Going back a few pages, the topic was brought up about low pass filters. It's important to have a low pass filter before an A/D to avoid aliasing. Ideally it would be a very sharp drop. If you are recording at 44.1k, you would like to capture everything up to at least 20k, not have a slow decay curve between 16-20k. This presented the question, how good is the one in my R4? In my SRC software? So I did a little home experiment which almost anyone can recreate, no special equipment required.
Cut to the punchline... 95% (maybe 99%) of what we have been discussing in this thread becomes a non-issue. The LPF on my recorder is plenty steep enough, so is my SRC, and the SRC's seem to do a really great job.
Generate a "standard sweep":
In Audacity, there is a menu option "Generate -> Tone". Create a 16k sine wave, 30 seconds is fine, amplitude of 0.2. Do the same for 17k, 18k, 19k and 20k. Now you have 5 sweeps and when you mix them all together it's an amplitude of 1.0, and these discrete building blocks combined. I did this at 44k because the anything in 16k-20k can be completely represented at 44k. But just for the heck of it, I repeated at 96k, and I was able to see that there really was no difference between the generated sweeps, so I stuck with the 44k one.
Run that into your recorder
I ran TRS cables from my sound card, an M-audio Profire Lightbridge, to my Busman Tmod R4. Not the best DAC, but I figured I would try the easy one first, and if I thought it mattered I'd try other ones I have (I later decided this was good enough). Press PLAY on the computer, press RECORD on the R4. Not synchronized, but as long as you get 25 seconds out of the 30, close enough. Do runs with R4 at 44k, 48k, and 96k. I also repeated the set using SPDIF out of the sound card > SPDIF into the R4. It's common knowledge that the R4 resamples DIGI in, so a 44k digi signal can be resampled to 96k... just curious how that would compare.
Compare the recorded sweeps:
Pulling up the recordings from the R4... the first thing I noticed is that I can see a visible difference... interesting. In the picture "3wavs" 44k is at the top, 48k mid, and 96k at the bottom. Then look at the individual plots. The 44k looks very similar to the standard, and the 48k and 96k plots have high frequency resonances which aren't in the original. By definition, that's distortion, but realistically I can't hear it, so it probably doesn't matter. Then look at the 20k peaks on the plots... they are all coming up to pretty much 0 db, including the one I recorded at 44.1k, which means there is not a low pass filter reducing it at 20k. It's hard to tell for sure in the graphs, but in the next post, I have the data exported as text files. If you compare the 20k peak on the generated standard to the 20k peak on the analog recording, it's off by about 0.1db. I'm going to call that a result of analog transfer rather than a result of LPF rolloff.
Then I resampled the 96k file to 44.1k using "ssrc high precision". That's what I've been using for years, first on linux, then on windows. Luckily it's one of the shining stars in Page's comp studies. The bottom line is that the resampled one is virtually identical to the standard, and the one recorded at 44.1k. I kind of thought these SRC software options were a necessary evil, the output of which is about like adding an analog gen. I'm proven wrong. They are excellent, at least in this sample.