MQA first unfold in Roon? MQA? [Delivered in 1.5]

[MusicFidelity (Guy)]

Encoding is non Shannon sampling; rendering is upsampling.

Encoding then decoding.

[MusicFidelity (Guy)]

Here it is…
A’ (0-24k) is 2x sampling at 48k (Shannon-Nyquist)
‘B’ (24-48k) is 1x sampling at 48k (Non-Shannon-Nyquist)
‘C’ (48k-96k) is -2x under-sampling at 48k (Non-Shannon-Nyquist)

[crenca]

So the claim MQA is making can be summarized as

1. encoding is 384(or less) > folded 24/48 MQA
2. decoding is MQA > 17(at best)/96, (with loss at the 22k border of fold apparently)
3. rendering is upsampling of 17/96 to match the original input (for what reason?) + DAC side MQA filters
   OR
4. rendering is decoding of lossy > 96 information + DAC side MQA filters

[miguelito (Miguel Barrio)]

Guys, distinguish between encoding and rendering. Both/all of you are correct. Encoding is non Shannon sampling; rendering is upsampling.

Exactly.

Non-Shannon sampling is a fancy way to say “I will encode exactly what’s below 48KHz sampling rate to ~15bits resolution, and I will encode the band up to 96KHz with lower resolution”. This is why this is lossy - the info below 48KHz sampling rate is preserved exactly (with 15bit resolution, and exactly in terms of harmonics) and the info above 48KHz and below 96KHz only approximately. The “first unfold” unpacks the 24bit/48KHz data into the two pieces above. The stream is at 24bit/96KHz out of this.

Past 96KHz - aka rendering - is pure upsampling - no real information is “decoded” here.

On the comment that the Dragonfly is limited to 96KHz - this is not a limitation in this case since all first unfolds are 96KHz (or 88KHz when the source file is at 44KHz). The internal DAC in the Dragonfly upsamples to 384KHz.

[MusicFidelity (Guy)]

A thing to note. A pure PCM at 24 bit 192k means it has up to 24 bit resolution all the way to 96kHz bandwidth. In the case of MQA, it only gives up to 15 bit from 0-24k(audio range) and 7 bit from 24 to 96k bandwidth(ultra-sonic range). Non Shannon sampling is always done at 44.1/48k at the ultra-sonic range to further reduce the bandwidth. This is the reason, why undecoded MQA is always at 44.1/48k. Therefore, MQA is technically a lossy compression system.

When decoding happens, all the 3 bandpass, ‘A’+‘B’+‘C’ are combined, the sampling frequency will be 192kHz, e.g.

‘A’ at 48k, 1x multiplier, ‘B’ at 96k, 2x multiplier and ‘C’ at 192k 4x multiplier after 3 bandpass are combined. I think this is not up-sampling but decoding.

[miguelito (Miguel Barrio)]

A thing to note. A pure PCM at 24 bit 192k means it has up to 24 bit resolution all the way to 96kHz bandwidth. In the case of MQA, it only gives up to 15 bit from 0-24k(audio range) and 7 bit from 24 to 96k bandwidth(ultra-sonic range). Non Shannon sampling is always done at 44.1/48k at the ultra-sonic range to further reduce the bandwidth. This is the reason, why undecoded MQA is always at 44.1/48k.

When decoding happens, all the 3 bandpass, ‘A’+‘B’+‘C’ are combined, the sampling frequency will be 192kHz. Therefore, MQA is technically a lossy compression system.

I think we are mixing sample rate and bandwidth here… The “real information” in an MQA encoded file only goes to 96KHz sampling rate / 48KHz audio bandwidth. Anything above this is upsampling.

The plot you show has audio bandwidth in the x-axis - so information between 0-24KHz audio bandwidth is encoded/decoded exactly, information between 24KHz and 48KHz audio bandwidth is approximate (this is what Non-Shannon means), and information between 48KHz and 96KHz is upsampled/guessed using specific upsampling filters. There are 32 of these filters, most of them look remarkably similar. Check the Archimago article I quoted above.

[danny (Danny Dulai)]

As Roon already passes through MQA for the MQA-compatible DAC to unfold, why do we even need Roon to do the first MQA unfold? We’ll need a MQA-capable DAC anyway, with or without Roon doing the first unfold.

There are DACs that can do the render but not the decode. The Audioquest Dragonflies come to mind. They require software to do the decode.

[Paula_Dickerson]

What kind of DAC do you have?

[miguelito (Miguel Barrio)]

As Roon already passes through MQA for the MQA-compatible DAC to unfold, why do we even need Roon to do the first MQA unfold? We’ll need a MQA-capable DAC anyway, with or without Roon doing the first unfold.

Three reasons:

1- For DACs that know nothing about MQA, you can still get the vast majority of the MQA decoding with the first unfold.

2- For DACs that only do rendering (as danny noted) you need a first unfold in software.

