Search form

[CS70]

Theory says that the higher rates are better.

No, the theory doesn't.

[hobbyist]

Theory says that the higher rates are better.

No, the theory doesn't.

Okay -- Theory , with a perfect recovery is perfect.

But there are always errors in real life ; and in real life higher sampling rates could benefit the accuracy of the result IF the A/D errors and clock jitter are small enough.

[Hugh Robjohns]

...in real life higher sampling rates could benefit the accuracy of the result IF the A/D errors and clock jitter are small enough.

My bovine excreta alarms are sounding here...

Perhaps it would help if you could clarify what you mean by 'accuracy'.

Do you have any credible references for this enhanced sampling theory?

H

[James Perrett]

But there are always errors in real life ; and in real life higher sampling rates could benefit the accuracy of the result IF the A/D errors and clock jitter are small enough.

As I asked before - can you explain this point a little more? While the need for sample rates slightly higher than 44.1 or 48kHz is well established, I don't quite understand why extremely high sample rates are required for a system where the highest frequency required to be sampled is of the order of 20kHz.

[hobbyist]

But there are always errors in real life ; and in real life higher sampling rates could benefit the accuracy of the result IF the A/D errors and clock jitter are small enough.

As I asked before - can you explain this point a little more? While the need for sample rates slightly higher than 44.1 or 48kHz is well established, I don't quite understand why extremely high sample rates are required for a system where the highest frequency required to be sampled is of the order of 20kHz.

Because the real physical reconstruction of the analogue signal is not perfect.

There are errors in the timing due to clock jitter, and errors in the amplitude due to A/D/A errors. There is not just a mathematical point that is being used.

This changes what was a nice neat point to a smooshed blob so during the recreation that makes it harder to recreate perfectly.

[Eddy Deegan]

Because the real physical reconstruction of the analogue signal is not perfect.

It doesn't have to be. It only needs to be good enough to accurately capture enough of the waveform such that the reconstructed result is as subjectively good to the listener as the original anlogue signal was.

In the case of human hearing, 44.1Khz is more than sufficient in the vast majority of cases. At this point resolution is more important than frequency and 24-bit is more than enough for the task. Though I may be wrong here my understanding is that 24-bit resolution is effectively 20 to 21-bit due to the noise floor.

For some applications higher rates may be useful, but that's what oversampling is for and as per another recent thread about plugins this is a well understood mechanism which can be catered for within normal digital environments.

[Hugh Robjohns]

Because the real physical reconstruction of the analogue signal is not perfect.

That's true... but then nothing is perfect. The reconstructed analogue signal from a tape recorder or vinyl disc player ain't perfect either. Noise and distortion will always manifest to some degree. The only question is how small do the errors need to be to be inaudible... and we know the answer to that.

There are errors in the timing due to clock jitter, and errors in the amplitude due to A/D/A errors.

Yes... but neither are magically solved by a higher system word-clock rate. Actually both are barely measurable these days, and they are definitely well within the acceptable limits such that they are far below audibility in any conventional application. Quantising amplitude errors were effectively eliminated with the introduction of delta-sigma converters, and system clock jitter was wrestled under control by improved clock generator and clock recovery designs long ago.

Added to which, if the jitter element is random it manifests as a (fractionally) increased HF noise floor, which is hardly going to upset anyone even if they could hear it! If it's not random (but instead related some some periodic function) then the circuit design needs to be reworked.

Either way, though, a higher system word-clock rate means that any inherent clock jitter becomes a significantly higher proportion of the sampling period, and this makes the problem much worse, not much better!

This changes what was a nice neat point to a smooshed blob so during the recreation that makes it harder to recreate perfectly.

Oops... there goes that alarm again...

H

[CS70]

Theory says that the higher rates are better.

No, the theory doesn't.

Okay -- Theory , with a perfect recovery is perfect.

But there are always errors in real life ; and in real life higher sampling rates could benefit the accuracy of the result IF the A/D errors and clock jitter are small enough.

Yes, the math is clear: sampling at double the Nyquist rate maintains all the information in the sampled signal, so the theoretical reconstruction is perfect. There is no information loss. So you don't need anything more.

As of these bothersome engineers, yeah the poor fellows don't seem to be able to mach the beauty and perfection of the math. How annoying. But you gotta understand why and what!

