• 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!

AAC's quality...

Robin L

Master Contributor
Joined
Sep 2, 2019
Messages
5,211
Likes
7,589
Location
1 mile east of Sleater Kinney Rd
...can you provide any scientific article with references that confirm this? I don't want any Monty from 'youtube'.
"Any Monty" from YouTube does a fine job, thank you.

Any representation of a soundwave represented as a series of ladder steps is a misrepresentation of how digital audio works. Nyquist works, our ears are physically limited to an upper limit of 20khz at best, higher bits rates are good for production and postproduction but basically don't do much for the final product. I'm not sure what higher sampling rates do, seeing as they are outside our range of hearing, but DBTs have shown that there are no audible differences between High-rez and Redbook.
 

ZolaIII

Major Contributor
Joined
Jul 28, 2019
Messages
4,069
Likes
2,409
@UKPI Opus (encoder) on X86 (AMD64) by default first does input conversion to FP32 (if there are unsolved IEEE MCU erratums build is fixed point and converts to 32 bit INT). It does the processing with increased precision and then converts down to 16 bit integer (or Q17) for packing. Not entirely sure but I think output can be also forced as 32 bit FP (--float option).
On X86 platform even with mentioned erratum FP build whose default. As discussed for the platform whose also disclosed for Opus lib.
On the other hand ARM had so much erratums that it used fixed point builds only.
I hope that whose informative enough for you. As I actually run away from IT to audio (and hire) I leave up to you to found up how things have gone regarding ARM and FP support and inform me, best regards.
 

Grooved

Addicted to Fun and Learning
Joined
Feb 26, 2021
Messages
679
Likes
441
"Any Monty" from YouTube does a fine job, thank you.

Any representation of a soundwave represented as a series of ladder steps is a misrepresentation of how digital audio works.
Correct if I'm wrong but I would say more that the ladder step is a representation of "how digital audio" works, but that it's misrepresentation of "how analog audio coming from digital audio" works
 

earlevel

Addicted to Fun and Learning
Joined
Nov 18, 2020
Messages
545
Likes
776
Correct if I'm wrong but I would say more that the ladder step is a representation of "how digital audio" works, but that it's misrepresentation of "how analog audio coming from digital audio" works
No, it's actually a misrepresentation of both.

Consider sampling—taking an instantaneous reading of an analog signal. That reading is truly only valid for the instant it represents—it does not represent the entire period till the next sample.

But, I have to add something that usually goes unsaid, and muddies the water a bit. Like I said before, to make the sampling process practical, we "sample and hold" the input—this is just so that the super-accurate measurements we take (24-bit sampling is insanely accurate) have time to settle to a stable reading. The reading is still of a point in time, we just lengthen the time to make it easier to read.

More significantly, outputting accurate impulses is hard to do with accuracy. So we do, essentially, output steps instead of impulses, ahead of filtering. But unlike the sampling process, this does truly f'up the result. But, the error it causes is known precisely and is easily compensated for.

In other words, for DACs, we essentially do it wrong then fix it—it's simply easier to do that than to to try to do it right in one shot. Extending the impulses to steps causes a sinc-shaped frequency rolloff at the top end. It's easy to compensate for with inverse filtering. I could elaborate on why it's so much easier and repeatable to make steps than impulses, if asked, but I'm already wordy.

Of course, modern converters are usually even more complex, but I'm describing the most fundamental aspects common to all, including the basic R-2R ladder DACs.

So, yeah, it gets a little sticky when complaining that "steps" are wrong, despite steps being used in practical implementations. But the fact is, if you're describing sampling theory, viewing as steps is wrong. Even if we might use them when it comes time to build a practical converter.
 
Last edited:

pkane

Master Contributor
Forum Donor
Joined
Aug 18, 2017
Messages
5,632
Likes
10,205
Location
North-East
Correct if I'm wrong but I would say more that the ladder step is a representation of "how digital audio" works, but that it's misrepresentation of "how analog audio coming from digital audio" works

Here's an example of digital samples and the properly (sinc) reconstructed waveform from these samples:

index.php


If you add a thousand more point samples in between each of the two samples, the final output waveform will not change.
 

NTK

Major Contributor
Forum Donor
Joined
Aug 11, 2019
Messages
2,656
Likes
5,819
Location
US East
Correct if I'm wrong but I would say more that the ladder step is a representation of "how digital audio" works, but that it's misrepresentation of "how analog audio coming from digital audio" works
Some DACs, as their first step in the D-to-A conversion, output stairstep waveforms (zero-order hold, or sample-and-hold). This results in a modulated frequency response in the reconstruction (sinc response), which needs to be reversed. All these are parts of the inner details of the DAC chip.

See this presentation from TI for more details.
 

Head_Unit

Major Contributor
Forum Donor
Joined
Aug 27, 2018
Messages
1,341
Likes
688
Yep... MTV. The old days of a thing called television.

I just miss all of the record stores that used to exist seemingly everywhere. Now I have to drive 30 minutes to find an okay one.
I just tap into my phone and listen to all kinds of stuff ;)
As for this "television" you mention, is that the thing they faked the moon landing on? Someone told me there used to not be YouTube but of course I didn't believe such nonsense...
 

2M2B

Active Member
Joined
Mar 9, 2021
Messages
118
Likes
73
AAC the best codec with near MP3 level support if you use the QAAC 2,71. I can't tell lossless from 160kbps VBR AAC 99.1% of the time, I've yet to see a sample break 320kbps VBR AAC after trying QAAC again. It blows away Vorbis & MP3 at 256kbps VBR and can match Opus at 160kbps VBR.
 
Top Bottom