MQA – Further explained

Sometimes things are simpler than you think. Also sometimes you have all pieces of the puzzle and you still do not see the big picture. MQA seems to fall into this category.

Some people notice the possibility to employ DRM on the extended part of the download version, see this rant on Metal-Fi. This way we get a near CD quality unprotected version with the full MQA quality controlled by DRM. I have to admit I had not really thought about this. I had already dismissed the download version as useless (see my previous post on MQA), which is clearly the case from a customer point of view. However from the distributor point of view the download version is quite interesting.

A long time ago (at least 9 years) I had a discussion with Charles Altmann, designer of many small audio devices and the Tera Player. He asked why you need a brick wall or low-pass filter in front of the DAC. The reason was he made a sampler without it and he found it sounding better than ones with such a filter. I explained about the folding of the frequencies around the Nyquist frequency (see also my previous post on MQA). He stated that in the real world the energy of the frequencies above the Nyquist frequency is very low, especially on the 96kHz sample rate he was using.

What do we know about MQA so far:

  • According to Mr. Stuart MQA is transparent from the analog sound in the studio to the analog sound in your home.
  • Also equipment that does not support MQA decoding will benefit from the improvement made by the MQA encoding.
  • Recording (sampling) without analog brick wall filter sounds better than with such filters.
  • Pre and post echo’s caused by analog or digital filtering are distorting the sound and are therefor not transparent.
  • Avoiding pre and post echo’s is better than removing them using an apodizing filter.

So the most simple conclusion is that MQA is just recording/mastering/sampling without any filters at the ADC at 176.4/192 kHz and then performs the processing as described in the previous post: down sample to 88.2/96 kHz using a 6 tap lowpass filter that only attenuates the highest frequencies that would be mapped to the lowest frequencies. All other frequencies are retained by folding around the Nyquist frequency. The resulting 88.2/96 kHz signal is reduced to 16 bits using dithering and noise shaping. This is the streaming version of MQA. This version can then be lossless compressed to a 44.1/48 24 bits download version. The HDCD style encoding of information in the LSB seems to be only used for authentication purposes, and possibly as DRM.

As mentioned above, in real life there will be not much (if any) energy in the higher frequencies, so the above method can be simplified to just recording to 88.2/96 kHz 24 bits without any filters in the ADC stage. Then apply the same dithering and noise shaping to lower the bit-depth to 16 bits. This will make lossless encoding much more efficient as the lower 8 bits of a 24 bits signal is pure noise and can therefor not be lossless compressed. And due to the noise shaping the 16 bits should be comparable to a 21 bits signal, so there should be no loss on sound quality.

This entry was posted in Audio Processing, Meridian. Bookmark the permalink.

5 Responses to MQA – Further explained

  1. David L. Rick says:

    Having read Stuart & Craven’s AES convention paper #9178, and being already familiar some of the cited references, it appears to me that an important aspect of MQA is that it’s using a non-Nyquist sampling and reconstruction scheme. The short spline filters don’t prevent aliasing. They actually allow it to happen in a way that is reversible under the assumption of a certain smoothness constraints on the original analog signal. Meridian calls this “folding”, partly because ” aliasing” would scare the audiophiles, and partly because it’s not conventional aliasing if it can be unambiguously reversed. A mathematician would call it subspace projection. Read the work of Michael Unser for details.

    As for the calibration based on original ADC characteristics, I’m pretty sure the reason is to choose an apodizing filter that best corrects the time smear of the original brickwall anti-alias filter. A complementary spline reconstruction filter is needed at playback time, so this information must be transmitted as a non-Nyquist audible watermark.

  2. David L. Rick says:

    Darn auto-complete! The last phrase should read: …so this information must be transmitted as a non-audible watermark.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s