For example, increased sample rate require a faster clock! And the faster the clock, the more small inaccuracies in timing become significant. So the higher the sample rate, the more accurate the clocking must be, which as the Romans found out, is quite a lot harder than making a less accurate one. So jitter is a bigger problem at higher sample rates: with the same components and tolerances, you're likely to have a more error in the reconstructed signal, not less - read: worse sound.

As Hugh can explain much better than I, the main issue is that real world lo-pass filtering (which is required to limit the signal to the 20-20K band) is not as perfect as it should be. This in turn may allow some ultrasonic content to alias back in the hearing band. Most mics roll off anyway pretty sharply after 20K, but of course there's some recorded signal that may have content there. 96KHz may help there. Lots of plugins do oversampling for that reason. Tough it helps only if you have ultrasonic content in the first place. I remember he mentioning drum cymbals as a typical source which can benefit (as it's more likely to produce significant ultrasonic energy, which survives the mic roll off attenuation, and may thus get aliased back in the hearing band).

On a practical view tough, I never heard anybody say "hey these cymbals sound odd, probably they've been recorded at 44.1Khz"

Also, keep in mind that 96KHz may put the analogue part of your chain under more strain - and especially with lower end kit, most of the testing has probably been done at the "regular" sample rates. To go over 96Khz for normal musical applications is ... not well thought thru

Unless it's marketing of course - preying on the honest desire for quality of customers to sell them stuff and even worse, convincing them that the bs is real..

[Ariosto]

So would the experts at SOS recommend 96KHz rather than 44.1KHz (or 48K)? At present I record at 44.1KHz and 24 bit.

I just want to record with the best possible quality and not worry too much about the technicalities, and leave that sort of thing to the really clever people who do such a great job with such brilliant engineering and the data it produces.

[CS70]

So would the experts at SOS recommend 96KHz rather than 44.1KHz (or 48K)? At present I record at 44.1KHz and 24 bit.

I just want to record with the best possible quality and not worry too much about the technicalities, and leave that sort of thing to the really clever people who do such a great job with such brilliant engineering and the data it produces.

Recording drums or anything with potential ultrasonic content (i.e. over 20Khz) yes. But just record, then you can still downsample in the DAW to 44.1 or 48 to keep file sizes manageable (as if there's an issue, it's the hardware filtering in the A/D converters).

But vocals or guitars, I wouldn't bother. A symphonic orchestra, yes.. That gong there..

[Hugh Robjohns]

So would the experts at SOS recommend 96KHz rather than 44.1KHz (or 48K)? At present I record at 44.1KHz and 24 bit.

I generally record everything at 96kHz....but mostly just because I can rather than because I am concerned about the 'quality' at lower rates -- although I do like the idea of my mics controlling the audio bandwidth, rather than a non-Nyquist filter in a converter chip. It pleases my engineering head that way...

However, I've recorded plenty of stuff at 44.1 and 48, and still do when appropriate, and I've never thought doing so had reduced the 'quality', so if 44.1k works for you, your clients and your workflow, stick with it. It avoids all that faffing about with down-sampling, reduces the archive storage demands, and makes file transfers quicker.

As always, let your ears be the final arbiter -- and trust them over all the Internet myths and BS!

H

[N i g e l]

Theory says that the higher rates are better.
No, the theory doesn't.
Okay -- Theory , with a perfect recovery is perfect.

For *practical* sample reconstruction, maths predicts a low pass filter effect on the source. The filter cut off is related to the sampling frequency, so the higher the sampling rate, the smaller the dropoff on a specific reconstructed waveform.

Because the [sampling theory] output sample stream pulses are moments in time [width tending to zero], they would have very little energy [amplitude x width] which is impractical for an electronic system to handle.

An easy work around is to hold the sample value from one sample instant to the next – a “zero order hold”. This is basically filling in the values between samples with a simple interpolation [drawing a horizontal line – which gives a “stepped waveform”]. The frequency response of the zero order hold is a sinc(x) function.

Better interpolations are available at the expense of more complex electronics -

1st order – linear interpolation, joining the sample points with a sloping line.
2nd, 3rd order and above.... increasingly complex curve fitting between samples.

Early DAC systems upped the sample rate by say holding the sample value for only ¼ period and adding 3 extra samples of zero value in the remaing ¾ period gap.
Todays 1 bit DACs use complex nth order filters. e.g. n=5

[Hugh Robjohns]

An easy work around is to hold the sample value from one sample instant to the next – a “zero order hold”.

