[Community of A/V Enthusiasts]

Effects of downsampling and comparison of native 16/44 vs. 16/88

Just out of curiosity, I decided to compare the spectrum of signals recorded at 16/44 vs. 16/88, and also compare native 16/88 to what the spectrum looks like when I downsampled that 16/88 wav file to 16/44. Here are the results

For both cases (native vs. downsampled, native vs. native), there was a difference in the peaks. They were These differences were slight but noticeable (about 2 dB). For native vs. downsampled, there was also an increase in the noise floor, meaning that recording music at 88 kHz, editing it, then downsampling it to 44 (44.1 kHz to be exact), may result in more noise than if the music were simply recorded at 16/44 in the first place.

I used some lower frequencies for one test and higher frequencies for a second test.

This is just a first look at such things, so what do you think?

▶ Reply to This

Replies to This Discussion

Permalink Reply by John E. Johnson, Jr. on September 22, 2009 at 2:27pm

Here are the graphs.

Attachments:

▶ Reply to This

Permalink Reply by Larry V. Hryshko on September 22, 2009 at 2:35pm: First thought......2 dB isn't that slight.; ▶ Reply to This

Permalink Reply by Tyler Stripko on September 23, 2009 at 7:01am: I agree with Larry. 2dB is a pretty big difference in the peaks. Increased noise floor and lower peaks could certainly lead to a sense of constrained dynamics and lack of clarity for the downsampled vs. original signal. Of course, you'd need to hear the two different samples back to back to compare, but I bet a lot of folks would be able to perceive the difference.; ▶ Reply to This

Permalink Reply by Larry V. Hryshko on September 23, 2009 at 12:39pm: Second thought...........one of the benefits of digital recordings is that the noise floor is typically extremely low, often vanishingly low. So, while knowing that I had an increased noise floor might be annoying, there is a good chance that it might also be inaudible, even if it was a few dB (e.g. -90 to -88 dB). In contrast, losing a few dB of dynamic range seems to be a far bigger concern. as Tyler mentions.; ▶ Reply to This

Permalink Reply by John E. Johnson, Jr. on September 23, 2009 at 1:11pm: What I am referring to is the difference in the relative height of the original peaks vs. the subtractive peaks. They are not the same relative to each other. For example, look at the Native 16/88 subtracted from the Native 16/44 graph. Notice that the peaks in the 16/88 spectrum (the yellow spectrum) are all the same height. But, in the subtraction results graph (the red graph), the peaks are at different heights. The first peak (red graph) at 10 kHz is at - 11 dB, the second one at - 13 dB, and so on. This implies that the recordings at different sampling rates will be audibly different. What that difference is in terms of the sound, I don't know yet. This is just the beginning of some experiments. The red graph is on the same scale as the yellow graph. The red graph represents the difference between the two spectra. If the two spectra were exactly the same, the results would be a flat line at 0 dB. Noise is random between the two spectra, so even if the test signal data were exactly the same, the subtractive results would be a slightly jagged line at 0 dB instead of absolutely flat.; ▶ Reply to This

Permalink Reply by John E. Johnson, Jr. on September 23, 2009 at 1:20pm

OK, so now take a look at the following three graphs. All are with a 1 kHz sine wave test signal. They are at 16/44, 16/96, and 24/96. The 16/44 and 16/96 have pretty much the same THD+N, but at 24/96, the noise floor goes down, and the resulting THD+N is an order of magnitude lower. This gives more dynamic range. But, I am a little surprised that there is not really much less distortion going from 16/44 to 16/96, where the sampling rate is doubled.