• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

I tend to prefer slow roll offs and even the NOS mode - what does this say about my setup and preference? (Conclusion: back to fast linear)

OP
anphex

anphex

Addicted to Fun and Learning
Forum Donor
Joined
May 14, 2021
Messages
662
Likes
870
Location
Berlin, Germany
Oh boy, I didn't notice this article back then -_-
What a gold mine of knowledge. Thank you, Will try out L-Fast for a while.
Kind of ironic since something always bothered me when I was trying out the Gustard X16 at first. It's the reason I started this whole thread and switcharoo.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,223
Likes
17,800
Location
Netherlands
This example is contrived and unrealistic, because the source signal is heavily compressed and distorted. When the source is that messed up to begin with, how much can it matter what kind of filter you use? Decorating a turd.
Contrived to show the differences, yes! Unrealistic? Probably not. Heavily compressed and distorted is the hallmark of the modern music producer (and has been now for at least two decades).

But yes, best to stay away from such abominations as much as possible.. sadly that seems to be quite a challenge nowadays.
 
OP
anphex

anphex

Addicted to Fun and Learning
Forum Donor
Joined
May 14, 2021
Messages
662
Likes
870
Location
Berlin, Germany
Hey guys, I am back to NOS again haha.

New DAC.
New power amps.

And again I went back and forth between fast linear, slow linear and NOS. Keep in my that I would never use NOS without severe oversampling, currently 384 Khz @ 32 bit. If Windows would allow more I'd run it even at 768 Khz.

With this settings it's just nicer than the sound you get with a reconstruction filter. If I had to explain it, it's clear without being annoyingly bright, cleaner transients. Maybe this is what all vinyl purist are trying to achieve? A non-digital sound?

I know, it's weird and nonorthodox. Please be gentle with your pitchforks.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,223
Likes
17,800
Location
Netherlands
Well, you know the drill by now... Do a double-blind test ;)
 
OP
anphex

anphex

Addicted to Fun and Learning
Forum Donor
Joined
May 14, 2021
Messages
662
Likes
870
Location
Berlin, Germany
Can I even do that alone without biasing myself? I need someone to try to trick me, right?
Other than that I can already with confidence say that I can hear the different filters due to the "brightness". The steeper the filter, the brighter the sound. I know, I know, now you could say that it's due to a NOS very early roll off... but at 384 Khz?
After switching back from NOS @ 32/384 to fast linear it feels like there was shaped "white noise" added to the top end of the signal.

Note that I am not endorsing R2R or NOS by itself, but questioning the necessity of a reconstruction filter when having the option of extremely high sampling rates.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,223
Likes
17,800
Location
Netherlands
Isn't the Windows upsampling supposed to be super shit? Possibly you just like that...

 
OP
anphex

anphex

Addicted to Fun and Learning
Forum Donor
Joined
May 14, 2021
Messages
662
Likes
870
Location
Berlin, Germany
Interesting! I will be looking around for a DSP software that intercepts the signal, does the "nice" sinc interpolation and then sends it back upsampled to windows. There should be something like that around.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,223
Likes
17,800
Location
Netherlands
Interesting! I will be looking around for a DSP software that intercepts the signal, does the "nice" sinc interpolation and then sends it back upsampled to windows. There should be something like that around.
Here is another fun one:
1655144251230.png

There should be some useful info here:

 
OP
anphex

anphex

Addicted to Fun and Learning
Forum Donor
Joined
May 14, 2021
Messages
662
Likes
870
Location
Berlin, Germany
Here is another fun one:
View attachment 212568
There should be some useful info here:

Hmm considering this post, using EAPO with the correct installation should already have disabled the nasty APO from windows that causes all those issues. It's already running for my FIR-filtering to fix the room modes, then followed by the loudness curve module.

1655148609230.png
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
... But yes, best to stay away from such abominations as much as possible.. sadly that seems to be quite a challenge nowadays.
Friends don't let friends listen to music that has been squashed with heavy dynamic compression.

