| 
                                    
                                        | 
                                                
                                                
                                                    |  |  
                                                    
                                                    | xandcg 
  
   Rio de Janeiro, BR.
 Posts 219
 Joined on 09-07-2014
 
 
                                                            
                                                                | Post #: | 7 |  
                                                                | Post ID: | 22290 |  
                                                                | Reply to: | 21483 |  
                                                                |  |  |  |  
                                                    | fiogf49gjkf0dI found this description of how MQA works: 
 http://www.hifiwigwam.com/showthread.php?127311-Tidal-Announce-MQA-Streaming&p=2217736&viewfull=1#post2217736
 
 
 | | It is not a secret, what MQA is and does. 
 Here we go, in decreasing levels of abstraction.
 
 1) MQA is a kind of origami.
 
 That doesn't help you.
 
 2) MQA is about substituting triangles for rectangles.
 
 That probably doesn't help either, although it is a brilliantly succinct way of describing MQA.
 
 3) A bit more practical then.
 
 Start (ideally) with a 192/24 recording.
 
 Reduce this to 96/24 by means of a very short (i.e. fast) low pass filter.
 Such a filter allows some aliasing to happen, but since we assume that there is not
 much real audio in the 48-96kHz region the amount of aliasing is low.
 
 Now take the 96/24 recording.
 
 Grab the lower 24kHz band, at 24 bits, and noise shape it into a 48k/16b format.
 Put the result in a 48/24 container. Hey, we now have 8 unused bits per sample!
 
 Go back to the 96/24 original. Grab the upper band, i.e. 24 to 48kHz. Resample it at 48kHz.
 This is lossless because the sampling theorem says that any signal with a bandwidth smaller than Fs/2 can be
 sampled at Fs. So we end with the ultrasonic information in a 48/24 
format. We now assume that, again, there isn't really much ultrasonic 
info at all, so why let it occupy 24 bits? Not needed, so slam it down 
to 8 bits. We now have 48/8. Scramble these into pseudo noise and store 
the result in the lower bits of the 48/24 container we created in the 
previous step.
 
 
 Voila: 192/24 packed into 48/24.
 
 For replay: undo in reverse order.
 
 Very smart. Utterly useless
 | 
 | 
 
 
 Think for yourself, do not be sheep.
 |  |  |