The No.1 Website for Pro Audio
88.2 - 96kHz Vs 44.1-48kHz (a thread to end them all!!)
Old 4 weeks ago
  #631
Gear Addict
 
haysonics's Avatar
 

Quote:
Originally Posted by Yannick View Post
And of course there is useful information between two sampling points... this information is stored in the data ! It is reconstructed by the DA converter, only bandwidth limited !!!!!!
How is the useful information inbetween the sample points captured (and stored in the data) if you haven’t sampled it? Sounds like the “reconstruction” is a mathematic assumption of what lies inbetween based on the sampled points.
Old 4 weeks ago
  #632
Not an assumption, rather a reconstruction of everything below the Nyquist frequency. Not a shunning ump to be seen....
Old 4 weeks ago
  #633
Gear Addict
 
haysonics's Avatar
 

Nope, the “reconstruction” is partly assumption unless the information inbetween the sample points is encoded in the sample points. And Nyquist adherents don’t want to entertain that idea so that leaves you with missing information. If that missing information is higher frequencies than we can hear then its not an issue but if it is also missing amplitudes then you have a problem that upping the bit depth wont fix.

The problem is more to do with AD and oversampling was a godsend but not THE END.
Old 4 weeks ago
  #634
Lives for gear
 
chrischoir's Avatar
 

Quote:
Originally Posted by haysonics View Post
Sounds like the “reconstruction” is a mathematic assumption of what lies inbetween based on the sampled points.[/SIZE]
Not at all an assumption. but rather "interpolation" between 2 sample points. Based on frequency and amplitude, a SIN function has a very specific mathematical definition which makes it easily re-calculable with 100% accuracy.
Old 4 weeks ago
  #635
Quote:
Originally Posted by Tom Barnaby View Post
You answer is very useful because it will help us to tackle a key question.

You say that the amplitude in the time domain and the frequency content are the same thing. This may look trivial, but I think it is not.

amplitude variation and frequency content are not the same.
See, this is exactly what the Fourier theorem proves. For any continuous signal, they are the same thing.

Good luck at demonstrating the opposite, and debunking a whole theorem! (that's easily worth a dozen Nobel prices! )

In fact, the Fourier transform is the conversion method between time and frequency domain (and the inverse Fourier transform covering the other way around). These domains are just different perspectives on the same, unambiguous, underlying two dimensional data.



https://en.wikipedia.org/wiki/Time_domain
https://en.wikipedia.org/wiki/Frequency_domain

I think you oversee that the frequency domain holds both info about frequency magnitude and phase magnitude. The Fourier transform yields complex numbers (i.e. numbers made of two values, one representing freq magnitude, the other the phase magnitude), please take a closer look at what it does:

https://en.wikipedia.org/wiki/Fourier_transform


The main difference between time and frequency domain is the X axis: Either absolute time (hence, time domain), or frequency (hence, frequency domain). Further, the frequency domain is really drawn/imagined as two plots: freq magnitude and phase magnitude (remember: Freq domain is expressed in complex numbers).

Last edited by FabienTDR; 4 weeks ago at 05:59 AM..
Old 4 weeks ago
  #636
Gear Guru
 

Quote:
Originally Posted by haysonics View Post
If that missing information is higher frequencies than we can hear
...it is...

Quote:
Originally Posted by haysonics View Post
then its not an issue
...it's not an issue...

Quote:
Originally Posted by haysonics View Post
but if it is also missing amplitudes
...there are frequencies without amplitude???...

Quote:
Originally Posted by haysonics View Post
then you have a problem that upping the bit depth wont fix.
That doesn't even make sense.
Old 4 weeks ago
  #637
Lives for gear
 
Monkey Man's Avatar
 

Quote:
Originally Posted by Tom Barnaby View Post
You made me having a doubt, so I just made a listening comparison between a song recorded at 96 khz and another from a 44.1 khz CD. The more I listen to them and the more I hear the difference.
It sounds like you compared two different recordings, that is, different songs, artists or actual recording sessions. If this is the case the test is of zero value, Tom.

