upsampling while mastering?

[wxq99]

First, I would like to thanks Bob Ketz for the great mastering book, thanks for sharing the valuable experience.

While, I read that the higher sample rate "sounds" good is because the inperfect digital filter when the signal pass the A/D and D/A converter. I would agree on this opinion because the rule of nyquist frequency. And for 44.1KHz sample rate, the frequency limit is 22.05KHz and it is possible to record and reproduce the information below it, theoretically. The difference is because the digital filter.

However, bob also mention that that higher sample will reduce the distortion during the post production(the point is actually by J.A.Moorer) since some nonlinear process such as filter, EQ and compression will add less distortion to the signal. I just want to know if anyone have do an objective test to prove the point? I have searched the AES database but not find a paper to explain it. If it is true, should we mastering a 44.1KHz session in 88.2KHz or higher for a better result? Thanks.

[wxq99]

bump, is that abstruseness for people here?

[Bob Olhsson]

I don't think you can generalize. Some digital signal processing sounds better at higher sample rates and some doesn't. Some methods of changing the sample rate cause more harm than doing the signal processing at the lower rate.

This is why very high quality monitoring is crucial as is learning to watch out for tiny gain changes that can fool you into believing something that sounds the same or worse sounds better.

[wxq99]

I know that people at this forum are not all professional but I feel sorry that no one but Bob Olhsson and uros help me. I did a simple test yesterday.

I downloaded a reference 44.1KHZ/32BIT 1Khz test tone, dithered it to 24BIT using UV22. You can see that the signal is very good for test use. I upsampling it to 192K and the FFT shown the SRC I used is good, some new frequency is above 40KHz and will not result in audiable difference

[wxq99]

I tested some compressor plug-ins and the results are very same, so I only put one on. You can see after upsampling to 192KHz, the new frequency imported by the compressor is much lower though the THD do not change.

However, it will improve sound quality significantly. The new frequency around the base tone and 2nd, 3nd harmonic will blur the stereo image which as the same as jitter's effect.

[dcollins]

Originally Posted by wxq99

However, bob also mention that that higher sample will reduce the distortion during the post production(the point is actually by J.A.Moorer) since some nonlinear process such as filter, EQ and compression will add less distortion to the signal. .

EQ and filtering are linear processes, and aren't really helped by oversampling as they should not create harmonics.

Limiting and compression are non-linear processes and DO benefit from oversampling as your graphs show.

DC

[lu432]

Originally Posted by dcollins

EQ and filtering are linear processes, and aren't really helped by oversampling as they should not create harmonics.

Limiting and compression are non-linear processes and DO benefit from oversampling as your graphs show.

DC

Dave thank you for this insight. Could you please explain this further. I am very interested as to the reason why. What do you mean exactly by non-linear? Could you please explain this with more detail?

[Cellotron]

Originally Posted by dcollins

EQ and filtering are linear processes, and aren't really helped by oversampling...

Dave -
I'm going to have disagree with you on this as there are in fact some digital eq designs that take advantage of having less steep anti-aliasing filters during processing after upsampling to provide better sounding results - especially when equalizing higher frequencies - as this allows there to be less "pre-warping" where the filter's shape loses it's symmetry as the frequencies head towards the Nyquist frequency - more info at Equalizer - Sonoris Audio Engineering

Here's a pic of a filter with and without the effects of "pre-warping"



Best regards,
Steve Berson

[wxq99]

It is basic knowledge that both EQ and filter are not linear. Only gain change and summing are linear processing.

[dcollins]

Originally Posted by Cellotron

Here's a pic of a filter with and without the effects of "pre-warping"

Yeah yeah, but I think it's better to look at this from a harmonic-generating perspective.

A much bigger difference than whether you get a particular frequency response without oversampling, imo.

Can't fight Nyquist.


DC

[dcollins]

Originally Posted by wxq99

It is basic knowledge that both EQ and filter are not linear. Only gain change and summing are linear processing.

Really?

DC

[iquinn]

Originally Posted by wxq99

I tested some compressor plug-ins and the results are very same, so I only put one on. You can see after upsampling to 192KHz, the new frequency imported by the compressor is much lower though the THD do not change.

However, it will improve sound quality significantly. The new frequency around the base tone and 2nd, 3nd harmonic will blur the stereo image which as the same as jitter's effect.

if those improvements are happening below -110dbfs, even at loud'ish listening levels, will they realistically translate into audible differences?

also, if your DAC's highest sample rate is only say 96khz, but you're working in 192khz, will the re-sampling at that stage negate those improvements?

[wxq99]

Originally Posted by dcollins

Really?

DC

LP and MP EQ will not import harmonic distortion but they produce none harmonic noise. A good EQ will import very low noise even the sampling rate is at 44.1K. You can do the test yourself.

[wxq99]

Originally Posted by iquinn

if those improvements are happening below -110dbfs, even at loud'ish listening levels, will they realistically translate into audible differences?

also, if your DAC's highest sample rate is only say 96khz, but you're working in 192khz, will the re-sampling at that stage negate those improvements?

-110db is audible if you use some high quality monitor. And dont forget the noise will accumulate during the mixing session, so if you do lot's processing, after mixed the whole session, you may hear the difference.

If you need some analog processing and need D/A or A/D, the resampling may has negative effect if the algorithim is not good, for these, you can check on src.infinitewave.ca. Although the test result varies from what I tested, SRC function in some of the software are really awful. However, as I read in Pro Audio maillist last year, if the coefficient is just 1:2, 1:4, most plug-ins and DAW will work without problem and my test proved the point is correct, so I think if are working on CD medium, you can try upsampling to 176.4KHz and resampling to 88.2KHz during the analog procession if you pay attention to the quality. But upsampling the session only to 88.2KHz and avoid multiple SRC will save a lot of time.

I am going to design an ABX double bind test to see if the improvement is really audible

[iquinn]

Originally Posted by wxq99

-110db is audible if you use some high quality monitor. And dont forget the noise will accumulate during the mixing session, so if you do lot's processing, after mixed the whole session, you may hear the difference.

thanks, i guess i'm thinking of it in a mastering context though.
but even at that, I think my tinnitus kicks in just above -110!

Originally Posted by wxq99

so I think if are working on CD medium, you can try upsampling to 176.4KHz and resampling to 88.2KHz during the analog procession if you pay attention to the quality.

i'm not sure that scenario would work for me as both my analog and digital
processing are running in the same session.
I've already done some tests on tracks were the original sample rate was 44.1, were I upsampled to 88.2 and then ran sessions at both sample rates to produce two 16/44.1 files.
There were certainly differences between the two resulting files, but in the context of the material, i wouldn't say the 88.2 version was always better sounding, only different.
But I appreciate not all SRC's are of a high quality.

[wxq99]

Sorry I messed up. The difference might be very small if using very high quality processsor because some of them will upsampling internally. And I also found some of the A/D and D/A converters are not supposed to working on the sample rate over 48KHz, they produce more distortion when working at higher sampling rate.

Considering the error contributed by the SRC, maybe not all system will really benefit from upsampling?

[Boomshanka]

Originally Posted by wxq99

Considering the error contributed by the SRC, maybe not all system will really benefit from upsampling?

.. as per the third post above, from Bob Olhsson.