View Single Post
Old 28th December 2007   #76
Dan Lavry
Gear addict
 
Join Date: May 2005
Posts: 437

Quote:
Originally Posted by MickeySmid View Post
Thanks, Dan!!... Is it not also then wizer to record higher than 44 if we run a lot of plugins in our DAW? These can then also procces the 96khz recordings and add things like corolation (vintage warmers etc), and get a better result like you state above...

So if we use a lot of plugs it's def. best to record above 44?!

(This is an endless discussion... I know. Somehow I trust mr. Lavry... !!!

thanks,
Mickey
Hi Mickey,

Thank you for the nice comment.

As a rule, digital processing is very transparent for linear processing. As a rule, alias distortions show up when doing digital non linear processing, such is the case unless the process was designed to be free of aliasing, and the AD10 digital alias free emulation ™ is the only unit around to offer that. To achieve alias free distortions, one needs to work at higher sample rate, but that is only one little requirement. Unless the process is specifically designed to be alias free, going to higher sample rates may help some, but will not eliminate the aliasing.

IN ANALOG, any non linearity generates high frequencies, and such non audible energy can be filtered out (to prevent it from impacting gear and speaker performance).

IN DIGITAL, any non linearity “tries” to generate high frequencies, but as we know, when the frequencies exceed Nyquist, they “fold back” into the audible range. Aliasing can not be filtered out. By the time it happened it is already too late to do anything about it. Unlike analog, doing something "later" is not an option.
At first one may think that moving Nyquist up will solve the problem, but such approach proves very disappointing. Why? While the higher frequency harmonics do tend to have reduced amplitude, the amplitude reduction is pretty small for the ear, which is near logarithmic in its response to amplitude. A 1% on a scope is difficult to see, but for the ear it means only -40dB…

So digital is great for accurate summing, EQ, reverb… and all linear processing. But digital is terrible for limiters, tube and transformer emulation and other non linear processing (unless it is customized to work without aliasing).

What is the definition of linear process? For someone with math background it is a simple equation. I will tray to explain it without math, in a somewhat loose manner:

Say you wish to run some process on some some entity.
You can process the whole entity and get some outcome, lets call it outcome A.
Now lets break the entity into two or more pieces, and run the same process on each of the pieces. After doing so, lets combine (add) all the individually processed pieces, and we end up with an outcome B.
If the outcome A is the same as outcome B the process is linear. If, however, the outcome A is different then B the process is not linear.

Examples:

Say I have 2 channels and I boost the gain of each channel by 3dB, then I sum them into one channel. The result will be the same if I added the channels first, then boosted the combined single channel by 3dB. Thus the process of gain and attenuation is a linear one, and works well with digital processing.

Now lets take 2 channels and run them through a tube. The combined signal has larger range (think of the peak to peak span), causing the combined signal to “bent” more. The same with a transformer, where the impact is greatest when you “push the device” towards full magnetization. When running each channel at a time, the levels are lower and the coloration is much lower as well. When you add 2 channels with low coloration in each, the total will have less coloration compared to the single combined larger signal.

So a linear process is when the “sum of the processed parts” is the same as the “processed sum of the parts”. That includes EQ and reverb. Tube and transformer emulation is non linear and so on.

Most DAW processing is linear (mixing, EQ, reverb …) so DAW work is very fine, be it at 44.1KHz or higher. But regarding digital non linear processing, unless the process was very deliberately crafted to avoid or greatly reduce aliasing, I would look for an analog method.

Regards
Dan Lavry
Dan Lavry is offline   Reply With Quote