It'd be difficult-enough to discern any differences using carefully-controlled, scientific A/B techniques, let alone when comparing two entirely-different recordings.

Quote:
Originally Posted by Tom Barnaby View Post
44.1 khz sampling means a sampling point every 20 ms or so . Is it right ?
No, it's every 0.023ms, Tom, approximately 1/50th of a ms. Huge difference.

Quote:
Originally Posted by haysonics View Post
If that missing information is higher frequencies than we can hear then its not an issue but if it is also missing amplitudes then you have a problem that upping the bit depth wont fix.
It is information of higher frequency than we can hear, mate; the "useless" stuff is filtered out before the sampling takes place.
Old 4 weeks ago
  #638
Gear Guru
 

Quote:
Originally Posted by FabienTDR View Post
please take a closer look at the Fourier transform:

https://en.wikipedia.org/wiki/Fourier_transform
Not gonna happen. We've been here a million times and some people just won't learn. Or they troll. Whatever. Same crap.

I enjoy reading your posts btw. Very well written and informative.
Old 4 weeks ago
  #639
Here for the gear
 
TomStevens's Avatar
 

This thread is on a never-ending loop with the same old arguments & theories.

There are great reasons to record, mix, & master above 44.1khz, but it's not about more sample points to capture the sound. Whether it's bit depth or frequency range, our ears are limited in what we can perceive.
The real strengths of working at higher sample rates are similar to our reason for recording above 16 bit depth. There are advantages DURING the digital production process which can preserve fidelity and reduce unwanted artifacts.

Here's what one of the industry's best converter designers and best plugin developers have to say on the subject. How much effect any of this will have on your final product really comes down to your workflow and the gear / software you're using. So in reality, what's "best" for you or anyone else depends entirely on your circumstances.

Quote:
Originally Posted by Dan Lavry
88.2 or 96 KHz are preferred rates for audio quality.

Good conversion requires attention to capturing and reproducing the range we hear while filtering and keeping out energy in the frequency range outside of our hearing. At 44.1 KHz sampling the flatness response may be an issue. If each of the elements (microphone, AD, DA and speaker) limit the audio bandwidth to 20 KHz (each causing a 3dB loss at 20 KHz), the combined impact is -12dB at 20 KHz.

At 60 KHz sampling rate, the contribution of AD and DA to any attenuation in the audible range is negligible. Although 60 KHz would be closer to the ideal; given the existing standards, 88.2 KHz and 96 KHz are closest to the optimal sample rate. At 96 KHz sampling rate the theoretical bandwidth is 48 KHz. In designing a real world converter operating at 96 KHz, one ends up with a bandwidth of approximately 40 KHz.
Taken from: Lavry Engineering 2012 white paper - The optimal sample rate for audio quality

Quote:
Originally Posted by andy_cytomic
I recommend tracking at 88.2 and then downsampling to 44.1 for final CD release for 3 main reasons:

1) The aliasing caused by your soundcard's half band filters is moved way up into the inaudible frequencies
2) You will have extra headroom to contain aliasing for all your non-linear plugins
3) Any plugins that require further oversampling can do so with more gentle and cpu friendly oversampling filters, so your audio will sound better, and you will have lower latency
Quote:
Originally Posted by andy_cytomic
Track and mix at 88.2 kHz or 96 kHz

This is a great compromise between file sizes, memory usage, and quality. If you have lots of source material at 44.1 kHz that you want to use without changing its pitch, then 88.2 kHz is probably best since there will be lots of upsampling to do - which is easiest to not screw up if you do multiples of 2, but only one conversion down to 44.1 kHz or 48 kHz.

At 88.2 kHz / 96 kHz you have enough frequency headroom to considerably improve audio quality of most effects plugins, lowering aliasing to inaudible levels. Unless you are doing heavy distortion or compression or other non-linear processing you should be fine.

