Where does quantization noise come from.

[Tell]

Here you have a good explanation how quantization noise works and sounds.
(The juicy part starts at around 4:35)

And I can really recommend to watch the whole series, the guy explains the digital audio really well

 

[Mulder]

Understanding quantization error as such is not very difficult I think, and many have explained it in this thread. But one question I had a hard time understanding once in a while, and as I think many others find it difficult to understand intuitively - is why the reconstructed signal at all becomes the same as the analog signal as in the original situation. If there is a quantization error now, why does this error give rise to "noise" - why does it not instead lead to the final curve getting a certain deviation compared to the original, ie the digital sound deviates from the original in a way corresponding to the quantization error. This interpretation of the error is what is sometimes claimed within HiFi-Folklore. I think the most intuitive conclusion is that the quantization error leads to a somewhat distorted curve, not that it leads to Noise. It was pretty tricky, I remember that I thought once in time to understand this, and I really just understand it just purely mathematically, ie I don't know what an educational non-mathematical explanation would look like. I may be wrong, but isn't this question the thread starter wondering?

 

[voodooless]

Here you have a good explanation how quantization noise works and sounds.
(The juicy part starts at around 4:35)

And I can really recommend to watch the whole series, the guy explains the digital audio really well

Except for the fact that this picture is dead wrong:

A telltale sign that something is wrong, is when you see staircases

 

[Tell]

Except for the fact that this picture is dead wrong:
View attachment 365418
A telltale sign that something is wrong, is when you see staircases

I know what you're getting at, but just watch the explanation and you'll understand

 

[antcollinet]

I know what you're getting at, but just watch the explanation and you'll understand

The video is understandable. The problem is the depiction using stair stepped waveforms is misleading, and simply perpetuates one of the myths of digital audio.

If he replaced the stair steps with lollipop representation it would be fine.

 

[Tell]

The video is understandable. The problem is the depiction using stair stepped waveforms is misleading, and simply perpetuates one of the myths of digital audio.

If he replaced the stair steps with lollipop representation it would be fine.

Yeah those lollipops are everywhere in his videos, it's just that this is a good way of explaining quanatization noise.

 

[voodooless]

If he replaced the stair steps with lollipop representation it would be fine.

And then also fix the resulting error graph

 

[torgeirs]

The question of the original signal and the quantization noise is interesting.
For me the original waveform + noise waveform is easiest to understand intuitivly.
A real world example is (simplified) when you talk to someone you know on the telephone. Then you recognize the voice. If the reception gets bad you still recognize the voice but you also hear the irritating quantization noise. So the brain seperates the original signal and the quantization noise.

 

[antcollinet]

And then also fix the resulting error graph

Of course.

it's just that this is a good way of explaining quanatization noise.

It can equally (and more accurately) be explained with the correct representation of time discrete samples.

 

[antcollinet]

but you also hear the irritating quantization noise.

If the line is just getting bad, it is not quantisation noise. It is some other form of noise.

But the example is still valid. Just as any noise signal (eg mains hum, or telephone line interference, or crosstalk etc etc) is simply added to the signal, so is the quantisation noise. It is just added in the digital domain, rather than the analogue domain.

 

[Chromatischism]

Yeah those lollipops are everywhere in his videos, it's just that this is a good way of explaining quanatization noise.

quantization noise

Let's stick with not using the term "noise" because it leads to confusion.

https://www.ni.com/docs/en-US/bundle/daqhelp/page/quanterror.html

 

[solderdude]

Well... sometimes we do enjoy seeing this kind of visual illusions even we know exactly what is the truth. Isn't it?

Nothing wrong with enjoying illusion(ists).
That, however, has nothing to do with technical performance nor with VFM.
'sound stage' and 'depth' however are just illusions the brain creates from 2 sound sources and basically are an illusion which also happens to be different for each person and is not related to quantization error nor noise either.

 

[torgeirs]

If the line is just getting bad, it is not quantisation noise. It is some other form of noise.

But the example is still valid. Just as any noise signal (eg mains hum, or telephone line interference, or crosstalk etc etc) is simply added to the signal, so is the quantisation noise. It is just added in the digital domain, rather than the analogue domain.