I think you're talking about the non-return-to-zero (NRZ) approach where the sampling pulses are the same width as the sampling period -- and that does result in a windowing effect which cases a very small amount of roll-off at the high end of the amplitude response... but it's nothing that is going to cause anyone sleepless nights! -0.1dB at 20kHz is perfectly acceptable...

Better interpolations are available at the expense of more complex electronics

Yes, there are alternative sampling schemes, but I don't think anyone has successfully demonstrated a genuine requirement of benefit with them in audio applications... yet.

H

[Ariosto]

Thanks for the helpful reply Hugh. I will just experiment with 96KHz but probably go back to either 44.1KHz or 48KHz - unless I can actually hear a difference, which I probably won't!

[Kwackman]

So would the experts at SOS recommend 96KHz rather than 44.1KHz (or 48K)? At present I record at 44.1KHz and 24 bit.

I am NOT an expert (and this isn't false humility!) I record 24bit and 44.1 KHz. The 44.1 KHz choice is for a couple of reasons that may be incorrect, irrelevant or out of date or all of the above.
1. If I burn to CD I don't need SRC.
2. The main virtual instruments I use are recorded at 44.1 KHz, so I don't want to place more stress on my CPU making it do hard sums.

[Hugh Robjohns]

Makes perfect sense to me!

[ConcertinaChap]

It all ends up at 44.1KHz anyway or it does for me, CDs still being the largest part of what I produce. I don't pretend to have golden ears so since I can't hear any difference I take comfort in the Nyquist argument and carry on contentedly.

CC

[CS70]

It all ends up at 44.1KHz anyway or it does for me, CDs still being the largest part of what I produce. I don't pretend to have golden ears so since I can't hear any difference I take comfort in the Nyquist argument and carry on contentedly.

CC

Just for clarity' sake, the main argument for 96Khz is about recording, not anything downstream (unless it's outboard).

During the physical A/D conversion (that happens just the once) and then less than perfect filters, ultrasonic energy of sufficient level could alias back in the 20-20k range and be imprinted in the recorded samples, so that when the signal is reconstructed, it's audible.

Once the samples are in the bag (or on the disk ) you can convert to a lower rate in software and - if the conversion algorithm is good - you have no probs and smaller files.

The same might happen when an effect (a plugin) produces new frequencies, some of them potentially ultrasonic - hence internal oversampling in plugins.
For the same reason, a plugin which does not take care of this might introduce aliasing artifacts in its output. So there is a tiny argument for working at 96KHz, but in practice most modern reputable plugins do just fine. And you ears are the final judge anyway.

[Wonks]

I’ve just re-read that article Hugh and wondered if the ‘50 feet’ was a typo

Not sure. I've checked my original copy text from 2004 and it says 50 feet in that, so if it was a typo it was mine originally... but I can't believe I would have included a fact like that without having a reliable reference for it because 50 feet is obviously a lot!

I've removed that number from the online article for now, and will try to get some confirmation as to how thick the isolating pads actually are.

H

I'm trying to find out, as my company (Arup) were the structural engineers and Arup Acoustics were also involved in some aspects of the design (not the concert hall design itself). I can remember reading a synopsis of the construction including the anti-vibration design at the time in an internal publication. But it was finished back in 1991, so almost certainly no viable digital copy will remain unless the hard copies have since been scanned.

[MOF]

Hugh Robjohns wrote:
MOF wrote:
I’ve just re-read that article Hugh and wondered if the ‘50 feet’ was a typo

Not sure. I've checked my original copy text from 2004 and it says 50 feet in that, so if it was a typo it was mine originally... but I can't believe I would have included a fact like that without having a reliable reference for it because 50 feet is obviously a lot!......H

by Wonks » Fri Jul 12, 2019 12:24 pm
I'm trying to find out, as my company (Arup) were the structural engineers and Arup Acoustics were also involved in some aspects of the design (not the concert hall design itself). I can remember reading a synopsis of the construction including the anti-vibration design at the time in an internal publication. But it was finished back in 1991, so almost certainly no viable digital copy will remain unless the hard copies have since been scanned.
Top

My understanding of this type of construction is heavy duty coiled springs with neoprene rubber on top (and bottom?) are used between the real floor and the floating floor. Maybe Hugh put a ‘ instead of a “ (plus springs) or maybe it was 50 cm or maybe it was just 5’.