• 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!

High Resolution Audio?

solderdude

Grand Contributor
Joined
Jul 21, 2018
Messages
15,891
Likes
35,912
Location
The Neitherlands
A Byte has been 8 bits (2 nibble) since well before 1980.
16 bits = 2 bytes ... it's digital ... it's logical.
There is a reason why the next format 'step' is 24 bits and 32 bits.

Philips most likely coined 14 bits because that's what they could actually build at that time.
Also ... 8 bits is packed in 14 bits on the CD surface itself EFM (Eight to Forteen Modulation).
 
Last edited:

GrimSurfer

Major Contributor
Joined
May 25, 2019
Messages
1,238
Likes
1,484

blueone

Major Contributor
Forum Donor
Joined
May 11, 2019
Messages
1,179
Likes
1,495
Location
USA
Digital electronics were using 8bit words from the beginning.

Not entirely correct... for example, the CDC Cyber 70/170 supercomputers used 12bit bytes in 60bit words. Six bit characters, and 18bit memory addresses. I remember that their assembly language, Compass, was especially fun. (I had a weird notion of what was fun in those days.) :)

But seriously, all I was trying to say was that in the late 1970s I sincerely doubt the engineers had anything in mind but ASICs and replacing tape and phonograph records with a media that was "perfect sound forever".
 
Last edited:

eliash

Senior Member
Joined
May 29, 2019
Messages
407
Likes
209
Location
Bavaria, near lake Ammersee
Guys, I have to share that with you as a hidden Vinyl fan.

The discussion about the thresholld levels was also extremly helpful for me in understanding different media format characteristics.
I always thought of Vinyl as a low resolution media with some 60dB of SNR, but somehow I could never let it go, because it somehow sounded good and sometimes better than digital media.
Now I have finally understood why a 60dB SNR number is not that relevant.
Based on the discussions about the human hearing split in 24 "Critical Bands" of 100Hz to 3.5KHz bandwidth with the important ones having about 100Hz to 1000Hz bandwidth, the idea of resolution limitation becomes a completely different viewing (better say listening) angle. 100 to 1000Hz bands individually perceived, reduce a broadband noise number of 60dB by 13 to 23dB. Especially in the 300-2000Hz range the Vinyl idle noise seems to be lowest anyway, so there is a great impact of improvement just by the numbers.
Having understood this, there is double fun listening to the newly released "Nils Petter Molvaer - Khmer" album on Vinyl, which I always liked, but it had that "disco-type" sound on CD, due to heavy digital clippling and compression.
Now the Vinyl release is a completely different story. I would say, it now sounds "artistic" and the listening experience is much improved for my ears.
CD to Vinyl in numbers (a/b side):
- CD True Peak (i.e. extrapolating digital clipping): +0.6/0.7dB over full scale!
- Vinyl True Peak (dBs over 8cm/s reference level at 1KHz): +4.7/5.2dB (which is no issue for the cart)
- CD Peak-to-Loudness: 14.6/14.9dB
- Vinyl Peak-to-Loudness: 17/17.2dB

Somehow it feels anachronistic to me, obtaining a better sound from the technically less performing media, but that seems to be the way things develop over time...
 

SIY

Grand Contributor
Technical Expert
Joined
Apr 6, 2018
Messages
10,386
Likes
24,752
Location
Alfred, NY
Preferable =/= better.

And when you take into account the work by Mike Uwins ("Analog Hearts, Digital Minds"), sound may not actually be what's driving the preference.
 

blueone

Major Contributor
Forum Donor
Joined
May 11, 2019
Messages
1,179
Likes
1,495
Location
USA
Philips most likely coined 14 bits because that's what they could actually build at that time.

I do suspect that when Phillips started their TDA1540 design it was significantly more cost-effective to build a 14 bit DAC than a 16 bit one. Since consumer electronics were the target markets I can easily see Phillips making what it thought was a sound business decision.
 

Costia

Member
Joined
Jun 8, 2019
Messages
37
Likes
21
What does "true peak" for analog mean?
AFAIK it's inter-sample peaks for digital.
 

eliash

Senior Member
Joined
May 29, 2019
Messages
407
Likes
209
Location
Bavaria, near lake Ammersee
I haven't measured it... and I don't think it is all that relevant to any conversation between two human beings.

Whatever measurement I captured as a 50-something year old person who did cold water diving in the Navy (pressure changes are very hard on the ear drums when one is congested or affected by a cold virus) as a young man and warm water surfing as an older man (white noise in the surf zone very hard on ears) would be meaningless to somebody else.

Human hearing is reliable but lacks accuracy and precision. It's response to frequency varies (which raises the question that @andreasmaaan asked, why 400 Hz? This is outside the most sensitive part of human hearing range). It's sensitivity changes throughout the day and is affected by temperature, humidity, mood, stress levels, etc.

All of this questions the relevance of the kind of listening test that you appear to be performing... Why does this matter to you and what "hard data" are you seeking to capture?

