• 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!

Article: Understanding Digital Audio Measurements

scott wurcer

Major Contributor
Audio Luminary
Technical Expert
Joined
Apr 24, 2019
Messages
1,501
Likes
2,821
I was referring to this post:
https://www.audiosciencereview.com/...-digital-audio-measurements.10523/post-292443

In fact I am completely fine with white noise and impulse response measurements, and any signal that can be stored in a specific file format.

There is at times IMO some confusion between "illegal input" and simply the image rejection of the reconstruction filter.

If you generate white noise at say 8x oversampling and use the inverse FFT to remove everything above 22050Hz there is nothing in the region just above fs/2 in the data.
 

JR4321

Member
Joined
May 21, 2017
Messages
54
Likes
24
I read much, but not all of the article. I might have missed the answer, but my question is:

If you have a 24bit audio stream and are playing it on a DAC that is capable of say 19 bits of range, are you missing the quietest parts(they're cut off from the stream), or is it compressed such that there is not as many dBs between the loudest vs quietest? Thank you.
 

trl

Major Contributor
King of Mods
Joined
Feb 28, 2018
Messages
1,967
Likes
2,523
Location
Iasi, RO
The 24bit audio is the digital representation of the song, while the "19 bits of range" is the actual signal-over-noise (SINAD). SINAD is influenced a lot by the analog parts from the audio chain, because electronic components are adding noise and distortions over the original sound, but also the jitter and digital artefacts could affect the final SINAD as well.

I would personally don't try to compare the SINAD with the bitrate of the audio files.

For me, if the audio chain I'm listening to has a SINAD of at least 16-bits on the listening levels (e.g.: equivalent of an AVG 85dB for both headphones and speakers) then I would be very pleased.
 
Last edited:
OP
amirm

amirm

Founder/Admin
Staff Member
CFO (Chief Fun Officer)
Joined
Feb 13, 2016
Messages
44,368
Likes
234,388
Location
Seattle Area
I read much, but not all of the article. I might have missed the answer, but my question is:

If you have a 24bit audio stream and are playing it on a DAC that is capable of say 19 bits of range, are you missing the quietest parts(they're cut off from the stream), or is it compressed such that there is not as many dBs between the loudest vs quietest? Thank you.
You are. But the limit of capture and reproduction in audio is about 20 to 21 bits. It is very unlikely that you have that much useful information in recorded music as noise takes over the low order bits. So 19 bits is fine too.

And no, nothing is compressed. If SINAD is higher, then you are stepping on any music bits with either noise, distortion or both created by the DAC.
 

pozz

Слава Україні
Forum Donor
Editor
Joined
May 21, 2019
Messages
4,036
Likes
6,827
I read much, but not all of the article. I might have missed the answer, but my question is:

If you have a 24bit audio stream and are playing it on a DAC that is capable of say 19 bits of range, are you missing the quietest parts(they're cut off from the stream), or is it compressed such that there is not as many dBs between the loudest vs quietest? Thank you.
If you have a DAW or an editing program try playing a track at regular volume and then lowering it by 80dB (~13 bits, or well within CD range). You will likely hear nothing.

But to answer your question. 19 bits or ~114dB of range will has to be defined. Linearity, SINAD, SNR or DAC capability?
  • 114dB linearity means that signals below this level will not be produced with the correct volume. The values will be unstable and fluctuate randomly.
  • 114dB SINAD means that below this level either noise or distortion will begin to play along with the music.
  • 114dB SNR is the same as the above, but ignoring distortion (which is usually above the noise floor of the DAC).
  • 114dB of DAC capability means that the DAC will cut any signal below this level out entirely. I think one or two measured units had this problem.
 

JR4321

Member
Joined
May 21, 2017
Messages
54
Likes
24
If you have a DAW or an editing program try playing a track at regular volume and then lowering it by 80dB (~13 bits, or well within CD range). You will likely hear nothing.

But to answer your question. 19 bits or ~114dB of range will has to be defined. Linearity, SINAD, SNR or DAC capability?
  • 114dB linearity means that signals below this level will not be produced with the correct volume. The values will be unstable and fluctuate randomly.
  • 114dB SINAD means that below this level either noise or distortion will begin to play along with the music.
  • 114dB SNR is the same as the above, but ignoring distortion (which is usually above the noise floor of the DAC).
  • 114dB of DAC capability means that the DAC will cut any signal below this level out entirely. I think one or two measured units had this problem.
Thank you for the details. How about in the case of the latest review, here (Topping E30). What definition, above, would the 19.5bits fall under?
 

JR4321

Member
Joined
May 21, 2017
Messages
54
Likes
24
The 24bit audio is the digital representation of the song, while the "19 bits of range" is the actual signal-over-noise (SINAD). SINAD is influenced a lot by the analog parts from the audio chain, because electronic component are adding noise and distortions over the original sound, but also the jitter and digital artefacts could affect the final SINAD as well.

I would personally don't try to compare the SINAD with the bitrate of the audio files.

