|
|
Post by Chuck Elliot on Feb 2, 2011 7:36:12 GMT -5
FWIW, I have wanted to do this for some time. This is a very sort snapshot of both the CD(44.1/16) and DVD-A(96/24) versions of the same song. Source: Steely Dan - Two Against Nature - Cousin Dupree. Dots are the sample points.  44.1/16  96/24 |
|
|
|
Post by putt4doh on Feb 2, 2011 8:19:25 GMT -5
Very cool visual representation...
I assume the dots represents points of data. So logically, more dots=more data. More data means better SQ?
|
|
|
|
Post by Nemesis.ie on Feb 2, 2011 8:20:21 GMT -5
Nice - of course the real questions is at what point do we actually start hearing a difference. IMO 48KHz/24bit is probably pretty optimum/a good balance in that regard.
|
|
|
|
Post by Porscheguy on Feb 2, 2011 8:28:33 GMT -5
What does this mean in reality?
|
|
Erwin.BE
Emo VIPs 
It's the room, stupid!
Posts: 2,268
|
Post by Erwin.BE on Feb 2, 2011 9:09:27 GMT -5
What does this mean in reality? Fascinating! The real-world soundwave, AKA "analogue" is a fluent sineform, ie has an infinite amount of dots. Digital sound takes a predetermined amount of dots of that sineform ("samples" it). The more dots, the more accurate one can later reproduce (mimic) the original sinefrom. You see that the 24/96 file produces a much more convincing sinewave, where the 16/44 (CD) file is quite rude in comparison. Not that you actually hear the difference on a clockradio, you need decent speakers. And indeed, even on my rather expensive, reveiling and neutral stereo, 24/48 sounds quite fantastic. The few files I have for comparison, don't show audible difference to 24/96. As always, better to have a fantastic recording on CD, than rubbish on High res. 16/44 can be very good, but there is no room for error, unlike with 24 bits, where there's more bits to waste. How long is the sample anyway? 0.01 second or so? |
|
|
|
Post by Chuck Elliot on Feb 2, 2011 9:44:48 GMT -5
Even less than that, about 1 ms. I needed to do that to see the data points.
One item of interest to me is that on the 3rd valley back, the 44.1 completely misses the amplitude of the -peak because it is between samples.
|
|
NorthStar
Seeker Of Truth
"And it stoned me to my soul" - Van Morrison
Posts: 0
|
Post by NorthStar on Feb 2, 2011 9:49:49 GMT -5
Reminds me of the people that use dots to mimic replicas of real people from computers and animate them based on those dot points (the more the better) to be used in films. You know...
* They first use real actors in suits with several dots on it, then they are recorded in action, and finally transfered to computers for finalisation of the full scale real life actors. ...And those dots are luminous, similar like in your graphs.
...Like in "King Kong" for example (Peter Jackson's one).
Just sayin'.
|
|
geebo
Emo VIPs
"Too bad that all the people who know how to run the country are driving taxicabs and cutting hair"
Posts: 24,211
|
Post by geebo on Feb 2, 2011 9:58:36 GMT -5
Even less than that, about 1 ms. I needed to do that to see the data points. One item of interest to me is that on the 3rd valley back, the 44.1 completely misses the amplitude of the -peak because it is between samples. Very very interesting. Thanks for posting. I wonder what the ball park frequency is thats represented here... |
|
|
|
Post by Chuck Elliot on Feb 2, 2011 10:03:25 GMT -5
Reminds me of the people that use dots to mimic replicas of real people from computers and animate them based on those dot points (the more the better) to be used in films. You know... * They first use real actors in suits with several dots on it, then they are recorded in action, and finally transferred to computers for finalisation of the full scale real life actors. ...And those dots are luminous, similar like in your graphs. ...Like in "King Kong" for example (Peter Jackson's one). Just sayin'. Avatar was another step up from even this. I wonder how long it will be before we have completely virtual actors that you can't distinguish from real actors? |
|
|
|
Post by Chuck Elliot on Feb 2, 2011 11:40:30 GMT -5
Even less than that, about 1 ms. I needed to do that to see the data points. One item of interest to me is that on the 3rd valley back, the 44.1 completely misses the amplitude of the -peak because it is between samples. Very very interesting. Thanks for posting. I wonder what the ball park frequency is thats represented here... I calculate that the "very close to" sine wave in the second half is about 13kHz |
|
geebo
Emo VIPs
"Too bad that all the people who know how to run the country are driving taxicabs and cutting hair"
Posts: 24,211
|
Post by geebo on Feb 2, 2011 11:51:07 GMT -5
Very very interesting. Thanks for posting. I wonder what the ball park frequency is thats represented here... I calculate that the "very close to" sine wave in the second half is about 13kHz Very much in the audible range then. And it could only get worse from there! |
|
|
|
Post by shayking on Feb 2, 2011 12:36:45 GMT -5
notice how smooth the 24bit samples roll vs the 16bit sharp* peeks.
thats what we always want. (crossovers/cpus/almost all eletros)
you can Hear* it to with a good set of fronts.
|
|
|
|
Post by Mr. Ben on Feb 2, 2011 14:32:26 GMT -5
Are you just taking the sample points and connecting them with a line? A DAC doesn't do that. A DAC tries to fill in the gaps.
