Gearspace.com - View Single Post - MQA discussion at Denver RMAF
View Single Post
Old 2nd September 2017 | Show parent
Here for the gear
Originally Posted by mcgilroy ➡️
I also think Aureliac has an unclear stance on MQA.
Auralic and Klinkt Beter figured out you can use the sox library to process MQA files. Sox is a high quality resampling and DSP library.

What we found is when upsampling the 24/44.1 MQA distribution files from 2L.no with sox, the sound difference with he original DXD masters tends to be very small. Some prefer the DXD, some prefer the upsampled distribution files.

This was on done on Amphion's best speaker, the Krypton 3, which goes further than the Amphion Two 18, which is a great studio speaker.

When comparing DXD with MQA on a mytek, there's more echo & reverb. Voices sound polished.

KIH #46 - MQA's missing link? | DAR__KO

MQA does not encode frequencies higher than 48 Khz as their format is basically 24/96 or 24/88.2 where the lower half of the 48 Khz or 44.1 Khz audio band is sampled at 17 bits, and the upper band is lossy, borrowing encrypted bits from the remaining 7 bits, which only an MQA decoder can decode.

Not all agree on the 17+7 bits distribution, some claim it's 15+9, some say 16+8.
For MQA cd, it's 13+3 which is for sure.

So for sampling at multiples of 44.1K

- MQA is max 17 bits
- audio frequency range stops just below half of 88.2 Khz sampling = 44.1 Khz
- any ultrasonic content above 44.1 Khz is lost
- ultrasonics are lossy

So for sampling at multiples of 48K

- MQA is max 17 bits
- audio frequency range stops just below half of 96 Khz sampling = 48 Khz
- any ultrasonic content above 48 Khz is lost
- ultrasonics are lossy

So MQA can never describe a 50 Khz ultrasonic tone.

So what happens with MQA files based on source files that were in a higher sample rate than 24/88.2 or 24/96 ?

From the first unfold which renders a 24/88.2 or 24/96 PCM format with an effective resolution of only 17 bits, MQA upsamples using a very simple minimum phase upsampler to the original resolution that can run on the very cheap processor inside an AudioQuest dragonfly.

They also apply weird anti-ringing filters which kills the post-ringing tail of the minimum phase upsampler. They have 32 filters to do this, and the MQA files knows which pre-defined filter to use:


This filter is what makes MQA sound thinner, with more distortion. They are trading one mistake for another.

So the alternative is to upsample MQA files directly with a high-quality minimum phase upsampler instead, and ignore the weird filters that try to get rid of post-ringing.

Post-ringing is not the problem, pre-ringing is, as our hearing is sensitive to pre-ringing, not post-ringing.

What we found is when upsampling MQA with minimum phase, without using an official decoder, it sounds more true to the original, than when running it through an official MQA dac like the Mytek Brooklyn.

We used files from 2L.no to do this test, as they have both the original DXD files which have been sitting there for years, and since recent, the MQA versions.