For those plugins that are generating lots of harmonics then if you are already at 88.2 kHz / 96 kHz the oversampling process is made much easier on cpu and latency than having to go from 44.1 kHz or 48 kHz. This is because the filters involved have to be very steep for this tricky first step, so doing it offline and Smilies
rendering the results is the way to go.

Most hosts and plugins don't handle oversampling, or if they do they don't do it very well.

If you pitch a sample up or down then this is just another name for sample rate conversion. Actually it's harder to do right than a static 44.1 to 88.2 conversion since most of the time the pitch is being modulated by and envelope or lfo. Since many people prefer high voice counts instead of good sounding sample playback quality usually suffers. If you write lots of music that uses sample playback and you change the pitch of samples you should really check into the quality of your sample playback plugins as this could really help the quality of your tracks.

All sample rate conversion introduces latency

This is true of DACs and ADCs as well as fully software based resampling, no exceptions. Some resampling methods sacrifice phase linearity to reduce latency but it is never removed.

Inter-sample wobbles and peaks are always there even though you can't see them most of the time
Old 4 weeks ago
  #640
Lives for gear
Quote:
Originally Posted by mattiasnyc View Post
"Experts" is exactly what those who created the theorem and devices were. Why don't you trust them?
I do think it's more useful to use proper test procedure and learn yourself, rather than blindly trust. If one has the time of course.
Old 4 weeks ago
  #641
Gear Maniac
 

Quote:
Originally Posted by chrischoir View Post
Not at all an assumption. but rather "interpolation" between 2 sample points. Based on frequency and amplitude, a SIN function has a very specific mathematical definition which makes it easily re-calculable with 100% accuracy.
The interpolation process you are talking about only works if the sampled signal behaves like a sine between two sampling points. It must have a sine form in order to be predictable.
I tried to show above that, because of the mismatch of elementary sine waves in the time domain, the behaviour of the global signal was more complex than a sine.
Old 4 weeks ago
  #642
Gear Addict
 
haysonics's Avatar
 

Quote:
Originally Posted by Tom Barnaby View Post
The interpolation process you are talking about only works if the sampled signal behaves like a sine between two sampling points. It must have a sine form in order to be predictable.
I tried to show above that, because of the mismatch of elementary sine waves in the time domain, the behaviour of the global signal was more complex than a sine.
They don’t get it because they are essentially reductionists and they only see frequencies. For them amplitude only relates to frequencies. You cant represent a 3D world to folk that only see 2D and fluctuation between positive and negative values. And that’s OK. It doesn’t have to be them vs us, as long as we are all making great tunes.
Old 4 weeks ago
  #643
Lives for gear
Quote:
Originally Posted by mattiasnyc View Post
Right. When they're added it's ONE signal. ONE waveform.
This is also physical reality and 100% analogue in the actual room of air. Of course already when recorded via a duo of microphone diaphragms there is some alterations being made. Maybe it's those stereo aspects @ Tom Barnaby is after, but as far as I know this has already been addressed many times over and 44.1 kHz does not have time domain issues as far as I know. Which is not a reason to not test nor does it mean under certain limitations there couldn't be such issues, but I think not for the reason of bandwidth limitations.

Just to make it clear, @ Tom Barnaby disbelieves 44.1 kHz reproduction in the consumer format and not only recording of signals. It's way easier to argue for the merits of oversampled recordings, but maybe not for "time domain" reasons. My take is that it makes it easier to mix and if you mix in oversampled territory it makes sense to record with the same sample rate. None of that suggests consumer formats need the same sample rate in order to be able to reproduce the final mix.

Last edited by Mikael B; 4 weeks ago at 01:39 PM..
Old 4 weeks ago
  #644
Lives for gear
Quote:
Originally Posted by Tom Barnaby View Post
,
… you can't ignore that thousands of engineers use 96 khz for recording because they think it sounds more authentic.