Ok, Ok, no issue, I was just interested how big an individual spread under similar conditions (and similar age) could be...and I had my stress tinnitus as well...
Why I am so interested in these listening tests has its root in a strange personal experience starting some 10-15y ago, music suddenly sounded distorted to me. That went hand in hand with the acquisition of these new Dynaudio speakers newly installed in our living room (I am currently still running them, but mechanically modified and moved to my attic listening area). There was no previous installation in that living room before and I did mostly listening in the car or by IEMs since we moved in our house some years before. So I was double disappointed, because I could not find the reason for that distortion and listening fun was´nt there.
After leaving my job, I had more time to look at the issue and I read a lot of scientific books about hearing, to understand what was going on. I learned a lot on top of that, which I could not properly finish for myself during university, but initially did not find any answer for my problem. That was when I was starting to experiment by my own and I can tell you that was painful and time consuming, but finally the answer was there:
- Early room reflections (from walls and ceiling)
- Some sort of Doppler-effect caused IMD of the high frequency (tweeter) band, through enclosure vibration (caused by moving masses of the bass speakers).
I could handle both and that´s the status, I like listening to music again, because it sounds good again.
Regarding the 400Hz tone, this is just a relict from the past old vinyl and tape age and I have somehow sticked to that when going digital.
 

eliash

Senior Member
Joined
May 29, 2019
Messages
407
Likes
209
Location
Bavaria, near lake Ammersee
What does "true peak" for analog mean?
AFAIK it's inter-sample peaks for digital.

True peak, as far as I have understood it, is when digital clipping is detected, an algorithm tries to restore the original waveform and displays a calculated value above 0dBfs. That is quite properly described on the danish tc-electronic pages by several white papers and conference reprints by their staff.
True peak for analog means that I am tapping my amp´s analog recording output for vinyl or digital sources, convert that to digital again (currently by a cheap tiny 24/96 ADC box, but a bit modified to insure proper operation and calibration) for the Clarity audio analyser. I was quite astonished how well these original clipping events survived the double conversion, that is why I was mentioning it. It gives you also the ability to check, how vinyl and digital source material compares (I described that just before for an example).
 

Costia

Member
Joined
Jun 8, 2019
Messages
37
Likes
21
True peak is not digital clipping, it's kinda the opposite.
It's when all the digital samples are below 0db - and thus are not clipped.
But when reconstructed, the analog signal does go above 0db, which DACs probably cant reproduce correctly.
It can also clip if any digital processing is done on the signal, such as upsampling.
The original PCM itself is not clipped.

But if you have an analog signal that is below 0db and then you sample it to digital PCM.
The sampled PCM+true peak shouldn't be clipped either, since the original analog signal was always below 0db.

So I don't understand what you are comparing with the vynil true peak measurement.
 

luxnulla

Member
Joined
May 8, 2019
Messages
7
Likes
7
Yeh, he's one of the good ones. His final comments on MQA didn't seem quite accurate though, i.e. he forgot to mention that the supersonic content is lossily compressed.
MQA had to be almost tortured by a bunch of acute observers to provide/confirm technical details on how the mqa chain works since neither the much boasted "whitepaper" nor interviewes with key people disclosed it. I guess at the time of publication Waldrep wasn't aware of that.
 

eliash

Senior Member
Joined
May 29, 2019
Messages
407
Likes
209
Location
Bavaria, near lake Ammersee
True peak is not digital clipping, it's kinda the opposite.
It's when all the digital samples are below 0db - and thus are not clipped.
But when reconstructed, the analog signal does go above 0db, which DACs probably cant reproduce correctly.
It can also clip if any digital processing is done on the signal, such as upsampling.
The original PCM itself is not clipped.

But if you have an analog signal that is below 0db and then you sample it to digital PCM.
The sampled PCM+true peak shouldn't be clipped either, since the original analog signal was always below 0db.

So I don't understand what you are comparing with the vynil true peak measurement.

I understood true peak calculation from digital clipping differently (especially from a music consumers perspectice as described) and it is often associated to real 0dB peaks as I found, as e.g. on CDs when Foobar´s DR-meter shows 0dB max. level also.
I also saw some digital program material which seem to have been clipped, but was e.g. mastered at -1dBfs, then the true peak calculation would be enaged as well, from what I can remember from the documention around it, but I cannot prove that right now.
For analog true peak measurment that true peak collapses just to a peak reading (I hope so, that´s what I ought it for originally). Actually it looks that the use-case of the Clarity meter was not intended for analog monitoring, since it has digital inputs only, but it was much less expensive than meters, which have an analog input. I added that analog input capability by myself by that adapted 20bugs ADC via S/PDIF for the interim, aiming at a decent and affordable universal AD/DA converter to be hopefully finally released next month...
 

Costia

Member
Joined
Jun 8, 2019
Messages
37
Likes
21
It's associated with 0db samples because it's not a problem otherwise.

Its just how sampling and reconstruction works in general.
The signal get sampled at constant intervals, and the sample doesn't always fall at the highest peaks of the signal.
It will still be reconstructed perfectly, including the peaks that go above the samples, as long as it's within the Nyquist freq. limits.

