Maybe stupid question about Digital PCM Filter and oversampling

[Vamp898]

There are tons of discussions about if people can hear the difference in DAC filters or not.

Unrelated to if these are able to be distinguished, just assume someone can for the sake of this question

Wouldn't oversampling to >=88.2kHz just irrelevant that? Why not just resample everything to >=88.2kHz and don't care?

Did i overlook something? I mean it doesn't really matter in the end anyway as most people will never hear the difference between the DAC filters but given they do, as soon you upsample, these effects should all be gone (not gone but way beyond 30kHz) or not?

Somehow i see no reason why upsampling should not solve this unless i overlooked something.

 

[colonelkernel8]

No. If you don't filter them the aliases could still appear in the audible range. But since you've moved the sampling frequency so far away from the target bandwidth you can use a simpler, lower order reconstruction filter since you have way more real estate.

 

[theveterans]

No. If you don't filter them the aliases could still appear in the audible range. But since you've moved the sampling frequency so far away from the target bandwidth you can use a simpler, lower order reconstruction filter since you have way more real estate.

Or allow for more aggressive noise shapers too if you desire

There are tons of discussions about if people can hear the difference in DAC filters or not.

Unrelated to if these are able to be distinguished, just assume someone can for the sake of this question

Wouldn't oversampling to >=88.2kHz just irrelevant that? Why not just resample everything to >=88.2kHz and don't care?

Did i overlook something? I mean it doesn't really matter in the end anyway as most people will never hear the difference between the DAC filters but given they do, as soon you upsample, these effects should all be gone (not gone but way beyond 30kHz) or not?

Somehow i see no reason why upsampling should not solve this unless i overlooked something.

Any modern DAC these days internally oversamples to at least 384KHz with built-in filters.

R2R DACs use the zero hold order step after after oversampling and filter reconstruction then relegates the bits on resistor ladder while DS DACs reduces the bits to about 5-7 bits (depending on DAC chip model) and oversamples to DSD frequency rates or even higher (on some DAC models) through a delta sigma modulator then the analog signal is filtered through a low pass filter

 

[Vamp898]

Or allow for more aggressive noise shapers too if you desire



Any modern DAC these days internally oversamples to at least 384KHz with built-in filters.

R2R DACs use the zero hold order step after after oversampling and filter reconstruction then relegates the bits on resistor ladder while DS DACs reduces the bits to about 5-7 bits (depending on DAC chip model) and oversamples to DSD frequency rates or even higher (on some DAC models) through a delta sigma modulator then the analog signal is filtered through a low pass filter

They do oversampling but that doesn't help with that issue. My D50III has the newest, modern, best ESS DAC that exists doing 8x oversampling (8x48 or 8x96) but it will still start to roll-off at 15kHz with an slow roll-off filter. If you resample to 96kHz before sending it to the DAC, it will start to roll-off at 30kHz.

The Fast Roll-off Filter will have an Stop band of 0.55 FS of the input signal (no matter if 0, 2x, 4x or 8x oversampling). As far as i understand, the only advantage of the oversampling is better image rejection, roll-off with slow roll-off will still happen in the audible band even with 8x oversampling as seen here

So i would have to resample the signal before it gets to the DAC, the oversampling inside the DAC doesn't help with this behavior.

Better measure than trust, this is the result of my D50III with an 48kHz input signal and the fast(!) roll-off filter with 8x oversampling

The role off happens shorty after 20kHz, the 8x oversampling has no effect on this, it stays at FS*0.55

 

[theveterans]

If music does NOT contain anything beyond 22KHz, you would still have no issues with FR using the slow roll off filter whether using hi-res file or not. However, if the hi-res file has content above 22 KHz, yes, the DAC will resolve it since the low pass filter will now start cutting off at half 30Khz like you mentioned. That's why some DACs have a sample rate indicator to let you know it's changing where to place the low pass filter.

There's a subjective debate of software oversampling to feed the DAC at 768KHz or 1.5 MHz for example which triggers its low pass filter to far ultrasonics allowing for complex noise shapers such as 5th order, 9th order and even 15th order noise shaper to function correctly and not create images/distortion

Other DACs omit this and just cut-off at 30Khz regardless of source file input (e.g. DACs that are DSD based such as Playback Designs or Nagra or EMM Labs)

 

[eq1849]

There are tons of discussions about if people can hear the difference in DAC filters or not.

Unrelated to if these are able to be distinguished, just assume someone can for the sake of this question

Wouldn't oversampling to >=88.2kHz just irrelevant that? Why not just resample everything to >=88.2kHz and don't care?

Did i overlook something? I mean it doesn't really matter in the end anyway as most people will never hear the difference between the DAC filters but given they do, as soon you upsample, these effects should all be gone (not gone but way beyond 30kHz) or not?

Somehow i see no reason why upsampling should not solve this unless i overlooked something.

The better question perhaps:

in the real world, is a DAC that oversamples 44.1khz to 88.2khz more or less accurate than simply feeding it a native 88.2khz file?

 

[theveterans]

The better question perhaps:

in the real world, is a DAC that oversamples 44.1khz to 88.2khz more or less accurate than simply feeding it a native 88.2khz file?