I think a better method would be to sample the output of a player at high resolution. In other words, play a CD, and sample the output at 24bit/192khz, and plot that. Then do the same thing, but with the 24bit/96khz disc. I'd bet they look a lot closer than your graphs above.
|
|
|
|
Post by Chuck Elliot on Feb 2, 2011 15:14:02 GMT -5
....A DAC doesn't do that. A DAC tries to fill in the gaps...... ....I'd bet they look a lot closer..... The DAC part of a DAC unit does exactly that. It's the low pass filter after the conversion that smooths things out. I bet they'd be closer too, but not the same. |
|
|
|
Post by hokienation on Feb 3, 2011 7:27:24 GMT -5
cf,
Help out the uniformed....to me it looks like you used less sample points for the 44.1 and therefore, have less curvature, no?
|
|
|
|
Post by Nemesis.ie on Feb 3, 2011 7:54:27 GMT -5
I think that is the whole point - the "frequency" quoted is not the audible range covered (although you do need a minimum of around 2 x the samples of the max frequency you want to sample to have success) but the number of sample points. 48KHz means there are 48000 samples taken every second. So for example, if you are using an EQ system for only your sub fequencies (10 to 200Hz as an example), at 48KHz you still get 480000 samples per second just for that range where you at most are only seeing 200Hz of audio frequency, which is probably plenty.  (That's "oversampling" by 240 times - at 20KHz you would only be getting 2.4 times the samples versus the frequency. People seem to confuse this rate (44.1KHz, 48Khz, 96KHz etc.) with the audio range, it is not, it is the number of samples/second. It might be clearer if it was called e.g. 48kSPS. The number of bits is then the range each sample has - 1 bit would only give you 2 points (max level and zero level), 16 bits gives you 65535 (i.e. that many steps between 0 and max) 24 bits gives you a little over 16 million which is a massive increase over 65535, so you can have much finer steps between one level and the next one sampled. High sample rates (e.g 96Khz, 192KHz) will have the most benefit when sampling high frequency sounds). I hope that makes it clearer for folks. |
|
|
|
Post by UT-Driven on Feb 3, 2011 11:18:12 GMT -5
Thanks for sharing this. It looks almost exactly as I imagined it would (not the specific waves for the Steely Dan song, but the difference in the curves). The 24/96 signal definitely looks more analog than the 16/44.1 signal.
Doug
|
|
|
|
Post by audiodragon on Feb 3, 2011 11:48:40 GMT -5
This is a very interesting and informative thread! I would love to see that same sample of music with a typical 128kb/s mp3 as the source, it would be a great illustration of how much information is thrown out by compression. Come to think of it, considering the sample size, ALL of it may be thrown out by compression It is amazing to me what the average person will consider acceptable, and even PAY FOR! (How many iTunes downloads can I fit on my iPod? ;D) |
|
|
|
Post by Vespid on Feb 3, 2011 12:15:51 GMT -5
I think a better method would be to sample the output of a player at high resolution. In other words, play a CD, and sample the output at 24bit/192khz, and plot that. Then do the same thing, but with the 24bit/96khz disc. I'd bet they look a lot closer than your graphs above. Of course they would. Your example would compare two 24 bit streams, at different sample rates. His original graph compares 16 bit to 24 bit. Big, big difference. **Edit** oops. I posted this before reading nemesis post above. See his post for a much better explanation! |
|
|
|
Post by stuofsci02 on Feb 3, 2011 12:34:45 GMT -5
This is a very nice post.
I think, however, that everything needs to be put into context. The higher sampling rate will definately reproduce a more accurate output. That said at sampling of 44.1K you can accurately reproduce frequencies of up to 22,050HZ. It is generally accepted that people cannot hear beyond 20,000 Hz, and so this should be sufficient.
What I think it much more interesting and can make more difference is the added bit depth. With 16 bit you can have only 65,536 different amplitudes. This is a lot, but probably still lacks in achieving the most out of digital. In contrary 24 bit has almost 17 million possible amplituted. This will allow the capture of even the sublest changes of output.
|
|
(That's "oversampling" by 240 times - at 20KHz you would only be getting 2.4 times the samples versus the frequency.