Would you call those people trolls ?
Don't put your beliefs into those engineer's mouths. You're claiming that 44.1 kHz REPRODUCTION is flawed. Don't pretend you have support where you don't have it!
Old 4 weeks ago
  #645
Gear Addict
 
haysonics's Avatar
 

Quote:
Originally Posted by mattiasnyc View Post
there are frequencies without amplitude
No but there are more factors involved than just frequency response. There is bit depth and there is time domain. For frequency response, you can limit your time domain to 44.1 and represent your frequency response perfectly. None of us are arguing against that. But you cant limit your time domain to 44.1 and capture all the “depth” no matter what your bit rate is and that is what you don’t get. And that’s OK in my book.

Think about it this way - if you increase your sample rate you increase your time domain accuracy and capture frequencies only bats will appreciate but you will also capture more data (sample more accurately) where your signals are placed in the time domain BUT that won’t make sense to you if you only think about your signals in terms of frequency response. If you can’t get past that I will go back to reading Chopra (0:
Old 4 weeks ago
  #646
Lives for gear
Quote:
Originally Posted by Tom Barnaby View Post
The interpolation process you are talking about only works if the sampled signal behaves like a sine between two sampling points. It must have a sine form in order to be predictable.
No. It needs to be within the frequency range being recorded. Any time domain issues are present already when recording with microphones. After that it's all electric signals converted to and from digital.
Old 4 weeks ago
  #647
Gear Maniac
 

Quote:
Originally Posted by Mikael B View Post
Don't put your beliefs into those engineer's mouths. You're claiming that 44.1 kHz REPRODUCTION is flawed. Don't pretend you have support where you don't have it!
I pretend that 44.1 kHz reproduction is not the most accurate that we can get.
I don't say it is flawed as far frequency reproduction is concerned. No audible frequency is missing and the sound will not be more crystal clear with a higher sample rate, but some nuances are not properly reproduced.

A lot of enginers or music consumers did the same experience when comparing with 96 or 192 kHz. The difference is more audible in the case of 192 kHz reproduction.
Old 4 weeks ago
  #648
Lives for gear
Quote:
Originally Posted by Tom Barnaby View Post
I pretend that 44.1 kHz reproduction is not the most accurate that we can get.
I don't say it is flawed as far frequency reproduction is concerned. No audible frequency is missing and the sound will not be more crystal clear with a higher sample rate, but some nuances are not properly reproduced.

A lot of enginers or music consumers did the same experience when comparing with 96 or 192 kHz. The difference is more audible in the case of 192 kHz reproduction.
I'd like to point out that all of this is testable with analogue nulling of signals. it would likely take some test setup to be able to do this in real time — mostly signal time differences need to be precisely accounted for — but it would be possible, I think.

If one cannot accept the results of (analog) nulling one is denying the physical reality. Because ALL signals behaves like that. It's a characteristic of the physical world.

Time differences between 2 tested signals means they clearly do not null, but that's not within the complexity of the tested signals, but between them. This must be adjusted for so nulling can take place. At least reproduction would be relatively easy to test with two audio interfaces set to different sample rates.



@ Tom Barnaby , are you willing to test? So go do it, publish how you did it so others can criticize your test setup and reproduce the same test and also post your results. We await your scientific contributions to the audio community. Isn't that after all what you set out to do?
Old 4 weeks ago
  #649
Lives for gear
Quote:
Originally Posted by haysonics View Post
No but there are more factors involved than just frequency response. There is bit depth and there is time domain. For frequency response, you can limit your time domain to 44.1 and represent your frequency response perfectly. None of us are arguing against that. But you cant limit your time domain to 44.1 and capture all the “depth” no matter what your bit rate is and that is what you don’t get. And that’s OK in my book.
You're being very ambiguous and cloudy.
Time domain? The changes of the waveform make up for that. Did you know that 44.1 kHz means you take 44100 samples per second? You see that word "second"? What do you think that refers to? That's the time domain. You're late to this. All of this has already been addressed and scientifically tested. I encourage you to make your own test for this and carry them out. Go do it!