IMHO, there's literally no such thing as "accurate" that's because it doesn't exist (no such thing as infinite bandwidth). Heck, on some tracks there's a preference for a slower filters while others it's brickwall very high attenuation filters. I do like the built-in closed-form filter from Schiit with very high attenuation without using mega-taps like those from HQPlayer and Chord DACs

 

[colonelkernel8]

IMHO, there's literally no such thing as "accurate" that's because it doesn't exist (no such thing as infinite bandwidth). Heck, on some tracks there's a preference for a slower filters while others it's brickwall very high attenuation filters. I do like the built-in closed-form filter from Schiit with very high attenuation without using mega-taps like those from HQPlayer and Chord DACs

Huh? Of COURSE there's such a thing as "accurate". You don't need infinite bandwidth to accurately reproduce audio...so it stands to reason that you can achieve accuracy without infinite bandwidth.

A reconstruction filter is just a low pass filter guys. There's no magic there. It's fine if it rolls off at 20kHz, even starting at 15kHz is totally fine. There is no good musical information at 15kHz.

 

[danadam]

My D50III has the newest, modern, best ESS DAC that exists doing 8x oversampling (8x48 or 8x96) but it will still start to roll-off at 15kHz with an slow roll-off filter.

Then use fast roll-off.

If you resample to 96kHz before sending it to the DAC, it will start to roll-off at 30kHz.

You realize that in the process or resampling a filter is also used and it will be fast roll-off if the resampler is any good.

 

[Vamp898]

Then use fast roll-off.


You realize that in the process or resampling a filter is also used and it will be fast roll-off if the resampler is any good.

I use fast roll-off, the idea behind this question is/was, if i upsample to 88.2kHz, it should no longer matter which filter i use at all, as all filters will roll off after 20kHz so i am wondering why DAC maker but the effort in to provide several filters if they could just resample to FS*2 and be done.

Is there any downside to that?

Huh? Of COURSE there's such a thing as "accurate". You don't need infinite bandwidth to accurately reproduce audio...so it stands to reason that you can achieve accuracy without infinite bandwidth.

A reconstruction filter is just a low pass filter guys. There's no magic there. It's fine if it rolls off at 20kHz, even starting at 15kHz is totally fine. There is no good musical information at 15kHz.

What you mean is "It does not matter for humans who are listening to music" in which you're completely right. I think his point was in theory.

According to this function

you need unlimited bandwidth to reproduce the signal accurate. As this doesn't exist in the real world as it would never produce any sound, you can not accurately reproduce an digital signal.

Simplification: 10*1/3 = 3.33 --> 3.33/(1/3)--> 9.99

For human ears, 9.99 or 10 does not matter, it will be the same, but in math 9.99 and 10 are not the same. 9.99 = 10 is wrong. So he is right, its irrelevant in the real world but the statement is true, you need unlimited bandwidth

 

[bigshot]

I just use the built in filter and it is transparent. You don’t need any other filters.

 

[Vamp898]

I just use the built in filter and it is transparent. You don’t need any other filters.

ESS DACs have 8 filters built in^^ and the default isn't even fast roll-off, its an minimum phase filter with non-linear phase because, according to ESS, that is what people preferred in listening tests.

So this here is the default filter

And according to several blind tests, people can spot the differences with an 2/3 preference. So either Minimum Phase or Fast Roll-off is not transparent (i assume the first one which is the default).

Cirrus does the same thing with their CS43198 and CS43131, Minimum Phase is the default filter.

So if these blind tests really have been done correctly, the default filter for most modern DACs is not transparent, so there is a reason to care

 

[71 dB]

The linear phase default filter is transparent. Minimum phase filters cause phase distortion that might not be transparent to human ears. What people prefer is not the same thing as what is transparent. If you prefer phase distortion then you probably prefer minimum phase filter over linear phase filters.

 

[theveterans]

I use fast roll-off, the idea behind this question is/was, if i upsample to 88.2kHz, it should no longer matter which filter i use at all, as all filters will roll off after 20kHz so i am wondering why DAC maker but the effort in to provide several filters if they could just resample to FS*2 and be done.

Is there any downside to that?

I don't see any downsides TBH. Most likely and even subjectively the filters matter less after feeding it with hi-res. Look at all the NOS DAC impressions out there that when fed with 88.2 or 96 KHz, resulted to audible difference in FR linearity

The linear phase default filter is transparent. Minimum phase filters cause phase distortion that might not be transparent to human ears. What people prefer is not the same thing as what is transparent. If you prefer phase distortion then you probably prefer minimum phase filter over linear phase filters.

Yes. This would be audible to some indeed no matter the sample rate of source

 

[colonelkernel8]

I use fast roll-off, the idea behind this question is/was, if i upsample to 88.2kHz, it should no longer matter which filter i use at all, as all filters will roll off after 20kHz so i am wondering why DAC maker but the effort in to provide several filters if they could just resample to FS*2 and be done.

Is there any downside to that?


What you mean is "It does not matter for humans who are listening to music" in which you're completely right. I think his point was in theory.

According to this function

you need unlimited bandwidth to reproduce the signal accurate. As this doesn't exist in the real world as it would never produce any sound, you can not accurately reproduce an digital signal.

Simplification: 10*1/3 = 3.33 --> 3.33/(1/3)--> 9.99

For human ears, 9.99 or 10 does not matter, it will be the same, but in math 9.99 and 10 are not the same. 9.99 = 10 is wrong. So he is right, its irrelevant in the real world but the statement is true, you need unlimited bandwidth

We're talking about audio reproduction. "theveterans" applying an unachievable universal standard for signal accuracy is counterproductive and effectively pointless. Music is made by and for humans, the bandwidth limitation is implied and obvious.