3- You cannot do any DSP on an MQA file that is to be fully decoded in the DAC (you’d destroy the encoding). If instead you do the first unfold in Roon, you can then apply DSP (eg room correction) to that high res stream. Audirvana allowed for this ever since it incorporated MQA.

[miguelito (Miguel Barrio)]

I chose my DAC because of its MQA filter and for the specific reason that it doesn’t use USB as the interface. I have no issues with Roon developing these options for those who desire them. I just want Roon to leave the option of passing thru the file untouched from Tidal or on those MQA files I have saved on my network .

They certainly will allow passthrough. Any other choice would be madness.

[thyname]

As Roon already passes through MQA for the MQA-compatible DAC to unfold, why do we even need Roon to do the first MQA unfold? We’ll need a MQA-capable DAC anyway, with or without Roon doing the first unfold.

There are DACs that can do the render but not the decode. The Audioquest Dragonflies come to mind. They require software to do the decode.

How do I know which DAC does the decode?

[miguelito (Miguel Barrio)]

How do I know which DAC does the decode?

Any DAC that says MQA compatible will do at least rendering. Full decoding in the DAC is harder to decipher from the specs - many would just be renderers but not be clear about it (eg iFi).

[andybob (Andrew Cox)]

Original topic for this thread was “Will Roon support MQA first-unfold”?
Answer : Yes.
Additionally, it’ll be done at no extra charge.
Hasn’t this thread run its course?

Yes, but closing MQA threads is like chopping the head off a hydra. People legitimately want to discuss MQA and for every thread we close two more seem to open. So we try to keep it to one or two active threads, even if they depart from the original topic into more general discussions.

[MusicFidelity (Guy)]

There’s a blog that dive into more details into this…

MQA – What is Meridian hiding?

The patent in question for bandpass that carry two separate bandwidth, 0-24k and 24-48k. The amount of bit used to encode is 16 and 8 respectively.

https://worldwide.espacenet.com/publicationDetails/originalDocument?CC=GB&NR=2503110A&KC=A&FT=D&ND=3&date=20131218&DB=EPODOC&locale=en_EP

[wklie (Peter Lie)]

How do I know which DAC does the decode?

Currently, Dragonfly, Berkeley and iFi perform MQA rendering only. Most of the other certified MQA playback equipment perform full MQA decoding.

The capability of Roon MQA core decoding is most useful to users who do not have MQA DAC, or has one of the above MQA renderers.

[miguelito (Miguel Barrio)]

The patent in question for bandpass that carry two separate bandwidth, 0-24k and 24-48k. The amount of bit used to encode is 16 and 8 respectively.

Correct, except there is other DRM and such data embedded, which ends up taking some bandwidth away. There are a multitude of studies around showing that the effective resolution is about 15 / 7 or thereabouts.

Also the post you quote is pretty old, and the “2nd version of MQA” does not exist anymore to my knowledge.

[miguelito (Miguel Barrio)]

The capability of Roon MQA core decoding is most useful to users who do not have MQA DAC, or has one of the above MQA renderers.

Or those with any type of DAC (MQA or not) who would like to do DSP on the stream.

[Ashley_Doane]

Or those with any type of DAC (MQA or not) who would like to do DSP on the stream.

For instance, the Dynaudio Xeo speakers, where the DSP occurs in the speakers. Here, DSP with active crossover is applied to the input PCM signal, which is converted to PWM and amplified. To the extent that DSP is a useful tool for matching speaker and room acoustic properties for the most accurate sound reproduction, best leave the DSP to the speaker engineers and keep the digital signal untouched as long as possible. And delivering the signal is where Roon comes in

[Alastair_Roberts]

There is so much misinformation in this thread. The first unfold to 96kHz is almost lossless and does not use non Shannon sampling. If it did there is no way they could claim its lossless nature. Plus the relevant patent etc does not indicate anything about Non Shannon.

The second unfold to 192 or above does use a non Shannon approach and there are plenty of patents and documents from MQA discussing it.

I can post links later if people are interested. Commuting right now…

[MusicFidelity (Guy)]

There is so much misinformation in this thread. The first unfold to 96kHz is almost lossless and does not use non Shannon sampling. If it did there is no way they could claim its lossless nature. Plus the relevant patent etc does not indicate anything about Non Shannon.

The second unfold to 192 or above does use a non Shannon approach and there are plenty of patents and documents from MQA discussing it.

Not true, MQA uses bit bandpass splitter that split the first 16 bit into 0-24k and remaining 8 bit into 24-48k bandwidth. When all the bits are used to code the information then the effective resolution is less than 16 bit at audio range and 8 bit at ultrasonic range after get unfold at 96kHz. This is already a lossy process.

Since MQA only limit up to 48kHz bandwidth, anything above is not coded. The result anything above that is ‘rendering’ or simply up-sampling to 192/384k.

The only time MQA will behave like a ‘lossless’ is when there’s no ultrasonic range to code; the upper remaining 8 bit can combine with the lower 16 bit to give a ‘24 bit’ in the audio range. This is only valid if the source of recording is PCM 24/48k, not 24/96k.