Or you'd rather debate people here so we can be impressed? I can assure we're all so impressed with your posts. So you can post about something else now. Mission accomplished.
Old 4 weeks ago
  #650
Lives for gear
Quote:
Originally Posted by haysonics View Post
No but there are more factors involved than just frequency response. There is bit depth and there is time domain. For frequency response, you can limit your time domain to 44.1 and represent your frequency response perfectly. None of us are arguing against that. But you cant limit your time domain to 44.1 and capture all the “depth” no matter what your bit rate is and that is what you don’t get. And that’s OK in my book.

Think about it this way - if you increase your sample rate you increase your time domain accuracy and capture frequencies only bats will appreciate but you will also capture more data about where your signals are placed in the time domain BUT not if you only think about your signals in terms of frequency response. If you can’t get past that them send in the chap with the gif of Alice In Wonderland and i will go back to reading Chopra (0:
Proofs !
Old 4 weeks ago
  #651
Lives for gear
Quote:
Originally Posted by dinococcus View Post
Proofs !
Tests!
Old 4 weeks ago
  #652
Lives for gear
Quote:
Originally Posted by Mikael B View Post
Tests!
With what:The ears?
Old 4 weeks ago
  #653
Gear Guru
 

Quote:
Originally Posted by Mikael B View Post
I do think it's more useful to use proper test procedure and learn yourself, rather than blindly trust. If one has the time of course.
In this case it would mean learning the theorem mathematically and accepting or changing it, then designing the circuitry (hardware) to be able to test it, then designing or finding out what hardware / software to use to test it all, then testing it oneself. If not then again just watch Monty's video.

And if what you're talking about isn't the above but what "experts" think it sounds like then you really up against a bunch of things that need to be addressed in order to make the tests valid - things that I'm willing to bet a lot of these "experts" have failed to address themselves.

Just because someone is an excellent mix engineer with records under one's name doesn't make one any less human and prone to being.. well.. wrong.
Old 4 weeks ago
  #654
Gear Guru
 

Quote:
Originally Posted by haysonics View Post
No but there are more factors involved than just frequency response. There is bit depth and there is time domain. For frequency response, you can limit your time domain to 44.1 and represent your frequency response perfectly. None of us are arguing against that.
Some of you are arguing against exactly that.

Quote:
Originally Posted by haysonics View Post
But you cant limit your time domain to 44.1 and capture all the “depthno matter what your bit rate is and that is what you don’t get. And that’s OK in my book.

if you increase your sample rate you increase your time domain accuracy and capture frequencies only bats will appreciate but you will also capture more data (sample more accurately) where your signals are placed in the time domain BUT that won’t make sense to you if you only think about your signals in terms of frequency response.
How - specifically - is a 1kHz sine wave captured with more "depth" using a 96kHz sample rate compared to a sample rate of 20kHz?

Like I said, please be specific in your answer.
Old 4 weeks ago
  #655
Gear Guru
 

Quote:
Originally Posted by haysonics View Post
You cant represent a 3D world to folk that only see 2D. For them it is just a fluctuation between positive and negative values.
One parameter is positive and negative values of the electrical current.

The other is those values over time.

What are the other physical parameters or "values" involved when sampling an electrical audio signal?

Please be specific and point out the physics of it all.
Old 4 weeks ago
  #656
Lives for gear
 
Yannick's Avatar
 

Quote:
Originally Posted by haysonics View Post
Think about it this way - if you increase your sample rate you increase your time domain accuracy and capture frequencies only bats will appreciate but you will also capture more data (sample more accurately) where your signals are placed in the time domain BUT that won’t make sense to you if you only think about your signals in terms of frequency response. If you can’t get past that then send in your gimp with the gif of Alice In Wonderland and i will go back to reading Chopra (0:
If you cant get past the common knowledge that the time domain accuracy of 44.1 k sampling rate is much higher than the samplerate, you can just stop here.

Go outside and buy a book and then study it !

