[Digital Enigma]

The filters used on most DACs are sufficient, because most DACs (excepting NOS) are audibly transparent. If you can’t hear it, it doesn’t matter.

So slow filters are also adequate because the aliasing is below audibility?

 

[gregorio]

When I tried adding a high shelf from 15kHz up to pink noise with EQ, I noticed noise that was clearly below 15kHz being added to the signal. After listening to an example of aliasing happen in production and how engineers control aliasing by using higher sampling rates to push the nyquist frequency higher, I compared that to what I was getting on my DAP and think that's what's happening here.

Think about that for a minute, what is the bandwidth of that pink noise you’ve generated at 44.1kFs/s? …. So, as there is no signal above 22.05kHz, you are boosting nothing above 22.05kHz, which results in nothing above 22.05kHz! How can that be aliasing? The only way it could potentially be aliasing is if it’s a modelled EQ, if it has non-linear distortion or is emulating non-linear distortion and is ancient or programmed by an amateur (who failed to employ oversampling). Almost certainly what’s actually happening is clipping, either in the EQ itself, something else downstream or inter-sample peak clipping, although it could possibly be IMD in the amp or transducers.

Since I got aliasing from that high shelf …

Aliasing can ONLY occur when recording using an insufficiently high sample rate, so in an ADC or when resampling to an insufficiently high sample rate. You’ve made an assumption that it’s aliasing because it sounds the same/similar as clipping and then everything that follows is based on that assumption, a classic audiophile trap!

So slow filters are also adequate because the aliasing is below audibility?

There is no aliasing in the DAC process, with slow or any other sort of anti-imaging filters. So as there is no aliasing then “yes”, the aliasing is definitely below audibility.

G

 

[Digital Enigma]

There is no aliasing in the DAC process, with slow or any other sort of anti-imaging filters. So as there is no aliasing then “yes”, the aliasing is definitely below audibility.

G

Yes as I understand, aliasing is created at the recording and mixing stage if not done correctly but even if there is aliasing that a slow filter will let through then it’s below audibility. That would explain why people use NOS and don’t complain about the sound.

 

[VNandor]

To spell out the difference between aliasing and imaging in more of layman's terms:

Aliasing is what could happen during recording when the recorded signal has a higher frequency than the nyquist frequency due to a bad anti-aliasing filter. When this happens, the higher frequency wraps back into the audible range instead of staying in the ultrasonic range. The consequences of a really bad anti-aliasing filter is an error signal in the audible range. If the nyquist frequency is at let's say 22.05kHz but the anti-aliasing filter only reaches the desired attenuation at 24kHz, the aliased signal will still stay above 20kHz, leaving the audible range completely untouched.

Imaging is what could happen during playback if the reconstruction filter isn't properly attenuating the signal above the nyquist frequency. The error signal in this case shows up above the nyquist frequency, well outside of the audible range. An anti-imaging filter with a poor attenuation will only cause problems if the gear after the DAC copes especially bad with ultrasonic noise and creates distortion in the audible range due to the extra harmonics in the >20kHz range.

The takeaway is that aliasing affects the frequency band where music is in a straight and direct way, while imaging only affects it in a nondirect way, if at all.

 

[gregorio]

Yes as I understand, aliasing is created at the recording and mixing stage if not done correctly but even if there is aliasing that a slow filter will let through then it’s below audibility.

When recording, it is done correctly, there’s no option not to do it correctly. I did hear some ADCs that didn’t do it correctly but that was when prosumer ADCs first started appearing in the early 1990’s. I haven’t encountered aliasing in even cheap prosumer ADCs for about 25 years or so and certainly not in pro ADCs. It is possible during mixing/mastering to induce aliasing, but it only occurs with certain processors and we watch out for it, in most cases where it could occur the processor automatically oversamples to avoid it. So if there is any audible aliasing in the master then it’s almost certainly there deliberately, unless it’s not professionally mixed or mastered.

That would explain why people use NOS and don’t complain about the sound.

How would it explain that? How does not having a filter remove aliases?

G

 

[bigshot]

People with NOS DACs don't complain because their sound is already degraded. They don't care about fidelity.

 

[bigshot]

So slow filters are also adequate because the aliasing is below audibility?

Any DAC that produces sound that is audibly transparent is adequate. Almost all of them do that... but not NOS DACs and ones with deliberately colored filtering. Thankfully, they're a tiny subset of the market. You can throw a dart at all the DACs on Amazon and be pretty sure you're going to get an audibly transparent one.

 

[Digital Enigma]

Ok so slow filter inaudible aliasing artefacts
Nos audible aliasing artefacts?

 

[Digital Enigma]

How would it explain that? How does not having a filter remove aliases?

G

I didn’t say it removed them I assumed it would be too low in level to be a problem with all this talk of -40db being enough.

 

[Digital Enigma]

I have a dragonfly cobalt floating around somewhere and remember seeing archimagos measurements and that uses a slow minimum phase filter, he talked about some imaging seeping in but when I listened to it didn’t sound rough like the measurements show.

http://archimago.blogspot.com/2019/08/measurements-dragonflies-audioquest.html?m=1

 

[Erick Muller]

People with NOS DACs don't complain because their sound is already degraded. They don't care about fidelity.

And if I don't care about fidelity and I like my music colored, is there an inherent problem with that?

 

[bigshot]

Just use a regular oversampling DAC. You don’t need selectable filters. At best they are transparent. At worst they color the sound. If you want to color the sound, do it with signal processing at the end of the chain, not with monkeying with numbers at the head.

