• 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!

What are all the variables that impact DAC quality?

Killingbeans

Major Contributor
Joined
Oct 23, 2018
Messages
4,088
Likes
7,544
Location
Bjerringbro, Denmark.
With frequency, it's different. If you want to reproduce 20 000 Hz, you don't get any ''help'' from starting at say 19 000 Hz. It's the same thing as starting at 20 Hz, or dead silence for that matter.

Exactly. The "speed" of an electrical signal is defined by a bandwidth limitation and nothing more.
 

mhardy6647

Grand Contributor
Joined
Dec 12, 2019
Messages
11,211
Likes
24,170
Ok:



Analyze my little experiment as you like.



1. Audio Track. One full-scale bit. It doesn't get any "faster" - it is an "illegal" signal, though...
2. DAC output (RCA) - uses Focusrite ADC for capture
3. Amplifier output (differential) - sure does look like the DAC waveform - uses Focusrite ADC for capture
4. Martin Logan electrostat - using the amplifier - using UMIK-1 ADC - measured at listening position
5. JBL LSR 308 - using its own ADC and amplifier - then using UMIK-1 ADC - measured at listening position

Recorded traces amplified in REW as required to match amplitude.

Dots are individual sample values.

I notice the "ringing" at the DAC/Amp is not reflected in the speaker output.

Trace for JBL is inverted at REW for comparison.


View attachment 93759
This may be the coolest experiment/data I have ever seen on the internet -- and I am not being facetious.
 
OP
J
Joined
Oct 31, 2020
Messages
27
Likes
16
Ok:



Analyze my little experiment as you like.



1. Audio Track. One full-scale bit. It doesn't get any "faster" - it is an "illegal" signal, though...
2. DAC output (RCA) - uses Focusrite ADC for capture
3. Amplifier output (differential) - sure does look like the DAC waveform - uses Focusrite ADC for capture
4. Martin Logan electrostat - using the amplifier - using UMIK-1 ADC - measured at listening position
5. JBL LSR 308 - using its own ADC and amplifier - then using UMIK-1 ADC - measured at listening position

Recorded traces amplified in REW as required to match amplitude.

Dots are individual sample values.

I notice the "ringing" at the DAC/Amp is not reflected in the speaker output.

Trace for JBL is inverted at REW for comparison.


View attachment 93759

Could someone please explain to me how can it be possible for the DAC to already by 0.99860 be reacting to a signal which isn't supposed to arrive until after 0.99865? Can the DAC predict the future?
 

andreasmaaan

Master Contributor
Forum Donor
Joined
Jun 19, 2018
Messages
6,652
Likes
9,399
Could someone please explain to me how can it be possible for the DAC to already by 0.99860 be reacting to a signal which isn't supposed to arrive until after 0.99865? Can the DAC predict the future?

Kind of... The DAC takes some time to process the signal. The output (which includes the "pre-ringing" that you see there) is delayed a few ms relative to the input (but this is not shown in Ray's graphs). In reality, the pre-ringing does not begin until after the DAC has received the input signal.
 
OP
J
Joined
Oct 31, 2020
Messages
27
Likes
16
Kind of... The DAC takes some time to process the signal. The output (which includes the "pre-ringing" that you see there) is delayed a few ms relative to the input (but this is not shown in Ray's graphs). In reality, the pre-ringing does not begin until after the DAC has received the input signal.

Thanks for your answer, I understand now.
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,198
Likes
16,981
Location
Riverview FL
Could someone please explain to me how can it be possible for the DAC to already by 0.99860 be reacting to a signal which isn't supposed to arrive until after 0.99865? Can the DAC predict the future?

The process of playback and recording delays all of the traces except the source trace.