Often with cellular phone the bitrate is reduced with bad reseption and the quantisation noise goes up

 

[torgeirs]

Another test is super low bitrate music codec. Say 32kbit. You still recognise the singer and instruments. But the quantization noise can be anoying. (Of course a little bit more complicated because of the sub band coding)

Or plain truncating like this

https://dspillustrations.com/pages/posts/misc/how-does-quantization-noise-sound.html

The funny thing at that example is that at 2 bits you can kind of hear a shadow of the music in the error waveform.
So the noise is correlated with the signal

 

[antcollinet]

Let's stick with not using the term "noise" because it leads to confusion.

https://www.ni.com/docs/en-US/bundle/daqhelp/page/quanterror.html

No, it is noise in that it is an unwanted signal added to the wanted signal. When you hear it, it (if dithered) sounds similar to tape hiss, which is also a form of noise.

It *is* noise.

 

[antcollinet]

Often with cellular phone the bitrate is reduced with bad reseption and the quantisation noise goes up

Must admit I've never heard that - but I'm prepared to take your word for it.

What I normally hear on a bad line is pops and crackles. Though I guess this is mainly where there is a landline at one end of the link.

 

[DonH56]

Understanding quantization error as such is not very difficult I think, and many have explained it in this thread. But one question I had a hard time understanding once in a while, and as I think many others find it difficult to understand intuitively - is why the reconstructed signal at all becomes the same as the analog signal as in the original situation. If there is a quantization error now, why does this error give rise to "noise" - why does it not instead lead to the final curve getting a certain deviation compared to the original, ie the digital sound deviates from the original in a way corresponding to the quantization error. This interpretation of the error is what is sometimes claimed within HiFi-Folklore. I think the most intuitive conclusion is that the quantization error leads to a somewhat distorted curve, not that it leads to Noise. It was pretty tricky, I remember that I thought once in time to understand this, and I really just understand it just purely mathematically, ie I don't know what an educational non-mathematical explanation would look like. I may be wrong, but isn't this question the thread starter wondering?

The image (output) filter integrates, i.e. smooths, the waveform so it is a continuous curve. The stair-step waveform before the image (output) filter is due to quantization and the high-frequency content creating those sharp edges as the DAC switches state (values). Sharp edges require very wide bandwidth. much greater than audio (~20 kHz). The output filter removes frequencies above Nyquist, half the sampling rate, so the edges of the steps no longer appear at the output and all you see is a continuous waveform. The difference is also integrated so all those little pieces of signal ("triangles") between the smoothed output and the pre-filter stairstep create the quantization noise, the lower limit of the noise floor and thus maximum possible SNR from a DAC.

The math is a little painful (calculus used) and is shown here: https://www.audiosciencereview.com/...snr-about-6n-db-come-from.46603/#post-1663671

 

[Tell]

It can equally (and more accurately) be explained with the correct representation of time discrete samples.

Yeah guess you're correct, but apart from that it's a good explanation and the whole series is the best I've ever seen on digital audio (except for Monty of course, but it's not a comprehensive).

Let's stick with not using the term "noise" because it leads to confusion.

https://www.ni.com/docs/en-US/bundle/daqhelp/page/quanterror.html

Well it sounds like noise except that it's affected by the input signal, so correlated noise. It's quite an established term afaik so why not use it?

 

[DonH56]

All the engineering societies and references I know use "noise" for quantization noise. Whilst mathematically related to the signal and sampling rate, in general there is no common denominator between the fixed sampling rate and the signal, so the noise tones (spectra) created are uncorrelated and distinct from conventional distortion. For testing it is often useful to use a signal with specific ("prime") relation to the sample rate and FFT size so that all quantization noise spurs land in the middle of an FFT bin, obviating the need to window the result, but a PSD (power spectral density) analysis of the result will still yield essentially noise with equal energy in every FFT bin.

Refer to IEEE Standards 1057, 1241, 1658, etc. (yes I am biased, as I had a hand in writing some of those).

 

[torgeirs]

Must admit I've never heard that - but I'm prepared to take your word for it.

What I normally hear on a bad line is pops and crackles. Though I guess this is mainly where there is a landline at one end of the link.

Yes maybe a bad example in a international forum. Guess it can be difference from network to network. This is what I refer to: https://en.wikipedia.org/wiki/Link_adaptation