You put the low pass filter in yourself (on the master bus at the end of the signal chain just before you render down to a lower samplerate). It doesn't matter what the filter is but you probably want it to be very high order (4 or 6 pole) so it attenuates high frequency sound faster.

The reason why you want to do this is because it eliminates inaudibly high treble and the harmonics of certain upper mid range instruments that sound over Nyquist, alias and fold back into an audible frequency range. The idea is to band limit your mix so that it is inside Nyquist before you start truncating samples. It won't alias if you do this properly.

People just refer to this as an anti aliasing filter. It has nothing to do with dither.

The POW-r modes you quoted are noise shaping algorithms for dither. Dither is basically a process where you add extremely low level noise to a mix just before you truncate bits and samples. When you throw away bits and samples you increase quantization error which creates spurious low level harmonics that are correlated to the original signal. With dither, you cover that correlated noise up with random dither noise which is not correlated to the original signal. It is supposed to sound more pleasant or so the theory goes.

The algos shape the noise so it is more or less present in certain frequency ranges. There are loads of these algos with different noise shaping functions like UV-22 and those POW-R ones (which are psychoacoustic so the noise is more present in frequency ranges that the human ear is less sensitive to).

When you dither the difference should not be dramatic and if your mix has a low noise floor you shouldn't be able to hear dither noise or quantisation noise at normal listening levels. If you can hear dither noise and psychoacoustic dither at normal listening levels in your mix then you are probably doing it wrong and the level of dither noise is way too high. You only want enough to cover the spurious harmonics introduced with quantization error and they manifest at a very low level indeed (we are talking on the order of -100dB or more).

You should never dither a mix unless you intend to render down to a lower bit depth and samplerate. Dither should always be the very last thing you do and you should never dither more than once (i.e. don't synthesize bass drums in soundforge, dither them, use them in a mix and then dither the mix and definitely don't dither a mix twice because you will be adding the same blanket noise to the signal twice).

If you don't know what dither is or what noise shaping algos are then you should not dither your mixes. Instead, you will want to dither test tones and read about the subject so that you are informed about it before you do it for real. If you do it wrong you can make everything sound worse. I don't know enough about dither to make informed decisions about using it on a mix so I pretty much never dither for real. I can follow the principles when they concern simple, linear waveshapes where I can see the quantization error on a spectrum analyser but I lose track of whats happening when lots of instruments are in the mix and some or all of them are harmonically complex.

For this reason I feel it is something best left to a mastering engineer who will be informed about how it works and what would work best for each mix. I don't feel as if this opinion will change as long as I can continue to make improvements to my sound design and mixing which undoutedly yield more dramatic results than me doing dither myself, even if it is done properly.

Last edited by Derivative on May-16-2008 at 12:47

I know what dither is but, I did not understand what you were talking about when you were saying "anti aliasing filters". And besides that, I just don't understand what the differences are between POW-r 1, POW-r 2,and POW-r 3. And if you say that the noise will be inaudible, what is the point in all these different types? In Live there are 5 ways to choose from to dither and I just don't see which one I am supposed to use.

When I was saying that I notice a difference, I mean that when running at 88.2 or 96khz I can hear the difference in the synths and effects,such as filters. And I would like to try and capture some of that sound that I hear, for when I put it on normal cd.

And as far as putting a low pass filter on, you said about 4 or 6 pole. I just bought Sugarbytes WOW filter which sounds great, but I do not know how many poles it is. I guess I should read the manual sometime soon. Or how can I tell if it is 4 or 6? It has many different filter types.

here is the link, awesome sounding
http://www.sugar-bytes.de/content/p...dex.php?lang=en

The synth should tell you either on the GUI or in the manual what filter types you get.

Filters are sometimes referred to by the rate of attenuation (36dB/octave). A single pole low pass filter attenuates signal with frequency above cutoff at a rate of 6dB/octave. Therefore a 2 pole low pass filter would be 12dB/octave, a 3 pole low pass filter be 18dB/octave (303 filter!) and a 4 pole low pass filter would be 24dB/octave.

If they are so lame as to not tell you what you are buying then you can always run 2 sine waves (osc1 and osc2, where osc2 is tuned an octave higher than osc1 and where both same amplitude) through the filter and output to a spectrum analyser. Then set the cutoff to the frequency of osc1. Read off the y axis and check how much the osc2 has its volume lowered by the filter. If its like 24dB lower than osc1 then you know the filter is attenuating frequency at the same rate as a 4 pole low pass filter. i.e. it is a 4 pole low pass filter.

It would be incredibly lame if you had to do that to figure out what filter you bought though. It must be in the manuel.

Regarding dither: I didn't say quantisation noise was inaudible. At normal listening levels you will not be able to hear it above the ambient noise in your room (most of which probably comes from the fans and motorised parts in your computer).

To hear quantisation noise from truncating a 24 bit source down to 16 bits you will have to listen very very loud.

POW-R 1, POW-R 2, POW-R 3 are all psychoacoustic noise shaping algorithms with different probability density functions. According to the wiki article:

POW-R #1: optimized for simple program of average dynamic range, such as spoken word.
POW-R #2: optimized for low dynamic range program, such as rock music.
POW-R #3: optimized for complex, high dynamic range program, such as orchestral music.

I couldn't say for sure whether that information is accurate. Consult POW-R's website to see if they provide details of the algos that they sell.

Edit: For those interested in dither by the way, the wiki article is excellent and explains it much better than I have with pictures and everything. go go go!

Last edited by Derivative on May-16-2008 at 18:44

Do you really have to apply a filter manually? Surely most DAWs will do it automatically for you?

Just a bit OT here:

I found in cubase that it's possible to work in 44.1 (less cpu usage) and then simply export the track to 88.2 or any other samplerate. the question is.. does it actually export to higher samplerate? because I'm using realtime export (using hardware..) and it doesn't seem to increase cpu usage at all, even though there are quite a few plugins active..
Anyone know? Or maybe there's a way to check it?

quote:

Originally posted by Eldritch Do you really have to apply a filter manually? Surely most DAWs will do it automatically for you?

It just explains the proccess of dithering. No need to use the good ol' 2pole on the mix
..unless you're going for an "old school" sound

quote:

Originally posted by LfmC Just a bit OT here: I found in cubase that it's possible to work in 44.1 (less cpu usage) and then simply export the track to 88.2 or any other samplerate. the question is.. does it actually export to higher samplerate? because I'm using realtime export (using hardware..) and it doesn't seem to increase cpu usage at all, even though there are quite a few plugins active.. Anyone know? Or maybe there's a way to check it?

I haven't tried it, but I wouldn't think you can go up after the fact. If you record something at a sample rate of 44.1, that is the raw "image" or sound... when you upscale it, it would be impossible to create 88.2 out of that, right? I'm thinking in terms of images... if you take a photo at 1024x768, the pixels are set. If you upscale it to 2048x1536 or whatever, the pixels are already set... the image can technically be upscaled, but it is distorted because there's no way to make more quality out of a recorded sample. Just a guess.