If those who make and master these recordings are going to do that, they might as well convert it to 8-bit when they're done. At least that would be honest. Given how consistently this heavy dynamic compression is used with most modern music, I wouldn't be surprised if 8-bit became trendy. They could call it "green", using less bandwidth to deliver the music, which saves processing power and electricity.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,223
Likes
17,800
Location
Netherlands
I wouldn't be surprised if 8-bit became trendy. They could call it "green", using less bandwidth to deliver the music, which saves processing power and electricity.
8-bit 96 Khz, heavily dithered into HF should work pretty well actually ;)
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
... 44.1 using a normal filter will have nothing above 22kHz (24kHz with some filters) but also when upsampled to say 176.4 kHz the audio bandwidth will just be 22kHz with nothing present till 88kHz.
When you put this through a NOS filterless DAC or a filtered DAC the actual output will be the same except for some small amplitude noise far, far above the audible range which no normal transducer will ever be able to reproduce.
Most DACs oversample CD quality (44.1 kHz) by at least 4x or more. However, we see in Amir's DAC tests that a NOS or lazy filter still passes plenty of HF noise as "low" as 22.05 kHz and up. If oversampling 4x to 176.4 means no noise below 88.2 kHz (for example), then where is that HF noise coming from?
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
Its the filter that does this and Amir measures this at 44.1kHz.
In recorded music there should be no signal over 20kHz so any aliasing should only exist above 24kHz.

The so called 'NOS emulating' or slow filters do not adhere to the sampling theorem. Using it (and risking HF noise) is a choice of the user.
That choice is usually based on looking at plots of square-waves or needle pulse test signals (which are never present in recordings) thinking they have 'superior' impulse response in audio.
The weird part is that such 'speed' in music resides in the 2-5kHz region so any perceived 'improvements' in attack cannot be caused by the filter.
In fact treble above 5kHz may be a bit too low occasionally.

When upsampling one needs a filter (usually sharp) which removes noise between 24kHz and say 100kHz but above the new sampling frequency there will be noise again.
That noise, however, will be attenuated much by the transducers.
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
I understand these NOS filters are incorrect or poorly engineered. However:
... In recorded music there should be no signal over 20kHz so any aliasing should only exist above 24kHz.
I have several recordings having energy above 20 kHz. I've also recorded some myself. It's not all that unusual, especially for close-miced recordings of castanets and other small percussion like instruments. Sometimes trumpet and a few other instruments. Of course that doesn't mean it's audible, but the energy is there, at least in some recordings depending on the mics used and positioning.
The weird part is that such 'speed' in music resides in the 2-5kHz region so any perceived 'improvements' in attack cannot be caused by the filter.
In fact treble above 5kHz may be a bit too low occasionally.
IME, ABX testing how filters affect high frequencies, from 12 kHz on up the difference is not tonality (as it couldn't be, since the harmonics are supersonic) but rather the timing or transients sound smeared when listening to castanets, jangling keys, etc. So for me personally, "speed" or timing consists in those frequencies near the upper thresholds of perception. Not 2-5 kHz, but above 12 kHz.
The so called 'NOS emulating' or slow filters do not adhere to the sampling theorem. Using it (and risking HF noise) is a choice of the user. ...
Here's where I have a question. Intuitively, without a filter, you're reconstructing little stair-steps defined by the samples. Each of those has infinitely sharp corners (discontinuous 1st derivative) which cannot occur in nature. Each of those corners forces a spray of high frequencies into the spectrum to form that sharp transition. The sample rate determines the highest frequency the samples can encode, and thus the lowest frequency of noise that must be filtered. If you oversample 44.1 4x to 176.4, then Nyquist is 88.2 kHz, so any noise must be > 88.2 kHz. For that noise to alias in to the passband, it would have to be at least (88200 - 20000) + 20000 = 156,400 Hz.

