Consensus about upsampling to 512 DSD

[audiventory]

4 hours ago, jabbr said:

I’m coming to the impression that once upsampled to DSD512, that SQ entirely depends on the analog electronics ...

 

I'd say that analog parts are important for any DSD sample rate.
Digital filtering (in addition to analog one) may be applied though.

[audiventory]

18 minutes ago, Em2016 said:

80dB of digital volume control can maintain precision in the digital domain yes, but not affect the analogue signal to noise ratio?

 

Correct digital volume control don't impact to signal/noise ratio for PCM and DSD.

[audiventory]

9 minutes ago, Em2016 said:

PCM too?

Yes.

 

9 minutes ago, Em2016 said:

No impact on the analogue signal/ratio?

Analog (analog circuits) signal/noise ratio is higher than PCM24 bit.

 

If PCM 16 bit, signal/noise ratio is unchanged too.

 

If no changes in digital domain, there are no changes in analog domain too.

 

Digital volume control should be done properly. See below.

 

9 minutes ago, Em2016 said:

80dB of digital volume control?

From 0 to almost minus infinity. Processing should be done in float point formats and rounded correctly at input and output.

[audiventory]

 

6 minutes ago, Em2016 said:

Better than 144dB SNR?

 

The good audio DACs have noise floor about -118...120 dB. May be there are some exceptions, but I don't know such examples.

 

The digital signal is not "stairs". When digital processing is applied, need to consider it as analog signal with infinite precision by level.

[audiventory]

6 minutes ago, Em2016 said:

I'm definitely no expert myself but volume levelling down 80dB with no impact on the analogue SNR seems too good to be true (whether PCM or DSD).

 

In digital signal theory many things, that may looks like unobvious.

 

Digital signal is processed like analog, by same formulas.

 

I.e. it is not matter, where the same formula is applied in digital or analog form.

[audiventory]

2 minutes ago, Summit said:

It must be really hard not to be able to trust your own perceptions and always feel the need to get them verified before you can decided what sounds good or not. The same with food, drinks, love or anything that involves our senses and preference, and which we maybe can have some form of bias to. I would probably go insane if I could not trust my senses and constantly doubt if they are real or just imaging. I guess am lucky that my ignorance of so called “prof” can be such a blessing.

 

If gain control is applied correctly, it absolutelly transparent for signal and, further, ears.

 

Example: If the gain control is applied with 64-bit float point precision, measurement tool with 64-bit float point precision don't detect the impact. To detect the impact, 3...10 times more precise tool is need. The precision far from -144 dB.

[audiventory]

1 hour ago, Em2016 said:

 

If we're talking about a 32-bit DAC, 64-bits precision software volume control is not the limiting factor here, right?

 

Neither 16 bit nor 24 nor 32 bit DACs are not limiting factors in digital gain control.

[audiventory]

2 minutes ago, Kal Rubinson said:

JRiver cannot directly upsample DSD-to-DSDx without an intermediary conversion to PCM

There is a stuff where it described?

[audiventory]

23 minutes ago, Kal Rubinson said:

Nope.  Things like that are gleaned from the JRiver YABB forums.    And, btw, my full statement was 

"Unless something has been changed lately, JRiver cannot directly upsample DSD-to-DSDx without an intermediary conversion to PCM. "

 

There are no known me technology (patents, publications) to alter DSD without PCM intermediate conversion.

But PCM should be considered in wider meaning than 24 bit / 3xx kHz.

 

So it is general matter, not only certain software.

Read details https://samplerateconverter.com/educational/dsd-dsf-dff-editor

[audiventory]

47 minutes ago, barrows said:

incoming DSD (say DSD 256 so 11.2896/1 bit)-convert to 11.2896/32 bit for volume control-convert to 11.2896/6 bits-convert to analog.

 

47 minutes ago, barrows said:

Now some would say this a is a conversion to PCM, but without decimation, so none of the usual artifact problems.

 

Artifacts are result of digital filtering, that allow to decimate.

 

Decimated samples consume lesser computing resources.

 

But non-linear processings without the digital filtering can cause audible  noise.

 

Gain altering is linear processing. But non-filetered input DSD's noise can reduce dynamic range of output sigma-delta modulator (convert back to DSD).

To avoid the dynamic range reducing, it is need to use the digital filter, but it cause ringing.

 

There is not ideal decision.

 

In my opinion, ringing is not big matter, because most musical stuff is smooth enough and I know nothing about serious researches of ringing impact to ears.

But broken stability of the sigma-delta modulator at high musical levels due overload is more probable and obviously audible after the dynamic range reducing.

 

 

47 minutes ago, barrows said:

Certainly with the power available in computers they could control DSD volume this way, and my understanding is that the only limitation would be the performance of the DS modulator, right?

 

If there is sigma-delta modulator without filter, then the modulator consume all resources. But filter of DSD noise can consume more resources than the modulator.

[audiventory]

8 minutes ago, mansr said:

Audiventory applies a very steep lowpass filter near 20 kHz (I don't recall the exact number). Those with a fear of filters should probably avoid it. 

As I said before:

• either fear of filters;
• or lower maximal signal/noise ratio || more probable audible intermodulations || more probable broken stability.

In my opinion, last 2 things are more practically available for perception, than ringing.

 

If we allow to lose N dB, we can allow non-filtered DSD processing to avoid of broken stability.

 

But no one of people who was tried wide band 100 kHz, was stumbled with audible intermodulation (additional noise).

 

When 100 kHz band is applied there is possibility to makes lesser steep filter.

