• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Delta-sigma vs “Multibit”: what’s the big deal?

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,834
Likes
16,496
Location
Monument, CO
A number of RF systems use the images to produce direct-RF outputs. The main drawbacks are reduced output due t the sinx/x curve and subsequently greater noise. They don't technically "violate" the sampling theorem, it's just another application of it.

For audio, to apply that sort of ultrasonic energy to the rest of my system seems like a bad thing... But my hearing rolls off well below 60 kHz so maybe I just don't know what I am missing. :)
 

AnalogSteph

Major Contributor
Joined
Nov 6, 2018
Messages
3,334
Likes
3,278
Location
.de
Speaking of chiptune, back in the mid-'90s you knew you had a fancy MOD player if it supported cubic spline interpolation in addition to 2nd-order and basic linear interpolation. It did make quite the difference in CPU usage, though I can't remember the specifics for the life of me.

A number of RF systems use the images to produce direct-RF outputs. The main drawbacks are reduced output due t the sinx/x curve and subsequently greater noise. They don't technically "violate" the sampling theorem, it's just another application of it.
It's called, guess what, a sampling mixer. You can do some neat things with these. A handful of essentially mechanically-tuned FM tuners used them to implement locking the LO to a fixed frequency grid of 100 or 50 kHz (made up by high multiples of fs = 100 / 50 kHz) instead of a traditional AFC. So you would get the frequency stability of a PLL-tuned set with the frontend selectivity of a mechanical varicap.

Stroboscopic speed indicators are another application of undersampling that may be a little closer to home for some. Here's the math if you're interested.
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,834
Likes
16,496
Location
Monument, CO
It's called, guess what, a sampling mixer. You can do some neat things with these. A handful of essentially mechanically-tuned FM tuners used them to implement locking the LO to a fixed frequency grid of 100 or 50 kHz (made up by high multiples of fs = 100 / 50 kHz) instead of a traditional AFC. So you would get the frequency stability of a PLL-tuned set with the frontend selectivity of a mechanical varicap.

I had not heard it called that but it's been a while. Another trick is to narrow the output pulses to be closer to an impulse than the typical ZOH stairsteps. You lose a lot of energy but the sinx/x rolloff is not such a big deal.

Technically, as long as the information bandwidth does not exceed Nyquist, you are not undersampling. It works for ADCs as well, with an appropriate (often gnarly) bandpass filter at the target RF center frequency and a sampling with sufficient bandwidth and low aperture error.
 

gvl

Major Contributor
Joined
Mar 16, 2018
Messages
3,425
Likes
3,980
Location
SoCal
I've been listening to multi-bit NOS DAC loan from my friend. A kitsune holo spring.

I have to say. I know this stuff measures worse, but I'm a sucker for NOS. I can listen for 7-8 hours without fatigue. Plus I don't hear the images/artifacting from not oversampling anyway. I really wanted to hate this thing..... :D it sounds great to me. The DX3 has more clarity but I get listening fatigue over time. The iFi black label had the same effortless fatigue-less sound but the holo spring is a much better DAC than the iFi.

I can definitely relate to this. I received my Khadas TB in December. I plugged it in, sat down, and flipped through a dozen tracks and it felt like "wow", I then disconnected the board and put it back in its box where it spent the next month sitting on a shelf. Only now I'm starting to take it in medicinal dozes with hopes my audio-sensory system will dull up a bit with use. Based on this experience I tend to think that a (really) well measuring DAC isn't necessarily a DAC for every occasion.
 

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
What kind of null levels does Paul's software reach ?
 

Esotechnik

Member
Joined
Mar 9, 2019
Messages
72
Likes
5
Location
Russia
TLDR - not going to read the whole 51 page thread.
First post:
Corr Depth: 9-35 dB - mean 2-6 bit real accuracy.

Idle tones, limit cycles, integrator leak - deltasigma-only artefacts.

What's the summary?
SAR ADC recording, R2R listening = linear time-invariant system.
>14-bit level accuracy and SFDR >90 dB is important for hi-fi audio and unattainable for deltasigma converters.
Good reproduction of 1 kHz test sine is a necessary but not sufficient condition for music.
 
Last edited:

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,521
Likes
37,050
First post:
"Corr Depth: 35,3 dB" - mean 6 bit accuracy.


SAR ADC recording, R2R listening = linear time-invariant system.
>14-bit level accuracy and SFDR >90 dB is important and unattainable for deltasigma converters.
You didn't get that from the posts over at gearslutz exactly.

I've posted results of I think 3 DACs which are sigma-delta and get better than 35.3 on the correlated null depth. The RMS difference is a better indicator anyway. The results in that thread also aren't giving you SFDR.

