Why is High Resolution Better?

[PeterSt]

... sort of. See here for the base of my thinking : http://www.computeraudiophile.com/content/relation-between-amplitude-and-sample-rate

 

As you will see in the last post I added in there today, I better withdraw my earlier "statement".

It was an idea I had at some stage, and it was still in my mind. Today I can't justify it.

 

It's just over my head ...

 

 

 

[PeterSt]

I will try to have a constructive contribution. Not sure whether that succeeds ...

 

Let me start with saying : of course hirez should be better. I see no reason why it shouldn't for theoretical reasons (and please skip my before ad-hoc response which seemed like a (bad) teaser).

Also, we all must assume well behaving DACs on the higher resolutions. If not, we are to be blamed ourselves.

So, what can prohibit hires from being better, *if* that would be the case (which clearly is what I am saying) ?

In random sequence of importance :

 

1.

We don't know how to rip the various formats.

Very likely, and personally I never sat down to improve it.I should though.

 

2.

Similar to 1., we don't know the source often. HDTracks is the example.

 

3.

We don't know how to interpret the "graphs".

Partly a theoretical aspect, but in the end of importance because what "you" interpret as good, I interpret as flawed.

Could be the other way around just as well.

 

4.

We use improper DACs.

Yes, it was my own assumption "that must be good", but we must not forget that a "re-interpreting" DAC is not the best for judgement. Not for listening, and not even for measuring.

Let's remember, hires material should not be subject to filtering anymore, and whether it will be afterall (in-DAC) we don't know.

See the difference with Redbook, which *needs* filtering, and which makes the SQ subjective to that application. Not so here, or at least it shouldn't be so IMO.

 

5.

Poor recording facilities.

It seems clear that when 20KHz mics are used, nothing hires will come from it. This may even turn out for the worse, because the distortion of 20KHz mics at capturing 25KHz music, will now be nicely captured in the hirez material.

 

6.

Poorly applied mastering techniques.

Like how 96 emerges from a 192 master.

I say this is 99% likely.

 

7.

2 channel rips from 5.1 etc. surround.

This just can't happen with software I know of. Still players (also hardware) can do it fairly well (without separate 2ch tracks).

 

8.

The hires is fake. Upsampled.

 

People may come up with more, but as how I see it, it is one, a combination, or even all of the above which makes hires not perform as could theoretically be.

 

If you carefully analyse the albums, there is always *something* wrong. This is so various that at least I can't tell the general flaw, apart from that it's always throughout in the album concerned. But there are also patterns visible in the recording companies.

 

#6 deserves special attention;

What I say (and see) is that it might be impossible to decimate e.g. 96 from 192 without applying the filters we know of, and which we don't want. The means of this filtering might be so commonly baked into the software concerned, that even when we'd want to avoid it, we can't. This is how 96 material pops up from our speakers with the exact same roll off as 16/44.1 does.

This time, however, it is about squeezed out high frequencies, while with 16/44.1 it is merely about squeezed in high frequencies but up to the roll off point. The result will not be the same I think, but it is beyond me how to reason it out (later maybe).

 

But let me add a most specalutive one (aim, shoot, but I hope you will miss);

 

I know what it takes to bring forward the high frequencies in the Phasure NOS1. This is not about loosing 3dB or something like that, while normally it is upto 6dB for Redbook (well, it helps), but this is merely about the energy it takes to get this done well. There is more, but I leave that to one's imagination. The point is, that I say that high frequencies require more energy (current) than low frequencies, which is as usual some opposite thinking. But, I guess people can do some math on this, and don't forget this is about D/A chips.

 

I really can't imagine what the difference will be for A/D chips (and everything needed to get the job done), that choking in its own energy needed. So, what I'm aiming at is the inherently more dull sound coming from 192 "native" recordings, just because the higher frequencies are attenuated all over.

Let me also refer to general specs of our audio apparatus which the most usually denote +/- 3dB for some (audio) range, which in my view is rediculous. I state -0.1dB (not even +) from just above DC to 20KHz which just is the measured truth, and audible. 6dB ? don't make me laugh ! But that will be true too, and audible.

 

So, the above too is needed for an "explanation" to explain it all, because otherwise I really can't for the 192 part.

 

I hope this was a constructive contribution, and not one with a negative view.

Peter

 

 

 

[sandyk]

"So, what I'm aiming at is the inherently more dull sound coming from 192 "native" recordings,"

Peter

Perhaps you need to use a different software player ? (grin!)

That certainly is not my experience, and due to my age, I am very sensitive to further loss of HF information.

Alex

 

[Teresa]

Some major and even some boutique labels have been caught trying to pass off 16/44.1kHz recordings by upsampling them to higher resolution. HDTracks are supposed to be checking these out before putting them up and some still make it through.

 

The audiophile labels provide bonafide high resolution recordings that are also well engineered and mostly natural sounding. Labels such as Reference Recordings, MA Recordings, AIX and Chesky for example and they are what should be used to appreciate what high resolution offers sonically.

