Because the differences are highly likely due to the DAC performing better when processing data in the 24 bit form. If you put the 16 bit data into the 24 bit format "box" then that variable is removed from the test.
I never found any difference to speak of, you sure it's not just adding more high frequency noise which gives that impression?
Of course you are right Scott, and it seems that I tried to impress everyone, including myself
Now, the differences in mV (for Ref. Voltage=10V) :
8bit (LSB=39.0625mV):
INL: +0.025LSB = 0.977mV
DNL: +/-0.02LSB = +/-0.781mV
10bit (LSB=9.765625mV):
INL: +0.7LSB = 6.836mV
DNL: +/-0.2LSB = +/-1.953mV
12bit (LSB=2.441406mV):
INL: +0.5LSB = 1.221mV
DNL: +0.1/-0.6LSB = +0.244mV/-1.465mV
I can counter that with this excerpt from the BB seminar
The point from my part was not the “R-2R against D-S” war.
Nevertheless, although IMO in that screenshot there is a too simplifying conclusion, the B-B datasheets of these two DACs are not point to point comparable.
http://www.ti.com/lit/ds/symlink/pcm1702.pdf
http://www.ti.com/lit/ds/symlink/pcm1710.pdf
George
Great. We can test that theory once he's demonstrated an actual audible difference with a straight 16/44.1/48 vs. 24/96 comparison. Personally, I doubt Richard himself could reliably hear a difference between 24/96 and a 256kbps MP3. But let's start with a straight 16/44.1/48.
se
It's going to vary for every person and situation, for just about every reason you think of! I was surprised myself, at the difference - but it's a pointless thing to do, because it means storing huge music files on hard drives - it's a dead end ...
The subjective effect is directly opposite to your suggestion - at low res the treble is just high frequency noise, various layers of anonymous hissing and tzzz'ng going on - when playing the high sampling rate variant that "noise" reverts back to being part of the musical note or sound that produced that part of the spectrum; the high frequency harmonics start belonging to the instruments, and what you're hearing makes more auditory sense, rather than being a blur.
Last edited:
Yup, if you have a target bandwidth, and a target s/n ratio, then the maximum voltage must be that much higher than the johnson noise of the resistor, it turns out that the power depends upon the s/n you aim for, if the resistor is higher then the maximum voltage goes up. so 22kHz and 32 bit noise floor requires a target I/V stage to dissipate 26 Watts for 28 bit true resolution. If you really want a full 32 bit for some bragging right then expect to have some 6.7 KW dissipated at maximum output. Or for a real swinging ..... for 96 KHz and noise at the level of the smallest bit, expect to dissipate 29.3 KW at the D/A, if you can keep it all at 300K that is...
Eh! are you saying that high frequency information which is somehow lost definition wise in the process of resampling can somehow be regained by upsampling again? I was under the impression that once information is lost it cannot be regained by resampling.
The information is always in the source "data" - if you were able to look at the waveform, as say computer data, and had the capability of analysing precisely what was there, then you would see that the musical waveforms made sense, it wasn't "noise". But, at the moment of presenting that data to the DAC, the low res version of the music file was causing the circuitry in that area to not work as well as it should, how the theory says it should behave - the analogue output is being "corrupted", it is not an accurate representation of the data in the music file - you hear "poor" sound. But upsampling turns the data into something "nicer" for the circuitry to deal with - definitely, no "new" information is being added - the dots along the waveform envelope look more "analogue" for a start - and less "noise" artifacts are added to the analogue signal.
What do you mean it "looks more 'analogue'"? Coming out of the reconstruction filter, it is analogue. Even without the upsampling. So how on earth can it look "more analogue"?
se
No, the digital data representing the waveform "looks" more analogue - take a 20kHz sine wave, and the data stream encoding that at 44.1 sampling rate, and plot the points on paper - much better still, in a waveform editor like Audacity - and join the dots, which the software does automatically. It looks a mess, very nasty! Now, upsample again and again, to ridiculously high sampling rates - still, pure, digital data - and in Audacity, etc, it looks like a scope capture of a pure 20kHz sine wave, from an oscillator - you could feed that digital data straight into an analogue amplifier, with zero audible artifacts.
Which is partly why a cheap DAC and circuitry can do a better job, with more "information" being fed to it ...
Which is partly why a cheap DAC and circuitry can do a better job, with more "information" being fed to it ...
I bet Frank could put Humpty back together again and by upsampling him he would be an even better Humpty than before the fall.
So what your saying Frank is that with more guesses you can arrive at a more accurate end guess.
Last edited:
All the information that's musically significant, that's for sure - and you would dump a whole lot of ultra high frequency hissing on the way - no great loss ...
.I did an exercise some time ago of taking a valid hires file, one of the demos available for download - not a fake, as poor Mooly discovered to his chagrin,
- with real hires data(!), and played with it. Chopped off the audible spectrum, only had the above 20kHz left, and did a pitch change downwards on it, several octaves worth, until everything was now audible, well within the range of a cheapo full range speaker, then amplify what was there by a large amount so I could hear it ... and listened. Not very interesting, I'm afraid - various percussive whacks and such, and vague ringing noises - relating it back to the full blown music signal, there was nothing much worthwhile in it ...
Upsampling is just another way of partially getting to that desired result of the "correct" analogue waveform - there are no guesses, 32 bit processing guarantees accuracy to levels way beyond what any analogue circuitry could ever achieve, in its wildest dreams ...
What's fascinating with the old cartoons is that they worried about fluidity of movement - the amount of drawing effort to get every movement to have an "organic" feel - dare I say it, analogue in nature,
.Later stuff, they realised they could get away with being lazy, only the vaguest suggestion of the actual stroke was necessary - the Japanese took this to a high form, minutes can go by without a single thing on the screen moving - except a mouth perhaps,
.Yes, "digital" cartoons - we're so used to them now ...
- Status
- Not open for further replies.