There are in fact sigma-delta converters that meet better than 90 db SFDR. So what are you going on about?
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,521
Likes
37,050
What kind of null levels does Paul's software reach ?
You'll need to be more specific. Just as an example, I've used it with all the same gear, only changing cables to see what difference the analog cables make, and you'll get results in the high 90 db or low 100 db range.

I know Paul's Deltawave gave results almost the same as Diffmaker in earlier versions. Within a couple db usually. I've not gone back with the current version to see how much that has changed. I think it probably does a better job of alignment and will get a bit better null than at the beginning.
 

Esotechnik

Member
Joined
Mar 9, 2019
Messages
72
Likes
5
Location
Russia
There are in fact sigma-delta converters that meet better than 90 db SFDR.
R2R DAC to DS ADC:
MSB Platinum IV Plus DAC ---> MSB Studio ADC master (DUJS)
4,9 dB (L), 4,9 dB (R) Corr Depth: 62,8 dB (L), 63,4 dB (R) Difference: -77,7 dBFS (L), -77,4 dBFS (R)

DS to DS:
Swissonic DA24 MkII ---> Texas Instrument PCM4222 Evaluation Module master (living sounds), DC filter off.
0,1 dB (L), 0,1 dB (R) Corr Depth: 82,6 dB (L), 85,5 dB (R) Difference*: -73.0 dBFS (L), -73.8 dBFS (R)
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,521
Likes
37,050

watchnerd

Grand Contributor
Joined
Dec 8, 2016
Messages
12,449
Likes
10,408
Location
Seattle Area, USA
First post:
>14-bit level accuracy and SFDR >90 dB is important for hi-fi audio and unattainable for deltasigma converters.
Good reproduction of 1 kHz test sine is a necessary but not sufficient condition for music.

There have a bajillion tests of DS DACs on this site that test for bit level linearity.

While few exceed 20 bits of real resolution, the few that test at <14 bits are notably few and usually called out as giving bad results.

Your hypothesis that DS is inherently limited to <14 bits of resolution is an extraordinary claim and requires extraordinary evidence to contradict prior fact discovery.
 

Esotechnik

Member
Joined
Mar 9, 2019
Messages
72
Likes
5
Location
Russia
There have a bajillion tests of DS DACs on this site that test for bit level linearity.
This is tests with strong averaging - up to 1M point FFT, good for 1kHz sine.
Music is close to a random noise signal. Converters should not spoil its statistical characteristics.
If the condition is met
U(t)in=a * U(t)out + b
(a&b = const) - this is linear system

Conversion time must be small and fixed.
Auto-correlation between samples should be minimal.

Multi-stage noise shaping = "all missing codes", non-linear.
SAR ADC and human hearing can detect this errors:
index.php
 
Last edited:

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,628
Likes
10,202
Location
North-East
What kind of null levels does Paul's software reach ?

Usually about the same, as @Blumlein 88 said. In the newer versions, you can also choose to correct for non-linear phase and amplitude errors. With these settings, the nulls can be significantly better with DeltaWave, sometimes by 10-20dB.
 

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,628
Likes
10,202
Location
North-East
Usually about the same, as @Blumlein 88 said. In the newer versions, you can also choose to correct for non-linear phase and amplitude errors. With these settings, the nulls can be significantly better with DeltaWave, sometimes by 10-20dB.

Here are the two results from Gearslutz quoted above, sent through DeltaWave:

R2R DAC to DS ADC:
MSB Platinum IV Plus DAC ---> MSB Studio ADC master (DUJS)
4,9 dB (L), 4,9 dB (R) Corr Depth: 62,8 dB (L), 63,4 dB (R) Difference: -77,7 dBFS (L), -77,4 dBFS (R)
DeltaWave: Difference (rms) = -80.02dB [-85.24dBA] Correlated Null Depth=84.53dB

DS to DS:
Swissonic DA24 MkII ---> Texas Instrument PCM4222 Evaluation Module master (living sounds), DC filter off.
0,1 dB (L), 0,1 dB (R) Corr Depth: 82,6 dB (L), 85,5 dB (R) Difference*: -73.0 dBFS (L), -73.8 dBFS (R)

DeltaWave: Difference (rms) = -86.62dB [-87.52dBA] Correlated Null Depth=87.94dB

The major improvement is in the correlated null, and that's because DW corrected for non-linear phase effects, usually caused by the type of filter used in the DAC.
 

DonH56

Master Contributor
Technical Expert
Forum Donor
Joined
Mar 15, 2016
Messages
7,834
Likes
16,496
Location
Monument, CO
SAR ADC recording, R2R listening = linear time-invariant system.
>14-bit level accuracy and SFDR >90 dB is important for hi-fi audio and unattainable for deltasigma converters.
Good reproduction of 1 kHz test sine is a necessary but not sufficient condition for music.

None of that is true but given the other posts and other folk already on this thread I feel no need to attempt to contribute nor correct the base assumptions. Seems like another "religious war" in the making.
 
Top Bottom