If so, then oversampling at high rates would obviate the need for a filter, since the HF noise would be at frequencies above the bandwidth of the amplifier. What am I missing here?
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
Using oversampling moves the noise upwards. Smaller steps in amplitude and the lowest frequency part of the steps is moved upwards. The HF noise is merely lower in amplitude as there are steps interpolated between the original samples.

Do you know of any over-samplers that have no filtering and only interpolate 'steps' between the original samples ?
Most over-samplers use at least decent filtering. This means only small 'steps' will be above Nyquist.
Those NOS DACs always have some low pass filtering going on anyway and the transducers themselves drop off above 30kHz (some higher, some lower) and thus 50kHz content will not even be present or too low in level to be harmful.

Oversampling, for most folks who are into NOS totally defeats the purpose of using one but technically vastly improves performance.

Filterless NOS is not adhering to the sampling theorem and delivers quite distorted signals some folks seem to like.
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
Using oversampling moves the noise upwards. Smaller steps in amplitude and the lowest frequency part of the steps is moved upwards. The HF noise is merely lower in amplitude as there are steps interpolated between the original samples.

Do you know of any over-samplers that have no filtering and only interpolate 'steps' between the original samples ?
Most over-samplers use at least decent filtering. This means only small 'steps' will be above Nyquist.
...
I know that, but it doesn't answer my question. The answer is related to the fact that filtering is an essential part of sampling theory. My question is exactly "how" this comes into play with proper oversampling methods. I need to study this further to articulate that.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,223
Likes
17,800
Location
Netherlands
@MRC01 maybe you can rephrase your question? I tried grasping what your after in several posts, but I don't exactly understand yet.

Are you asking if using NOS would be okay if oversampling to a higher rate first? Obviously pushing the noise into the HF is the point of oversampling in the first place, lowering the complexity of subsequent filtering.
 

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
...Are you asking if using NOS would be okay if oversampling to a higher rate first? Obviously pushing the noise into the HF is the point of oversampling in the first place, lowering the complexity of subsequent filtering.
Yes. My example above indicates that oversampling CD quality (44.1k) by 4x should have no audio noise below 88.2 kHz. And virtually all DACs internally oversample at least that much, usually more. If so, then why, when Amir measures them, do the NOS filters show noise from 22.05 kHz and higher? Should be impossible.

I believe the answer has to do with filtering. But I want a more precise answer than that. I need to study the method/algorithm for oversampling to know how/why this is the case.
 

voodooless

Grand Contributor
Forum Donor
Joined
Jun 16, 2020
Messages
10,223
Likes
17,800
Location
Netherlands
Yes. My example above indicates that oversampling CD quality (44.1k) by 4x should have no audio noise below 88.2 kHz. And virtually all DACs internally oversample at least that much, usually more. If so, then why, when Amir measures them, do the NOS filters show noise from 22.05 kHz and higher? Should be impossible.
Because those "NOS filters" emulate the NOS behavior. Basically (in most cases) the oversampling filter is made to suck and leak like a sieve ;)
I believe the answer has to do with filtering.
Indeed ;)

Edit: some more fun: https://www.audiosciencereview.com/forum/index.php?threads/nos-dac-in-band-distortion.11231/
 
Last edited:

MRC01

Major Contributor
Joined
Feb 5, 2019
Messages
3,424
Likes
4,030
Location
Pacific Northwest
Saw that already, and it's not the first time. Totally aware of that.
I'm looking to put my math/programming hat on and dig into oversampling method/algorithm to better understand exactly how/why the artifacts arise. Sure, I already know it's due to the Fourier transform of sharp corners introducing high frequency components. I've done those transforms, seen those results. And how the aliasing folds back like a mirror. The questions are (1) what exactly distinguishes the HF noise arising from those spurious frequencies, from the HF information encoded by the samples and (2) how does oversampling shift that distinguishing line?
 
Top Bottom