Upsampling and Aliasing.

[PaperBoat]

So? What conclusion do you think you can derive from this?

No upsampling in JRiver. Linear phase EQ, Limiter with no oversampling. Updating IFI Zen V2 with GTO filter.

 

[danadam]

How is a stair-stepped ZOH signal from a DAC not the exact definition of an aliased signal?

And how is it one? What is your definition of aliased signal?

Aliasing is when a high frequency signal becomes indistinguishable from a low frequency signal (during AD process or during downsampling) and effectively becomes the low frequency signal. You know, it gets a false name, an alias. After that the original low frequency signal can't be recovered anymore.

And here, with ZOH, we have a low frequency signal being copied, or imaged, as high frequency signal and staying its own thing. And the original low frequency signal stays intact (modulo slight roll-off caused by ZOH).

 

[voodooless]

No upsampling in JRiver. Linear phase EQ, Limiter with no oversampling. Updating IFI Zen V2 with GTO filter.

Linear phase EQ is debatable. Especially correction for room modes you can do minimum phase. That’s what the room modes are anyway. Or use some software to linearize phase separate from frequency response.

 

[fieldcar]

And how is it one? What is your definition of aliased signal?

Aliasing is when a high frequency signal becomes indistinguishable from a low frequency signal (during AD process or during downsampling) and effectively becomes the low frequency signal. You know, it gets a false name, an alias. After that the original low frequency signal can't be recovered anymore.

And here, with ZOH, we have a low frequency signal being copied, or imaged, as high frequency signal and staying its own thing. And the original low frequency signal stays intact (modulo slight roll-off caused by ZOH).

You know what. I'll concede my defeat. That makes sense. I was under the impression that ZOH before a reconstruction filter would itself create signal nulls (aliasing) near the upper limit of the sample rate.

You all have been patient. No hard feelings. Happy to talk with you all again soon.

Bowing out.

 

[danadam]

In my poor understanding: 48kHz upsampled to 384kHz downsampled to 48kHz = distortion. Am I right? Please correct me.

Here's white noise, band-limited to 20 kHz (in attachment), upsampled and downsampled:

sox "noise-20k.flac" "up.flac" rate -v 352800
sox "up.flac" "down.flac" rate -v 44100

Let's see if up-sampling actually did anything:

Ok, looks nice, now let's subtract the down-sampled version from the original and see what's left:

sox -m -v1 "noise-20k.flac" -v -1 "down.flac" "null.flac"
sox "null.flac" -n stats
[...]
Pk lev dB    -138.47
RMS lev dB   -172.13
RMS Pk dB    -165.73
RMS Tr dB    -196.32

And the spectrum:

So not much in the null, meaning that for all practical considerations the two are the same and that up-sampling and down-sampling didn't introduce any distortion.

- But, but...
- Yes, the signal I used has no content in the transition band but that's on purpose. The whole point of 44.1k sampling being 44.1k and not 40k is to have the transition band where filters can do their thing. I'm just showing there are no changes in the band where it matters, i.e., up to 20 kHz.

One thing to pay attention though, especially when up-sampling white noise , is clipping, so just make sure the original has enough headroom.

 

[Blumlein 88]

https://sdg-master.com/lesestoff/aes97ny.pdf

View attachment 295259

Yep, works at dcs, should know better. Wrong terminology is still wrong. It is common, but wrong.

Plus in that paper he makes claims for high sample rates that are out of the book for selling to subjectivists. No good attribution for the claim. Trying to persuade being slippery with the truth.

 

[Hexspa]

I apologise for my weak understanding in this regard... But I can't resist myself to get a little bit more air (using Fabfilter Pro Q3)... tight bass (using Fabfilter Pro MB) and clip protection without loosing openness (using Fabfilter Pro L2) in my musics. That's all I want without introduce pre-ringing and distortion.

You’re introducing nonlinearity with that limiter and multiband compressor afaik. A little is ok but I am under the impression that you can introduce harmonic distortion with dynamics processors which can manifest itself as aliasing.

By “adding air” You’re also slightly reducing the headroom and pushing your signal toward clipping. If you’re pushing up the top on modern masters, compressing the bass and chopping peaks with a limiter, you’re adding distortion.

 

[PaperBoat]

You’re introducing nonlinearity with that limiter and multiband compressor afaik. A little is ok but I am under the impression that you can introduce harmonic distortion with dynamics processors which can manifest itself as aliasing.

By “adding air” You’re also slightly reducing the headroom and pushing your signal toward clipping. If you’re pushing up the top on modern masters, compressing the bass and chopping peaks with a limiter, you’re adding distortion.

Could be solved with linear EQing?

 

[Blockader]

Linear phase EQ is debatable. Especially correction for room modes you can do minimum phase. That’s what the room modes are anyway. Or use some software to linearize phase separate from frequency response.

not all room effects are min phase. Peaks are usually min phase and dips are not.

 

[voodooless]

not all room effects are min phase. Peaks are usually min phase and dips are not.

How are they not? Stuff can't go back in time...

 

[AnalogSteph]

Comb filter effects are very typically linear phase phenomena. And no, FIR filters don't require a time machine to operate, they have latency.

 

[Blockader]

Comb filter effects are very typically linear phase phenomena. And no, FIR filters don't require a time machine to operate, they have latency.

thanks for saving me from trouble of explaining it.

this is also one of the reasons why EQing high Q dips with min phase filters is not a good idea.

A great article about room correction & excess phase & mixed phase of room responses: https://www.roomeqwizard.com/help/help_en-GB/html/minimumphase.html

 

[PaperBoat]

Is it necessary to enable oversampling in Fabfilter Pro MB to mitigate pre-ringing in Linear Phase mode processing?

 

[voodooless]

Is it necessary to enable oversampling in Fabfilter Pro MB to mitigate pre-ringing in Linear Phase mode processing?

If pre-ringing is audible or not depends totally on the filters applied, it's not contingent on oversampling.

 

[Hexspa]

Could be solved with linear EQing?

Interesting association but no. Nonlinearity has to do with the waveform going in as one shape and coming out a different one via a transfer function. I think this video goes into it:

Because you're adding new frequencies to every frequency, you'd never be able to remove them. Aliasing is the same except the added frequencies are inharmonic so they're basically really crappy noise.