More or less like this:

Signal sent from Audacity.
Signal formatted for USB
Signal sent on USB from PC
Signal received on USB at Focusrite
Signal formatted for optical output on Focusrtie
Signal hits a switch
Signal goes to DEQ2496
Signal goes to miniDSP OpenDRC-DI
Signal goes to DAC
Analog goes to Focurite ADC
Signal re-formatted for USB
Signal sent to PC
Data turned into TCP/IP packets internally
Signal read by Audacity

Or...

Signal sent from Audacity.
Signal formatted for USB
Signal sent on USB from PC
Signal received on USB at Focusrite
Signal formatted for optical output on Focusrtie
Signal hits a switch
Signal goes to DEQ2496
Signal goes to miniDSP OpenDRC-DI
Signal goes to DAC
Analog goes to preamp
Analog goes to amplifier
Analog goes to speaker
Speaker is 10 feet - 10ms - from microphone
Microphone does ADC
Digits formatted for USB
USB sent to PC
Data turned into TCP/IP packets internally
Signal read by Audacity

I manually aligned the peaks and amplified the recorded tracks so they could be visually compared.

1605486736984.png

You can do that.
 
Last edited:

mhardy6647

Grand Contributor
Joined
Dec 12, 2019
Messages
11,211
Likes
24,170
Kind of... The DAC takes some time to process the signal. The output (which includes the "pre-ringing" that you see there) is delayed a few ms relative to the input (but this is not shown in Ray's graphs). In reality, the pre-ringing does not begin until after the DAC has received the input signal.
That's good -- 'cause it would majorly creep me out if it was some sort of quantum entanglement clairvoyance thing.
:cool:
 

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
Could someone please explain to me how can it be possible for the DAC to already by 0.99860 be reacting to a signal which isn't supposed to arrive until after 0.99865? Can the DAC predict the future?
It is because the "impulse" as represented by the digital sample is not what you (intuitively) think it is. You need to understand it in terms of band-limited signals. Here is a quick review of the Nyquist-Shannon sampling theorem. From Wikipedia:
If a function x(t) contains no frequencies higher than B hertz, it is completely determined by giving its ordinates at a series of points spaced 1/2B seconds apart.

In other words, for a continuous time signal band-limited to B Hz, if it is sampled at at least 2B Hz, the discrete digital samples (let's ignore quantization issues at this time) will allow perfect reconstruction of the original signal.

Before Whittaker/Shannon/Nyquist etc. mathematically proved the sampling theorem, it was not intuitive clear that if one takes periodic samples of a continuous time signal, one can obtain enough information to reconstruct the continuous time signal perfectly from the discrete samples. For many "audio enthusiasts", they still aren't clear.

The lesser known other side of theorem is that, if we remove the band-limiting restriction, these digital samples can represent an infinite number of signals. Here is an illustration from Signal and Systems by Oppenheim and Willsky.

Oppenheim.JPG


Let's take the often used example of a square wave. A perfect square wave is not band-limited since its harmonic contents goes to infinitely high frequencies. Here are digital samples (the blue points) that "looks like" a square wave (pulse). The orange curve is the only band-limited analog signal that, when sampled with an ADC, will give you these digital samples. The "perfect square wave" is just one of an infinite number of non-band-limited signals that will match these samples. So why does it has to be a squarish looking one instead of an infinite number of other possibilities?

fig1.png


Coming back to DAC's. If a DAC is not reproducing the orange curve from the samples, it is not reproducing the signal exactly as what the samples are supposed to represent, but instead gives a substitute with some arbitrary alterations. Remember, the "perfect square wave" is just one of an infinite number of possibilities those samples represent, when the band-limiting restriction is ignored.

When people talk about "pre-ringing", there is no pre-ringing, it is what the signal, when band-limited, is supposed to be. You can only obtain these digital samples if your original analog signal, when it is in conformance with the band-limiting requirement, looks exactly like the orange curve.
 