For me, if the audio chain I'm listening to has a SINAD of at least 16-bits on the listening levels (e.g.: equivalent of an AVG 85dB for both headphones and speakers) then I would be very pleased.
Thank you. How would you go about measuring the bits/dBs/dynamic range of the entire system? How would you know what you're missing?
 
  • Like
Reactions: trl

pozz

Слава Україні
Forum Donor
Editor
Joined
May 21, 2019
Messages
4,036
Likes
6,827
Thank you for the details. How about in the case of the latest review, here (Topping E30). What definition, above, would the 19.5bits fall under?
1 bit is approximately 6.02 dB. So 19.5 bits is 117dB. Keep in mind what Amir said about the absolute noise floor being 20-21bits for all electronics. It's an unbreakable limit at the moment, quantum computing excluded.
 

trl

Major Contributor
King of Mods
Joined
Feb 28, 2018
Messages
1,967
Likes
2,523
Location
Iasi, RO
Thank you. How would you go about measuring the bits/dBs/dynamic range of the entire system? How would you know what you're missing?
https://www.audiosciencereview.com/...om-measurement-tutorial-for-dummies-part-1.4/
https://www.audiosciencereview.com/...om-measurement-tutorial-for-dummies-part-2.5/
https://www.audiosciencereview.com/...om-measurements-plots-and-help-with-rew.8601/

The most important things to measure on your audio system:
  1. Get a song with a good dynamic and pump up the volume when it reaches that level of audio energy that pleases you most (you can optionally measure the output dB from the listening position, if you want);
  2. Listen to your room, see what resonates most, search for reverberations, then find a way to damp and absorb those unwanted sounds (use at least a minimum treatment for your room);
  3. Turn off the music, leave the volume knob at the same position from step 1, take a 5 min. break so your ears to accommodate with room silence, then from the listening position try to identify if mains hum or hiss noise can be audible coming from your speakers.
If on step 2 you don't hear your room "vibrating" and on step 3 you can't hear any noise or hum, I guess there is no subjective reason to make any upgrade.

Of course, REW & a calibration mic would help a lot here, so you can better check and understand if there's something wrong with your audio system or not.
 

wwenze

Major Contributor
Joined
May 22, 2018
Messages
1,284
Likes
1,827
"Multitone Testing
An incorrect criticism against audio measurements is that we usually use a single tone and hence the result can’t possibly apply to music, which has countless tones. The argument misses the case that if a DAC distorts one tone, it will just as well distort one thousand. Anyway, to deal with the criticism, we can run more tones."

A recent few DACs measured have shown not 100% correlation between SINAD and multitone so I guess there is worth.

One thing I noticed however is that the "grass" of the Okto used in this article is much lower than the harmonic distortion components. That's when I realized the 32 tones themselves are harmonics of each other so we won't see harmonic distortion.

I'm wondering should we offset the tones' frequencies so they no longer hide harmonic distortion, then we can see a graph that contains both.
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,200
Likes
16,982
Location
Riverview FL
That's when I realized the 32 tones themselves are harmonics of each other so we won't see harmonic distortion.

I believe the tone frequencies have logarithmic spacing.

1585205461099.png


For example, there appears to be a tone at 100Hz. There is no tone at 300Hz, the third harmonic.

Harmonics are integer multiples of a fundamental.

Some may overlap, but not all.
 

Schackmannen

Active Member
Forum Donor
Joined
Aug 6, 2018
Messages
167
Likes
225
Location
Tucson, Arizona
"Multitone Testing
An incorrect criticism against audio measurements is that we usually use a single tone and hence the result can’t possibly apply to music, which has countless tones. The argument misses the case that if a DAC distorts one tone, it will just as well distort one thousand. Anyway, to deal with the criticism, we can run more tones."

A recent few DACs measured have shown not 100% correlation between SINAD and multitone so I guess there is worth.

One thing I noticed however is that the "grass" of the Okto used in this article is much lower than the harmonic distortion components. That's when I realized the 32 tones themselves are harmonics of each other so we won't see harmonic distortion.

I'm wondering should we offset the tones' frequencies so they no longer hide harmonic distortion, then we can see a graph that contains both.
Amir has posted the multitone test file before and IIRC the tones are generated so that none of them are an integer multiple of the others so harmonic distortion still show up in the test. I think the reason why the distortion sometimes looks lower in the multitone test than in the dashboard is because each tone in the multitone test has to be a lower amplitude than the single tone in the dashboard, otherwise they would sum up to over 0 dBFS and clip, which means that the distortion created by each tone in the multitone test is a lot lower in amplitude.
 