So if the max PCM value is at -6db and the true peak is 2db above that, the whole thing is still at -4db, so everything will work fine.
It's only becomes a problem when the PCM sample is already very close to 0db, then the true peak can take you above 0db and cause clipping somewhere down the line.

Example of a 0db 12khz signal sampled at 48000khz:
K6bynBX.png

1 is 12khz 0db sampled at 48000
2 is (1) upsampled to 384khz, to show how it would look like after reconstruction
3 is the same 12khz signal at 0db, but sampled with a delay (phase)
4 is (3) upsampled/reconstructed

They both reconstruct to 0db even though all the samples in the 48000 file with the phase are below 0db.

If the PCM samples of the second 48000 wav were at 0db, the reconstructed signal would go above 0db and clip.
It also shows that if you sampled analog signal didn't clip, the sampled PCM+true peak won't clip after reconstruction either.
 
Last edited:

eliash

Senior Member
Joined
May 29, 2019
Messages
407
Likes
209
Location
Bavaria, near lake Ammersee
Not sure if this makes sense in that context, to be honest and whether this is associated to the true peak definiton I am referring to.
I understood true peak measurement as an extension to the well known broadcasting type peak meter. The idea, fom my perspective, was to detect and analyse previously clipped program material, also on a peak meter. So they invented the true peak meter, which looks at consecutive clipped samples in digital material and does some math on it to calculate the previously clipped original signal peak. In contrast to this is the intersampling peak issue, which I do not overlook fully right now and which is not meant here.
 

edechamps

Addicted to Fun and Learning
Forum Donor
Joined
Nov 21, 2018
Messages
910
Likes
3,620
Location
London, United Kingdom
I think when you use the term "digital true peak" most people here (well, myself at least) will assume you mean detecting and quantifying intersample peaks. This is how that term is used in the top Google results for "digital true peak". It's not about trying to reconstruct already clipped signals, which is a completely different thing. You're free to use your own definitions, but then don't be surprised if no-one can understand you.
 

GrimSurfer

Major Contributor
Joined
May 25, 2019
Messages
1,238
Likes
1,484
I understood true peak measurement as an extension to the well known broadcasting type peak meter.

Are you thinking of a VU meter?
 

eliash

Senior Member
Joined
May 29, 2019
Messages
407
Likes
209
Location
Bavaria, near lake Ammersee
I think when you use the term "digital true peak" most people here (well, myself at least) will assume you mean detecting and quantifying intersample peaks. This is how that term is used in the top Google results for "digital true peak". It's not about trying to reconstruct already clipped signals, which is a completely different thing. You're free to use your own definitions, but then don't be surprised if no-one can understand you.

Ok, thanks,
I retreat and have another look at the definition of that true peak definition in the Clarity box tomorrow...
...meanwhile you may have a look here:
https://www.itu.int/dms_pubrec/itu-r/rec/bs/R-REC-BS.1770-4-201510-I!!PDF-E.pdf
which seems to be the or one of the origins of the definition and they are talking about intersampling...
 

eliash

Senior Member
Joined
May 29, 2019
Messages
407
Likes
209
Location
Bavaria, near lake Ammersee
Are you thinking of a VU meter?

The VU-meter is one type of meter to measure and quantify audio level, the peak meter is another one, which reacts faster, more suitable to avoid clipping in digital systems, true peak metering is an evolution of the peak meter and here the exact definiton, I have to admit, is currently not completely clear to me...good night...
 

eliash

Senior Member
Joined
May 29, 2019
Messages
407
Likes
209
Location
Bavaria, near lake Ammersee
Ok, thanks,
I retreat and have another look at the definition of that true peak definition in the Clarity box tomorrow...
...meanwhile you may have a look here:
https://www.itu.int/dms_pubrec/itu-r/rec/bs/R-REC-BS.1770-4-201510-I!!PDF-E.pdf
which seems to be the or one of the origins of the definition and they are talking about intersampling...
One last reply for today, have a look at this from your recommended paper, which I had in mind in the above. The oversampling math seems to reconstruct the clipped original signal and results in a "true peak" reading...anyway tomorrow I read that again entirely and now I enjoy the rest of the "Tatort" TV movie...

1560108791804.jpeg
 

Blumlein 88

Grand Contributor
Forum Donor
Joined
Feb 23, 2016
Messages
20,524
Likes
37,057
I would have gone also to 1.5dB if the max. number of available 12 switch positions would have been higher...
I had the same issue. I used to put two in series. One position just connected the second one straight thru. So that gave me 23 positions.

I later snagged some milspec silver contact 36 position switches at a ham show. I never used them.

I ended up basically putting the resistors in a ladder configuration similar to a R2R ladder DAC. I only used 7. So I had 128 positions. I put toggle switches in and you could adjust with excellent precision with only 7 switches. It was surprising how quickly the settings became second nature though at first it seemed chaotic.
 
Top Bottom