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
...
My illustration should make this easier to understand, if you can forgive my terrible handwriting skills.
Here is a better picture at your signal.
fig1b.png

You can see that there is a discontinuity in the slope when the lower frequency sine wave transition to a higher frequency one. The "abrupt" transition will produce higher frequency harmonics. Here is a look at what the band-limited signal that matches these samples should look like — there are some very minor oscillations.
fig1c.png

Here is what the built-in headphone output from my 7 year old Intel NUC produces, when captured with a scope (at 4 MHz sampling rate).

Waveform Screen.JPG
 

Beershaun

Major Contributor
Forum Donor
Joined
Oct 3, 2019
Messages
1,864
Likes
1,910
Thanks for starting this thread as a safe place for beginner questions about digital to analog conversion parameters. I think this question is on topic in a different dimension. Er. related to the idea of timing.

Is there any science regarding how accurate the timing of the above analog signal examples must be to be imperceptible to human hearing. Or how DACs address that? Ala Jitter and the idea of "time smearing" that Bob Stuart is talking about with MQA? https://audiogramii.wordpress.com/2018/03/31/time-smearing/

The idea being that humans are sensitive to proper timing of the signals down to 5-7ps and if the DAC cannot accurately recreate them at that level then the music can be distorted perceptibly in time between notes.

Disclaimer: I am not asking about the validity of MQA. I am asking about what the test is for timing accuracy and how good it must be to be imperceptible to our ears. Is it just Jitter?
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,198
Likes
16,981
Location
Riverview FL
Is there any science regarding how accurate the timing of the above analog signal examples must be to be imperceptible to human hearing.

Here is experimental evidence.

Using headphones and a "click" track.

One sample delay = 0.00002083 seconds

2 sample times delay on one channel at 48kHz starts to make a difference in the left/right location for me.

12 sample times or more makes it sound like only one side the headphone is working, though both sides are still outputting at the same level.

So, for me, today, 0.00002083 seconds (one sample time) borders on the imperceptible for location, 0.00004166 seconds delay is audible as a slight left/right location change (maybe perceived as 10 degrees off-center).

The same applies for speakers, though the numbers might be a little different, as both ears receive signal from both speakers.

1605497938767.png
 
Last edited:

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,521
Likes
37,050
Here is experimental evidence.

Using headphones and a "click" track.

One sample delay = 0.00002083 seconds

2 sample times delay on one channel at 48kHz starts to make a difference in the left/right location for me.

12 sample times or more makes it sound like only one side the headphone is working, though both sides are still outputting at the same level.

So, for me, today, 0.00002083 seconds (one sample time) borders on the imperceptible for location, 0.00004166 seconds delay is audible as a slight left/right location change (maybe perceived as 10 degrees off-center).

The same applies for speakers, though the numbers might be a little different, as both ears receive signal from both speakers.

View attachment 93840
So you might hear it clearly at 41.66 microseconds and might be near hearing it at 20.833 microseconds.

Some research into the matter indicates humans can detect an inter-channel timing change of 10 or 11 microseconds.

Timing of the digital signal as Ray knows isn't limited to a sample length of time. With 44,100 hz 16 bit it is well down below 100 picoseconds of inter-channel timing accuracy. So time smearing is generally a complete non-issue with any non-broken digital system.
 

blse59

Active Member
Joined
Feb 6, 2020
Messages
115
Likes
96
I believe the name of the chip is the biggest factor that affects quality. ESS Sabre <--- just by looking at the name you can tell it's sharp, sibilant, cold, clinical, and neutral.
 

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,383
Likes
24,749
Location
Alfred, NY
I believe the name of the chip is the biggest factor that affects quality. ESS Sabre <--- just by looking at the name you can tell it's sharp, sibilant, cold, clinical, and neutral.
Perhaps obsequious, purple, and clairvoyant?
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,521
Likes
37,050
I find AKM-based devices sound hot…

…cool would be better :confused:
A little liquid nitrogen around the chip will lower thermal noise, and chill down that sound just right. ;)
 
Top Bottom