RayDunzl

Grand Contributor
Central Scrutinizer
Joined
Mar 9, 2016
Messages
13,200
Likes
16,982
Location
Riverview FL
Last edited:

JimB

Addicted to Fun and Learning
Forum Donor
Joined
Sep 19, 2019
Messages
731
Likes
493
Location
California
Sorry if this has already been addressed somewhere on the site, but it struck me (red emphasis, mine):

AKM's two-chip DAC solution that separates the digital aspect from the analog was announced following a period of extensive tests promoted by the Japanese company using controlled listening tests. The results, as the company notes, validated the combination of the AK4191 premium delta-sigma modulator with multi-bit output chip for enhanced digital signal processing capability, when combined with with the AK4498 premium D/A converter with multi-bit modulator data interface, resulting in superior dynamic with vast playback resolution, as well as improved noise immunity and increased low-frequency noise performance.

AKM is now offering this solution for the audio industry, recommending the two-chip solution for high-end audio designs, active speakers, network audio and USB DAC applications. "By using two ICs, AKM has minimized the effects of digital noise within the analog output, resulting in a perceived improvement of the ratio of signal to noise. This improvement, while not easily quantifiable via traditional measurement techniques, is easily experienced during controlled listening tests," the company states.
 

pozz

Слава Україні
Forum Donor
Editor
Joined
May 21, 2019
Messages
4,036
Likes
6,827
Sorry if this has already been addressed somewhere on the site, but it struck me (red emphasis, mine):

AKM's two-chip DAC solution that separates the digital aspect from the analog was announced following a period of extensive tests promoted by the Japanese company using controlled listening tests. The results, as the company notes, validated the combination of the AK4191 premium delta-sigma modulator with multi-bit output chip for enhanced digital signal processing capability, when combined with with the AK4498 premium D/A converter with multi-bit modulator data interface, resulting in superior dynamic with vast playback resolution, as well as improved noise immunity and increased low-frequency noise performance.

AKM is now offering this solution for the audio industry, recommending the two-chip solution for high-end audio designs, active speakers, network audio and USB DAC applications. "By using two ICs, AKM has minimized the effects of digital noise within the analog output, resulting in a perceived improvement of the ratio of signal to noise. This improvement, while not easily quantifiable via traditional measurement techniques, is easily experienced during controlled listening tests," the company states.
Details from the company would be definitely interesting.
 

AntM

New Member
Joined
Apr 18, 2020
Messages
1
Likes
0
Hey all, just registered today and really interested in the discussion. Love the article and approach to measurement.

A few thoughts for debate on the more subtle aspects of digital audio and especially DAC measurement .... apologies if already covered.

- linearity is really revealing, not just because it may hint at the DAC architecture (sigma-delta, R-2R etc) but also whether the digital truncation stages are dithered correctly. So say if a 24b input is truncated to 16b without correct dither, a sine wave around the 16b point will start to look more like a square wave. Bad for sound quality!

This can happen with the output of digital volume controls, digital filters and intermediate stages of the DAC. You might hear this if you played a very very low level sine wave from a test track - it would sound more like a square wave.


- limit cycle tests: some DACs will generate tones if fed with a dc input or low level signal. This might be heard as a shuffling sound at really low levels.


- the max level square wave test to detect clipping is really important : modern tracks are often very “hot” i.e digitally compressed close to 0dB. I imagine some poorer DACs don’t have the digital headroom to cope - this means that any ringing in the digital filters (there is always some) will cause clipping.


- digital filter curves can be important. A slow roll off design will always cause some images above FS/2 that we’re not there in the original signal. This is not audible in itself, but might interfere with subsequent analogue stages and could excite the ultrasonic peaks in metal dome tweeters?


- there has been a lot of debate in the audio industry about digital filter design. The most popular Symmetrical (or at least non-minimum phase) FIR filters have pre-ringing, in other words the signal starts to appear before the main peak.

It has been argued that this is unnatural as there are no physical systems in the real world that do this, and so the human auditory system might not like it. Slow roll off filters tend to have less ringing, but have more transition-band imaging.... so there is a trade-off.

There are also some DACs with minimum phase filters, so no pre-ringing at all, but they tend to have more post-ringing. I’m other words there is more ringing after the main peak. This may cause problems with clipping if there is not enough digital headroom.


- Some DACs have built-in sample-rate converters, in other words they might always upsample to 48kHz, this could have consequences, especially if the conversion is asynchronous- all manor of evils can creep in. Wondering what tests could be the most revealing -

Apologies for the long post, cabin fever must setting in here! Happy to debate.
 
Last edited:

Denman

Member
Joined
Apr 27, 2020
Messages
37
Likes
50
Location
Amsterdam
@amirm
"Because the non-linearities in audio are quite small relative to the main signal (a 1 kHz tone in this case), we cannot see them in the sine wave in top right (time domain signal)."

Should be "top left"

I'm learning a lot here, thanks for this extensive guide!
 

ReaderZ

Addicted to Fun and Learning
Joined
Apr 14, 2020
Messages
618
Likes
414
"By using two ICs, AKM has minimized the effects of digital noise within the analog output, resulting in a perceived improvement of the ratio of signal to noise. This improvement, while not easily quantifiable via traditional measurement techniques, is easily experienced during controlled listening tests," the company states.

Isn't SNR easy to measure?
 
Top Bottom