Hi Bill,
I think PS Audio sees otherwise.
Remains a mistery to me why do you think MQA guys would know better what should be done on a code on FPGA chip, a code that they don’t know anything about…
MQA appeals or might appeal to the ones who use of-the-shelf dac ships…
Why do hi res recordings sound better? Your hearing peters out at 20k - mine at 62 goes to 12k and that was measured a couple of years ago so is likely worse now. So why does it sound better?
Much research has been done into it and the culprit is the brick wall anti-aliasing filter smearing transients. The higher that frequency of the brick wall filter the less smear you get. Now a DAC like the DS uses a filter at a phenomenal 10 times DSD - way beyond frequencies that will cause audible smearing. What you do is you put in whatever is coming in, then pad the data not that high out with zeroes, then apply the filter. Having read what the DS is doing its a bit more sophisticated, but that would be the effect:
http://www.6moons.com/audioreviewhttp://www.aes.org/e-lib/browse.cfm?elib=17501s2/psaudio/1.html
But at recording its another matter.
MQA does all sorts of tricks to attack that issue at the source. Depending on your background there is various explanations of exactly whats done. I will give the following link, but if it’s too advanced I know others:
http://www.aes.org/e-lib/browse.cfm?elib=17501
Just as an overview at 96k frequency it doesn’t take samples but rather, using a recording at higher frequency than 96k convolves it with overlapping triangle functions (actually the function is a bit more complex than that but no need to go into that in the overview). This has the effect of a transient not being interfered with as much. It has the effect of applying a 6db slow roll off filter from about 20k at a higher frequency like 192k (depending on how that convolving is done and the sample rate of the source) that does very little time smear compared to a brick wall anti aliasing filter. I will assume 192k here - but it can be a lot higher. It then chucks away the bits above 48k ie every second sample for a 192k source, which according to the Shannon sampling theorem means frequencies above 48k are folded back into the 48k range ie for 192k the frequencies from 48k to 96 k. But this is of little concern since due to the 6db rolloff, and amplitudes at that frequency are low anyway, it’s way below the noise floor.
Now we need to ‘unfold’ it to recover the 192k. The optimum way to do this depends on the function for convolving - the triangle function is very simple - it’s just linear interpolation. The way the DS works is it uses its own filter to upsample to 30bits PCM at 10 times the DSD frequency then converts it to DSD. Optimally it should use the filter that matches the convolving function it was encoded with ie simple linear interpolation for the triangle function. Indeed its even more complex than that. MQA takes into account the characteristics of whatever DA converter the DAC uses - it figures out the optimal filter from that. MQA has metadata on the convolving function used to create the file and its up-sampling needs to match that. The DS uses down-sampling and a delta zigma modulator to convert the 30 bits 10 times DSD PCM to double DSD for output. The output is very simple - it is just a filter that cuts out above about 80k. So the up-sampling to 30 bit 10 times DSD should use an optimal filter for MQA that matches the downconversion DSD converter and output filter.
What Roon is going to do is simply provide the 96K PCM. The unfolding to higher frequencies (ie any up-conversion done), if done in the DAC, will not be optimal. How will it sound - who knows. What I can say is in my experimentation with the DS it sounded very good. Although we didn’t do an A/B comparison (I will try and do one later) it was felt it was better than 96k masters. Why would that be? The lack of a brick wall filter in creating the MQA. Oh - on sources MQA encoded from masters that use brick wall filters they have tricks to compensate.
Thanks
Bill