But if we want to maximally keep all information below 100 kHz, there is need steep filter again. Steep filter need to avoid more degradation of dynamic range due DSD noise.

[audiventory]

9 hours ago, jabbr said:

Could you confirm under which circumstances the 20kHz filter or 100kHz filter are applied.

 

Suppose I have an SACD ISO and which to extract/convert to DSF, is there a filter applied?

 

Settings

 

The filter is selected manually.

 

a) In Settings > General > Filter mode list is present:

1. Optimized (cut ultrasound above 20 kHz)

2. Non-optimized (traditional for PCM || DSF, DFF band is about 20...27 kHz depend on input/output sample rate)

3. Non-optimized Wide (traditional for PCM || ISO, DSF, DFF band is about 20...100 kHz depend on input/output sample rate)

 

In optimized mode you can select minimal phase filter (up to current version).

 

b) When input and output sample rates are similar (example: ISO to DSD64) bit-perfect mode is there.

 

To activate bit-perfect mode it is need in Settings > General > to check "Don't make DSP..." switch.

 

 

 

Applications

 

1. If there is audible noise at playback, Optimized mode is recommended (including minimal phase option by subjective preference).

 

2. If there is no noise, any other mode may be checked for the best available subjective or measured sound quality (including minimal phase option)

Read details: https://samplerateconverter.com/content/how-improve-sound-quality

 

3. If there are no clicks and audible noise at playback of DSF files, extracted from ISO, bit-perfect mode may be applied: https://samplerateconverter.com/iso-converter/convert-iso-dsf-wav-flac-aiff#bit-perfect-iso-conversion

 

4. If there are clicks at playback of DSF files, extracted from ISO, one of click suppressing technologies with DSP is used.
Read details https://samplerateconverter.com/iso-converter/convert-iso-dsf-wav-flac-aiff

Warning: some clicks may be fixed via manual editing of audio stuff only:

https://samplerateconverter.com/educational/dsd-dsf-dff-editor

 

5. If there are audible noise at playback of DSF files, extracted from ISO, Optimized mode is recommended.

 

 

Resume

 

In general case, I recommend to use Optimized mode with linear filter (default settings).

But also I recommend to check sound in all other modes for each used audio system.

[audiventory]

18 hours ago, Miska said:

And for sure there's no need for steep filter, there's just no point in using such with DSD unless you are decimating.

 

Me seems, I have described cases when steep filtering need besides decimation.

[audiventory]

1 hour ago, jabbr said:

My I suggest that your nomenclature and descriptions are confusing to us in English

 

It may be not only language issues are complicated, but technical issues too.

 

1 hour ago, jabbr said:

May I assume that no filter is being used? Or no steep 20 kHz lowpass filter?

.

Filter is not used when "Don't make DSP..." is checked.

But without resampling filter there will dramaticaly degraded sound.

 

I plan to make non-steep filter. Though I consider aliases as more harmful, than ringing for musical signal.

[audiventory]

2 hours ago, jabbr said:

Hmm... I've got SACD derived ISOs and I want to split into DSF. Levelling DC is attractive but if this "dramatically degrades the sound" I'd just live with the DC offsets. I'm upsampling in realtime using HQPlayer, so really really really I don't want a steep 20 kHz filter ... that gets around the whole purpose of upsampling to start with?

 

Does leveling DC offsets degrade sound from base DSD64? (without filter)

 

I meant resampling filter. If resampling performed without filtering, sound is degraded.

 

DSD editing (fixing DC offset issue as example) without filtering can lead to:

- increased ultrasound noise level (can  cause audible noise);

- broken stability of sigma delta re-modulator.

 

One way there is using lesser steep filter, but it can't solve both these issues.

 

Broken stability may be solved via reducing loudness on several dB.

 

Increased noise level can not be fixed, unfortunatelly. But, fortunatelly, it is not issue of each audio system.

[audiventory]

2 hours ago, semente said:

Is it possible to make two files of the same piano track snippet showing those differences (ringing vs. aliases) filtered within the audible range at say 16 or 18KHz?

That could prove educational.

 

I thinks about re-working of the article about ringing audio. May be there I'll describe the matter in details with pictures.

 

[audiventory]

7 hours ago, tailspn said:

why is the DC offset an issue with you?

Look at pictures from real ISO (part "1.2. DC bias in some tracks as click reason") https://samplerateconverter.com/iso-converter/convert-iso-dsf-wav-flac-aiff

 

I don't know now why the offset is appear.

 

There in first track DS bias is rise smoothly. And al borders of latest tracks we observe non-zero level. Non-zero level is click. Click loudness depend on bias.

If remove DS bias via calculating it, we don't guarantee of removing of non-zero level at begin and end simultaneously.

Because DC offset may be changed during track time line.

 

So we can only to make transient with special form to smooth level changing from zero to original sample level in begin of track and back at the track end.

 

Special form is recommended to maximize of smoothness, to better click suppressing.

[audiventory]

 

2 hours ago, jabbr said:

Joni Mitchell 24/192:



Next converted to DSD128 (Audiventory):

 

Optimized mode looks more "scary" at spectrum than "Non-optimized" and "Wide". But I use it as default, because it it give more chances to better sound by my experience and feedback. Without looking to spectrum

 

Main original rule here is:

 

"To improve something we can degrade something".

 

Me seems, it may be compared with glasses with blue filtering. As I understand, the filtering give more clear vision, because fix overload factor.

This idea is used in radio communications.

 

In sound application we lose nothing (by current theory and experience) for human. Because inaudible range is filtered and intermodulations a fixed, potential dynamic range is wider (we can increase level of useful signal after conversion for same noise floor).