|
|
Post by moovtune on Feb 3, 2011 14:12:32 GMT -5
It looks like you used Adobe Audition for the snapshot. Are the number of points it creates in drawing the image set the same for each clip or does it change automatically depending on the clip sampled? In other words is the difference simply how the program draws the sampled clip? Is the magnification of the wav file the same in each case?
|
|
|
|
Post by Chuck Elliot on Feb 3, 2011 17:04:42 GMT -5
Assume the following: Blue - Sine wave is 20 kHz Red - Aligned to peak amplitude Transitions are sample point Green - Sample point shifted from peak Both sample rates are the same.  Both digitized tracks create a 20kHz square wave which post DAC low pass filtering converts to a 20k sine wave, same as the original. But, and it's a big but, the amplitude is wrong(green). Sample depth, 24/16, isn't going to fix this. Only increasing the sample rate. I'd love to have someone who really knows explain this to me, as I don't pretend to be an expert! |
|
|
|
Post by Chuck Elliot on Feb 3, 2011 17:07:03 GMT -5
It looks like you used Adobe Audition for the snapshot. Are the number of points it creates in drawing the image set the same for each clip or does it change automatically depending on the clip sampled? In other words is the difference simply how the program draws the sampled clip? Is the magnification of the wav file the same in each case? Images were screen captures from Sony "Sound Forge Audio Studio 10.0", then edited in Photoshop. Two different files adjusted by hand, or should I say mouse, to as close as possible to the same starting point. Graph above is from Excel. |
|
|
|
Post by Mr. Ben on Feb 4, 2011 7:16:58 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. You didn't understand what I proposed. cfelliot is showing us the differences between the inputs to a DAC. I was proposing measuring the outputs of a DAC (meaning the output of a CD player or DAC like the XDA-1, not the output of a single DAC chip). That's the more important piece, in my opinion, because that's what goes to the speakers. The differences aren't as dramatic at the output end, but I don't know how different they really are. |
|
|
|
Post by Chuck Elliot on Feb 4, 2011 11:06:29 GMT -5
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. You didn't understand what I proposed. cfelliot is showing us the differences between the inputs to a DAC. I was proposing measuring the outputs of a DAC (meaning the output of a CD player or DAC like the XDA-1, not the output of a single DAC chip). That's the more important piece, in my opinion, because that's what goes to the speakers. The differences aren't as dramatic at the output end, but I don't know how different they really are. Mr. Ben, I totally agree with you that what you suggest and is the "final" test! What I would love to see is the results of this test using a "known" test signal recorded at various bit depths and sample rates. This sample would be a mix of known frequencies. The measurement would be of harmonic and inter-modulation distortion between the different rates. Love to see the hell various compression formats cause here! Albeit, if I only had the test equipment to perform such a test! |
|
|
|
Post by Mr. Ben on Feb 4, 2011 16:16:31 GMT -5
You didn't understand what I proposed. cfelliot is showing us the differences between the inputs to a DAC. I was proposing measuring the outputs of a DAC (meaning the output of a CD player or DAC like the XDA-1, not the output of a single DAC chip). That's the more important piece, in my opinion, because that's what goes to the speakers. The differences aren't as dramatic at the output end, but I don't know how different they really are. Mr. Ben, I totally agree with you that what you suggest and is the "final" test! What I would love to see is the results of this test using a "known" test signal recorded at various bit depths and sample rates. This sample would be a mix of known frequencies. The measurement would be of harmonic and inter-modulation distortion between the different rates. Love to see the hell various compression formats cause here! Albeit, if I only had the test equipment to perform such a test! Yeah - I don't want to take anything away from what you've posted - it's interesting to see. But it makes me want to see more  I do have the ability to do the experiment I proposed, but I can record only up to 24bit/96khz, not 192khz. I've never taken the time because I listen to virtually nothing but vinyl, SACD, and CD though. |
|
|
|
Post by Chuck Elliot on Feb 4, 2011 16:59:13 GMT -5
Mr. Ben, I totally agree with you that what you suggest and is the "final" test! What I would love to see is the results of this test using a "known" test signal recorded at various bit depths and sample rates. This sample would be a mix of known frequencies. The measurement would be of harmonic and inter-modulation distortion between the different rates. Love to see the hell various compression formats cause here! Albeit, if I only had the test equipment to perform such a test! Yeah - I don't want to take anything away from what you've posted - it's interesting to see. But it makes me want to see more I do have the ability to do the experiment I proposed, but I can record only up to 24bit/96khz, not 192khz. I've never taken the time because I listen to virtually nothing but vinyl, SACD, and CD though. Not hard to set up, but the need for a high quality spectrum analyzer (about $4K) is the breaker. You just happen to have one? - LOL  |
|
|
|
Post by Vespid on Feb 5, 2011 6:55:49 GMT -5
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. You didn't understand what I proposed. cfelliot is showing us the differences between the inputs to a DAC. I was proposing measuring the outputs of a DAC (meaning the output of a CD player or DAC like the XDA-1, not the output of a single DAC chip). That's the more important piece, in my opinion, because that's what goes to the speakers. The differences aren't as dramatic at the output end, but I don't know how different they really are. Maybe so. I thought the point of the original graph was to visually point out the rather large differences between 16 and 24 bit? I agree that you would not see a lot of difference between 24/96 compared to 24/192, regardless if you compared the time vs amplitude sample points overlayed on the simulated wave(input) or the resultant waves themselves (output). I guess how much you would see would depend on the resolution of the graph no?  |
|
edrummereasye
Sensei
"This aggression will not stand, man!"
