Trouble with mixdown from 96khz

Question

I recently did the switch to 96 khz, and gradually working my way up to 192 khz, and i know a lot of you are going to say that i should just stick with 48 but i don't see the reason when my box goes up to 192. and when there is in fact a difference from 48 to 96 to 192. now i've made the recordings in 96khz, and when i mix it down to 48, it screws up the recording. i mean, it sounds good, but it doesn't sound anything like my song. it's weird. anyone know what this is all about? thanks in advance.

Answer 1

>Given that it is better to record and/or process at 24 bits (I have also read that sometimes there is no difference there), when should one expect a benefit from recording at sample rates of 98k or above and when should one expect benefits of recording at 24 bits?
Recording a 24-bit signal is essentially the same as recording a 32-bit one - the extra bits are scaling bits, and they only get invoked during processing operations. As for when you need to record at 24-bit as opposed to 16-bit - well, that gets slightly more interesting. If you are making an acoustic recording, then there's no way that you will be able to use all of even the 16-bit dynamic range up, never mind 24! That would give you a potential range of 96dB, and to get a venue that quiet... well, you can take it from me that you can't. BUT... that doesn't mean that you don't get benefits from 24-bit recording, because in a reverberant space that
is quiet, you will definitely capture reverb tails more accurately as they disappear into the natural dither the background noise is causing. So if you are making recordings like this, I'd say that it was worth it.
As for sampling at rates above 48k - well it's been researched, and proved conclusively that it's simply not worth it, because listening panels can't tell the difference, all other things being equal. Trouble is, that with a lot of kit, all other things
aren't equal, so equipment appears to behave better sometimes at higher sample rates. It is IMPORTANT to note, though, that this is a failure of the equipment to perform properly at lower rates, and not a function of the higher sample rate itself - that's irrelevant to the argument. What causes this has far more to do with things like dodgy Nyquist filtering, and generally poor component quality. Even some D-A converters are guilty from this point of view, running considerably better at higher rates than lower ones.
I have done critical listening to Apogee converters running at different sample rates into a good monitoring system, and you really can't tell the difference, whatever material is used. The entire Apogee chain was designed to run correctly at whatever sample rate you choose to use - and it proves to my ears (and everybody else who listens) that whatever differences people hear in lesser equipment are
not inherently due to any sample rate considerations. The other thing to bear in mind is that it's seriously difficult to reproduce even 20kHz in the average room with normal loudspeakers - so any system that can reproduce beyond this range is wasting its time and your processing power getting there - there is simply no benefit to be had at all - unless you are a bat/cat/dog, in which case it will annoy them quite a lot. (This has also been tested, incidentally!)
Hopefully, as converters improve generally, this will all inevitably become the accepted wisdom - and that day can't come soon enough as far as I'm concerned. The electronic tricks that you need to do to get over the filtering problems that have beset earlier designs are now better understood (this involves oversampling within the devices), and this approach is being used on most if not all of the newer devices that come to market - and now at a better price as well, which is why we are seeing them in more and more cheaper products; It's a technology that has come of age. These converters are now working well enough for the only remaining issues to be related to the analog components on the board, and it's now worth manufacturers choosing the best components they can, because the digital feeds to them can outperform them quite comfortably in many cases. But that, unfortunately, is what tends to make products more expensive, especially if you have to do it to a lot of channels.
I should also mention incidentally that the other thing that affects what is heard in some situations is the master clock stability in the system. In any given clock, you use the same circuitry for the whole thing, but increase the speed of it (by using less divider stages) for high sample rate running. When you do this, the
relative amount of jitter will reduce (although in effect it will remain the same). So it may well be that on systems with lower-grade clock circuitry, increasing the sample rate appears to make the conversion less 'veiled' because the relative jitter sidebands are reduced - but once again, that's down to poor design, and not inherently anything inherently due to the sample rate
per se. Designers are also recognising that this is the case, and clock stability is also improving in general.
>I understand that whatever word length the file is initially recorded at, the word length should be increased to 32 bits for processing, so you don't need to explain that part (unless I am mistaken).
No you aren't - that is quite correct.