 

[KinGensai]

Think about that for a minute, what is the bandwidth of that pink noise you’ve generated at 44.1kFs/s? …. So, as there is no signal above 22.05kHz, you are boosting nothing above 22.05kHz, which results in nothing above 22.05kHz! How can that be aliasing? The only way it could potentially be aliasing is if it’s a modelled EQ, if it has non-linear distortion or is emulating non-linear distortion and is ancient or programmed by an amateur (who failed to employ oversampling). Almost certainly what’s actually happening is clipping, either in the EQ itself, something else downstream or inter-sample peak clipping, although it could possibly be IMD in the amp or transducers.

Aliasing can ONLY occur when recording using an insufficiently high sample rate, so in an ADC or when resampling to an insufficiently high sample rate. You’ve made an assumption that it’s aliasing because it sounds the same/similar as clipping and then everything that follows is based on that assumption, a classic audiophile trap!

There is no aliasing in the DAC process, with slow or any other sort of anti-imaging filters. So as there is no aliasing then “yes”, the aliasing is definitely below audibility.

G

Oh, IMD makes a lot more sense given that context. So, given there's nothing occuring above 20k in the source noise, the only possible content that could occur above 20k would be generated harmonic distortion in the frequencies hit by that EQ and other various non-linear distortion artifacts caused by overloading the DAC? I'll try this again and take a close look at the waveform.

 

[KinGensai]

https://mega.nz/folder/9oABmIyZ#wkzZve6Edg2Af-nZDEmB2Q

To recreate what I was hearing, I tried generating pink noise in the same conditions I heard the artifacting from EQ and ran a comparative test in Deltawave. From what I saw in the original spectra, there was too much variation to pick out where possible IMD or aliasing(?) might have been happening, so I decided to run a more focused test and played a 16kHz test tone to modify with a high shelf without dropping pre-amp gain. This is directly from my V60 plugged in to my computer's ADC.

From what I'm seeing of the cleaner spectra due to using a single tone, I'm seeing a whole lot of artifacting that also extends in a linear fashion all the way up to the nyquist frequency, with some non-linear distortion being introduced at regular intervals going down the spectrum.

 

[gregorio]

Ok so slow filter inaudible aliasing artefacts
Nos audible aliasing artefacts?

No, one more time: Slow filter, fast filter, whatever filter, there are NO aliasing artefacts. Aliasing does not occur in the Digital to Analogue Conversion (DAC) process, only *potentially* in the Analogue to Digital Conversion (ADC) process. Maybe you’re referring to a NOS (Non-OverSampling) ADC but there is no such thing, by the end of the 1980’s all pro ADCs were oversampling ADCs and the NOS ADCs before then were never filterless anyway. The first pro ADC I bought was in about 1993 and as was standard at that time it had 64x oversampling.

I didn’t say it removed them I assumed it would be too low in level to be a problem with all this talk of -40db being enough.

-40dB is not enough. It’s enough by a few kiloHertz above the Nyquist freq (say by 24kHz or so, in the case of 44.1kFs/S) but you need increasingly more attenuation above that point, up to at least -80dB.

I have a dragonfly cobalt floating around somewhere and remember seeing archimagos measurements and that uses a slow minimum phase filter, he talked about some imaging seeping in but when I listened to it didn’t sound rough like the measurements show.

Unfortunately, you don’t seem to understand what the measurements show. To be fair, this is extremely common in the audiophile community and routinely exploited by marketers. If most audiophiles did understand what “the measurements show” then a large portion of the marketing/reviewing BS would instantly stop! There are two points that are commonly missed/misunderstood:

1. As I mentioned previously, these DAC measurements are measurements of how the DAC responds to certain test signals. Many of these test signals are designed to test the absolute limits of the DAC, they are not representative of the performance of the DAC in normal use. Some of the test signals used don’t even represent a “worst case scenario”, they’re even worse than a “worst case scenario”. The most obvious example of this is the impulse response measurement, the test signal is a Dirac Pulse but Dirac Pulses never exist in music (or other sound), so you will NEVER get that measurement when reproducing music/sound. In fact, the ringing you see in the impulse response measurement actually occurs quite rarely, the impulses we actually get in music typically produce no ringing at all and when we do get ringing, it’s always at a much lower level than with a Dirac Pulse. The test you’re referring to is not quite the same as the impulse response measurement, in the sense that a Dirac Pulse cannot actually exist as sound, while the near full scale 19 & 20kHz test tones could exist. However, in practice they don’t, or at least I’ve never seen it and if it does actually exist in any music recordings it’s incredibly rare. There are 19 & 20kHz freqs in music recordings but they’ll be a lot lower than full scale and therefore the images will also be a lot lower in level than this measurement, probably by at least 20dB and very possibly 40 or more dB lower.

2. Not reading/understanding the scales on the axes and/or not understanding how they relate to human hearing. In this example, the x-axis is frequency and the images produced in response to the 19 & 20kHz test tones are at 24.1 and 25.1kHz respectively. Baring in mind that healthy adult human hearing extends to about 16kHz, why are you expecting to hear something that “sounds rough” that is above 24kHz? Assuming you’re an adult, it would be very surprising if you could even hear the test tones (at a reasonable listening level), let alone the images which are ~5kHz higher. The only way you could potentially hear them with music recordings is indirectly, in the form of inter-modulation distortion (IMD) products below 16kHz, although you need to have particularly poor amp or transducers (or with very inappropriate settings) to produce enough IMD for it to be audible.

And if I don't care about fidelity and I like my music colored, is there an inherent problem with that?

None at all. There would only be an inherent problem if you then described that NOS DAC as high fidelity (or even mediocre fidelity) or as high (or good) quality with some implied reference to fidelity, because that would be an incorrect/false assertion. Unfortunately, we quite commonly see that sort of false assertion, which is a problem.

G