For the really hardheaded in here: do you really think a 20khz sine wave can only be sampled in ONE position ? So you are really arguing that the phase shift of a AD converter is up to 90 or even 180 degrees ? Did any of you look at the video posted above ? The guy even goes so far to use 100% analogue oscilloscopes to prove to all foolish harts out there that there is time resolution BETWEEN the sample points !

In short: if you increase samplerate you increase bandwidth, NOT the time resolution within the previous audio band.
Old 4 weeks ago
  #657
Gear Maniac
 

Quote:
Originally Posted by Yannick View Post
If you cant get past the common knowledge that the time domain accuracy of 44.1 k sampling rate is much higher than the samplerate, you can just stop here.

Go outside and buy a book and then study it !

For the really hardheaded in here: do you really think a 20khz sine wave can only be sampled in ONE position ? So you are really arguing that the phase shift of a AD converter is up to 90 or even 180 degrees ? Did any of you look at the video posted above ? The guy even goes so far to use 100% analogue oscilloscopes to prove to all foolish harts out there that there is time resolution BETWEEN the sample points !

In short: if you increase samplerate you increase bandwidth, NOT the time resolution within the previous audio band.

Instead always reading books, you should listen to the place of instruments in the audio space and ask yourself how present they feel.
This might help you in comparing recordings made at various sampling rates.
A good part of the difference can be found in this area.
Old 4 weeks ago
  #658
Lives for gear
Listen is not sharable. Everybody can tell what he want In good faith.
Old 4 weeks ago
  #659
Lives for gear
 
Yannick's Avatar
 

Come on man, we are all trying to be civilised here.

I have been doing exactly that my entire career, choosing microphones, placement, mic preamps, even cables (...). I have done the same tests with my gear concerning samplerates, the only discernable difference I got was recording at 32 KHz

How dare you come here, question well known and understood science as if you are onto something new, vomiting a continuous train of invalid thoughts, and then proposing how we should organize ourselves to start and make decent recordings ?

I am really starting to ask myself if you can even find the right side of a microphone or, dare I say it, find the record button.

I put you on my ignore list, so won't be responding to your uterlly useless posts anymore. If someone on the forum would point out to me that you have read and studied some stuff, maybe I'll read your posts again ... Bye.
Old 4 weeks ago
  #660
Gear Head
 
studio96's Avatar
>20khz sine wave can only be sampled in ONE position ?

Two last points: The issue with the theorems of Nyquist and Fourrier is, that both represenation and reconstruction assume theoretically ideal (endless) waves. With such a sine wave, one easily could reconstruct a 20kHz wave with only 40.001kHz sample rate because the AA filter will overcome it's transient behaviour and go into a steady state. Then (and only then) the output is stable at a given amplitude and phase delay. The same can be said with FFT processing by Fourier.

But this is not what music is about. In reality the frequencies in an audio stream quickly change in amplitude and this regularly make Fourier interpreation going false and artifact-free reconstruction of the wave by e.g. I-FFT nearly impossible! One would need either an infinite Fourier analysis window or an infinite density of the waves. Both is contradiciting.

Mapping this "theory" to the behaviour of both the incoming wave and the response of the anti aliasing filter, one quickly recognizes easily, that FIR-filters suffer from the same issue: We have to choose a compromise and never will reconstruct both amplitude and phase for all freqs the right way.

The commonly used IIR-filter (classical analog filters with RLC) thus are always in a transient operation mode an never reach a steady state. This means they "do what they want" or let's say it more precisely: They operate the incoming digital data according to their current state which is caused by the history of the former data. Well, there are methods to digitally preprocess the data in order to control the filters' behaviour in a better way, but this is limited.

These are the reasons, why frequencies with a large distance to Nyquist can be reconstructed most accurately and higher frequencies not as well. The higher the worse.

Of course you will get something out of the filter and it might sound good, but it is not, what was recorded.

Funnily some recordings with low sample rates and low quality players introduce high frequent artifacts that way, that the ear recognizes more "brilliance" like you get it from exiters. Large parts of the audience like that and rate such recordings as "better". This is a often observable issue .
Topic:
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump