original (has additional HF distortion).

Why, because the analogue output of a D/A converter is limited? Why should it be? You may be right, I'm no expert on D/A converters.

edit: We're not talking about conventional clipping here, but clipping due to the D/A conversion, so that's where my question comes from.

I believe this is wrong, or actually, not really explained as corresponding to reality. The DAC will follow the PCM representation as closely as possible, and thus add all kinds of HF distortion. Following up on the DAC should be an (as perfect as possible) lowpass filter, which will remove them again, and leave a "perfect" signal. This follows directly from DSP sampling theory.

The result of this process is more or less as you explain (though it isn't a spline curve, not in the mathemathical sense).

The problem of constructing the ideal as possible lowpass filter is one argument that is used for higher sampling rates (if you up the sampling rate, you can make a crappier filter and still get the same quality audio). It's a rather crappy argument though, because it's not as if we can't cheaply make good performing ones now!

The main reason why sampling rates > 44.1Khz is used is that at the hardware / player, the required transition bandwidth of the analogue low pass anti-aliasing filter is much - much wider and thus a simple low order analogue filter is sufficient. Building a low-order linear phase low pass filter is very simple compare to a high order filter!

It has nothing to do with psychoacoustics but electronics. If the sampling rate is 192 kHz, do not expect to hear above the usual 18 - 20 kHz range.

As for the case of 24 bits, I supposed 16 bits audio isn't enough to represent the entire dynamic range of the human hearing. At its most sensitive region, around 2 - 4 kHz, this dynamic range is actually greater than 100 dB.

Yes, of course it's limited. In practise it'll be limited to the supply voltage, i.e. 3.3V or something thereabouts.

Edit: Or a bit less depending on design etc. But in any case, the output voltage for sure is hard limited, it won't magically swing outside it's maximum range.

I know this argument, but does it have any bearing on practise? I mean, is this still a practical problem anywhere? It certainly doesn't seem to be! Even if we can't get the theorethical resolution because of limitations of real world filters, we certainly seem to be able to do well enough very cheaply that the results are perfect in a psychoacoustical sense.



Yes, and the amount of records that actually need such a dynamic range is? One in ten million?

As I already explained, I do not think that we need a dynamic range so far that the lower end of audibility corresponds with a higher end where the hearing of the listener is damaged by anything but very short term exposure. That's not only useless, it's actually dangerous.

Another thing to consider is that with proper dithering and noiseshaping, the effective resolution at 44.1kHz can be boosted by as much as 15dB. This means that a properly produced CD (unfortunately very few exist, but HDCD are an example) will have an effective SNR of 111dB.

How many ADC or DAC's have you seen that can attain that resolution?