Elektron Musical Instruments

>Well, yes and no. Normally you would apply alias-filter *before* the 
>decimation (the actual SRR), but you can also apply it afterwards. 

If it's applied afterwards the result will be aliasing. The 'internal 
sampler' wont know if the sample points it's about to digitize is a 
part of the desired audio spectrum or the frequencies above half the 
sample rate. The end result will be alias tones in the spectrum below 
half the samplerate. 

>For  
>example, if I have a rapid sin wave, it turns into random-like square 
wave  
>with large amount of SRR. But the result is aliased, the square wave 
is  
>not bandwidth-limited if the playback is not interpolated in anyway. 

Random-like square wave..?

Sample rate deals with frequencies. It's rather black and white, 
they're either below half the sample rate or above. Those below will 
be reproduced with 100% accuracy. Those above will contaminate the 
real sampling spectra with alias tones and must be filtered out for 
proper sampling. There is no quantization in the timeline of sampling, 
which is what sampling rate is concerned with. The sample clock is 
steady, the sample values are not moved back and forth in time to fit 
within a grid.

Quantization only applies to the level of each sample, the bit depth. 

 
>n fact, I take my words back. It would take more inspection to tell  
>hether the result in MM's SRR produces aliased waveforms or not. And 
it  
>isn't probably even worth it.

I'll parrot Daniel, if the SRR doesn't produce aliasing, it would be 
the same as a plain filter. If it doesn't sound like a filter but 
produces additional tones in the system, it's aliasing. No closer 
inspection needed. ;)
 

This may be old news, but these are often confused subjects. Hope this 
may clear things up a little.


Andreas

PS: perhaps the sampling theory paper at lavryengineering.com's 
support pages may be useful.