 

The major and boutique labels will of course will be a gamble as even if they use high resolution digital to record, there is no guarantee that the other equipment they are using is of high resolution quality. And it gets worse with overproduced commercial rock recordings, rock can be well made but I think it is rare, there are a few rock recordings on audiophile labels (still very rare) Check out Billy Burnette's "Memphis in Manhattan" on Chesky 24/96 download to hear what is possible with rock music.

 

[Miska]

I've written so much about benefits for example in context of waveform reconstruction that I leave those parts out.

 

Only remark I want to make on the subject is about information capacity - resolution. In audio, resolution is product of level resolution (word length) and sampling rate.

 

44.1 kHz x 16-bits = 0.7056 Mbps

44.1k x 24 = 1.0584 Mbps

48k x 24 = 1.1520 Mbps

88.2k x 24 = 2.1168 Mbps

96k x 24 = 2.3040 Mbps

176.4k x 24 = 4.2336 Mbps

192k x 24 = 4.6080 Mbps

352.8k x 24 = 8.4672 Mbps

384k x 24 = 9.2160 Mbps

 

DSD(64) 2822.4k x 1 = 2.8224 Mbps

DSD128 5644.8k x 1 = 5.6448 Mbps

DSD256 11289.6k x 1 = 11.2896 Mbps

 

Now, nice part in sampling rate is that it can be used to "oversample" the signal by using more sampling rate than needed from frequency band perspective and use the extra for increasing level resolution. Upsampling is commonly used to describe the action of producing one kind of oversampled version of the signal. Increasing word length (number of bits), OTOH doesn't make it possible to increase bandwidth. And a bit like in statistics, increasing size of the sample set allows more resolution in the result (smaller error) - this is quite a lot like the low level linearity improvement of delta-sigma. So instead of one sample value, there are group of values. Sample values are integers - what if you needed to represent sample value between 1 and 2? With four samples one could already have:

[1, 1, 1, 1] -> (1 + 1 + 1 + 1) / 4 = 1

[1, 1, 1, 2] -> (1 + 1 + 1 + 2) / 4 = 1.25

[1, 1, 2, 2] -> (1 + 1 + 2 + 2) / 4 = 1.5

[1, 2, 2, 2] -> (1 + 2 + 2 + 2) / 4 = 1.75

[2, 2, 2, 2] -> (2 + 2 + 2 + 2) / 4 = 2

...this is oversimplified idea of oversampling + noiseshaping. Note that you can rearrange the four values without changing the result, while the frequency content changes...

 

If we go to video and image processing which is technically similar area, resolution there is [width x height x bits (x fps)]. There too, increasing DPI/PPI - width&height can be used to increase colorspace even if there are only few bits/colors by using dithers and rasters. An example commonly shown on magazine and newspaper prints (and in natural way on film). This is similar to dither and noise shaping in audio. Even displays don't manage to produce the actual shades, but instead shades of red, green and blue as small adjacent colors which is then seen as integrated color by eyes due to natural "reconstruction filtering". "Retina display" takes this to the next level by making pixels small enough to make individual pixels disappear altogether.

 

 

[audiozorro]

You bring up an interesting point. If we had super-vision or super-aided vision like that of Lieutenant Commander Geordi La Forge with his VISOR ("Visual Instrument and Sensory Organ Replacement"), perhaps we would see the small adjacent pixels of red, green and blue.

 

In Star Trek the Next Generation, the VISOR device does not reproduce normal human vision, but does allow the character to "see" energy phenomena invisible to the naked human eye, as well as allowing him to view things at infrared and at microscopic levels. This also allowed the character to see human vital signs such as heart rate and temperature, giving him the ability to monitor moods and even detect lies. In one episode "Heart of Glory", Captain Picard keys the main viewer to Geordi's visor allowing him to see the way he does. Seeing all the overlapping and different wavelengths was highly confusing to the Captain, prompting him to ask Geordi how he was able to differentiate between them all. Geordi's response made the comparison of child hearing many different sounds at once and eventually being able to pick out what they needed; it is a learned talent.

 

So much for science fiction, but any engineer knows the difficulties and limits in designing sensors, ‘eyes’ and ‘ears’ for computers or robots. Mechanical sensors in many ways mathematically or numerically superior, but feeble or infantile compared to their human counterparts.

 

In the digital age we focus on numbers but there is more at play than just numbers. More (high resolution) is better because we remove or go way beyond previous limits. But is high resolution synonymous with high fidelity? I would argue that it is not. And isn’t that what we really want, high fidelity and the emotional satisfaction we get from listening to music (sorry but I just listened to Jim Smith’s Get Better Sound DVDs this weekend).

 

I am an advocate for higher resolution, but I am more of an advocate for higher fidelity. If one were to just look at the numbers for resolution, 24/96 PCM and DSD64 are far closer together than 24/96 and 24/176.4 or 24/192. But for various reasons 24/96 and 24/192 are apt to sound more alike than 24/96 and DSD64. Now I believe I read somewhere that DSD128 and DXD are sonically very similar which perhaps someone else can elaborate.