"I can hear them church bells pre-ringing!" - A (mostly) time domain investigation into DAC-like reconstruction filters

[modus2000]

Now read up about Intermodulation Distortion in the audio band caused by the cr*p let through due to failing to use a proper reconstruction filter.

You measured result of system together with speaker? Or you also look only to electronic measurement?
Does NOS DAC has this distortion?

 

[MaxwellsEq]

You measured result of system together with speaker? Or you also look only to electronic measurement?
Does NOS DAC has this distortion?

Yes

 

[modus2000]

Yes

You right (may be, I'm not a big specialist), but from the point of view of approximation there is no any sense to make a shape of signal more rounded

 

[voodooless]

but from the point of view of approximation there is no any sense to make a shape of signal more rounded

The goal is to make the best possible approximation. Simply making it “more rounded” isn’t what the reconstruction filter does. Maybe take sampling theorem 101…

 

[antcollinet]

Now read up about Intermodulation Distortion in the audio band caused by the cr*p let through due to failing to use a proper reconstruction filter.

Not to mention - speakers don't always limit to 20kHz - they sometimes go significantly higher and in any case the characteristic of the rolloff is horrible. - so as well as IMD the "system" will let through a bunch of imaging crap.

 

[RandomEar]

Because your ideas restricted only by electronic measurement of DAC and you don't care of the summary result of complete music system.
When you start evaluate behaviour of the whole system (but not only the DAC), you will understand that System already have mechanical filter which do approximation - speaker, which is mechanically restrict range 20-20Khz and which approximate square wave by it mass of membrane.
The Mass of speaker do all necessary approximation for amplified signal from NOS DAC

So, NOS DAC is not needed any filter (from the listening point of view), because all square shapes will be rounded by mechanical filter of speakers or headphones.
When you apply filter to DAC you add second filter to the system

But be fair - you never hear ultrasonic pre-ringing of delta-sigma DAC, because it will be also filtered by speaker mass

Yeah, I randomly picked the KEF R3 Meta, checked the specs and it says "38 Hz - 50 kHz (-6 dB)". That's FAR beyond 20 kHz. You should absolutely not rely on your speakers filtering ultrasonic noise close to the audible band. It's different for class D switching, which is typically around 300 - 600 kHz.

There's zero reason to use NOS (without prior upsampling and filtering in the source) and push that ultrasonic noise down your signal chain, where you do not know how the non-linearities of your components outside the audible band will handle it. Just use a proper band-limited reconstruction filter. It's literally free - in fact, most DACs with proper filtering are cheaper than the obscure and niche NOS DACs on the market

 

[modus2000]

Seems I was wrong and filter is not responsible for approximation. The upsampling is responsible for approximation.

 

[voodooless]

I always find this an excellent illustration.

 

[voodooless]

Seems I was wrong and filter is not responsible for approximation. The upsampling is responsible for approximation.

They are one and the same process. When oversampling, you’ll have to find the newly added sample values. You do that by band-limiting the existing samples using a brick-wall filter.

 

[RandomEar]

Seems I was wrong and filter is not responsible for approximation. The upsampling is responsible for approximation.

No. Just... read the OP. Of this thread. Which you commented on.

As pointed out by @voodooless, upsampling and filtering are one single step in actual DAC hardware. In software you can separate it, but upsampling without filtering is just a signal with mostly zeros - in case of 8x upsampling, 7 of 8 samples will be zero. You can't listen to that.

 

[modus2000]

View attachment 521625
I always find this an excellent illustration.

I do not see anything wrong with NOS - each square point is just "coordinate" for the fly of speaker membrane.
if you ever try to measure square wave or sine wave on the speaker, you will see there is no stairs or pre-ringing

I'm not sure but this rounding may round useful sharp signal when it non needed

So, I prefer to hear originally recorderd "stairs"

 

[voodooless]

So, I prefer to hear originally recorderd "stairs"

They were never recorded like that

 

[RandomEar]

I do not see anything wrong with NOS - each square point is just "coordinate" for the fly of speaker membrane.
if you ever try to measure square wave or sine wave on the speaker, you will see there is no stairs or pre-ringing

I'm not sure but this rounding may round useful sharp signal when it non needed

There is no rounding. It's a reconstruction of the ground truth. Again: Read the OP of the thread you're commenting on. It's explained there with nice graphs.

So, I prefer to hear originally recorderd "stairs"

There are no "stairs" in the signal. Samples are points, the values between them are not defined. A NOS DAC makes stairs out of them by incorrectly applying a "sample and hold" strategy.

 

[modus2000]

They were never recorded like that

Why? You can see it in any DAW when you record music.
Or you can record square wave or sine wave from another DAC

 

[voodooless]

Why?

Because you still did not pass sampling theorem 101!

 

[antcollinet]

Why? You can see it in any DAW when you record music.
Or you can record square wave or sine wave from another DAC

Here is 101. Can you watch, and try to understand. If you don't understand anything, feel free to come back with questions.

 

[modus2000]

Because you still did not pass sampling theorem 101!

Just read 101, you are near to persuade me to refuse listening my NOS DAC )))
I thought that samples is just voltage level, which after amplification move the driver of speaker.

If modern DAC calculate sin on the basis of deep mathematic, I start to respect it ))

Sorry, I never interested DAC so deep and thank you for patient explanation

 

[voodooless]

I thought that samples is just voltage level, which after amplification move the driver of speaker.

A sample value is valid only for the point-in-time it was measured at. In theory this timeframe is infinitely small. That is why the proper way to display these samples is just a line for every discrete time interval with a dot noting the value:

Anything in between is basically redundant information, and can therefore be reconstructed perfectly. At least in theory, in practice we can get super close, more than close enough, in fact.

Proper brick-wall filters are the magic that makes that happen. They are present on both ADC and DAC ends.

 

[modus2000]

interesteng, thank you
Previously I thought that modern dac just round the cornenrs ))

 

[voodooless]

interesteng, thank you
Previously I thought that modern dac just round the cornenrs ))

This really has nothing to do with modern DACs. These principles have been applied for decades, basically since the start of the technology.