Posts: 438
|
Post by edrummereasye on Feb 10, 2011 18:57:20 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. Actually...as I understand it, Redbook CD contains some additional "sub-channel" info, so you don't quite get 22,050 as the max frequency, more like 20-21kHz. However, the logic behind choosing 44.1 was exactly that - Nyquist's Theory is that you can reproduce a given waveform by sampling at twice the highest frequency it contains. So, 44.1, if it all was used for audio, would pretty much cover the range of even well-above-average human hearing. As I said, I've heard it doesn't quite all get used for audio, though OTOH, I've never heard anyone claim their ears were good enough to hear what gets clipped, on a well-produced CD (meaning well-recorded, well-mixed, well-engineered, and played back on equipment that is adequate to the task). For those who are struggling to understand, perhaps this will work...to record a CD, one sample containing 16 bits of information is taken every 1/44,100th of a second. In 24/96, one 24-bit* wide sample is taken every 1/96000th of a second...the graph shows the frequency with which samples are taken, not the number of bits...so there is more space "between" the samples in 44.1kHz sampling. You can also see from these graphs why older versions of Windows were often evil, and why one must be cautious and/or use ASIO even today - look at the graph from the CD. Now, imagine that *that* waveform (or one reconstructed from it), and taking a sample every 1/48000th of a second...you're starting with one that is not perfect, then you are taking samples that often fall "between" the originals...sampling an interpolation much of the time. Then trying to construct a new waveform from *that*...ugly. To the OP: Good stuff, thanks for posting! I've long tried to picture this in my head, I always tended to picture the samples as being very close together, with small spaces in between...guess it's a matter of how closely one looks! This gives an interesting perspective and has sparked a good thread. And may I complement your tastes...I've seen Donald & Walter twice now live, and been listening on CD for many a year. |
|
edrummereasye
Sensei
"This aggression will not stand, man!"
Posts: 438
|
Post by edrummereasye on Feb 10, 2011 19:12:00 GMT -5
You didn't understand what I proposed. cfelliot is showing us the differences between the inputs to a DAC. I was proposing measuring the outputs of a DAC (meaning the output of a CD player or DAC like the XDA-1, not the output of a single DAC chip). That's the more important piece, in my opinion, because that's what goes to the speakers. The differences aren't as dramatic at the output end, but I don't know how different they really are. Maybe so. I thought the point of the original graph was to visually point out the rather large differences between 16 and 24 bit? I agree that you would not see a lot of difference between 24/96 compared to 24/192, regardless if you compared the time vs amplitude sample points overlayed on the simulated wave(input) or the resultant waves themselves (output). I guess how much you would see would depend on the resolution of the graph no? His graph shows the sample points in time; in one graph, the samples (points) are 16 bits wide, in the other, 24 bits; but they are both represented as points. In other words, no, his graph doesn't show the effects of 16-bit vs. 24-bit; it shows that sampling every 96000th of a second yields more points than sampling every 44,100th of a second. (A bit more than twice as many, as mathematics would suggest - 96/44.1 = ~2.18). |
|
|
|
Post by bobbyt on Feb 11, 2011 1:30:27 GMT -5
All you can see on the graph is how many samples are taken. 96khz has a little over twice as many as 44.1khz. That does increase your accuracy in trying to capture every peak and trough.
It doesn't show 16 vs 24 bit, which is a huge difference. Each sample has to fit into a discrete value, and going from 16 to 24 bit gives you 256x as fine values (65k vs 16 million).
|
|
