Does it not mean that MQA uses sampling with a shaped kernel over more than the width of one nominal Nyqvist sample?
No, they just run decimation (sample rate reduction) using their filter (kernel) at mastering stage to reduce the source to 88.2/96 kHz rate for encoding. Then they use shaped high level dither to hide distortion from the leaky filter and trying to maintain dynamic range at lower frequencies despite only few bits. They use the entire top octave for filter roll-off (and aliasing), because they think that frequencies above 20 kHz are not useful so those can be sacrificed to keep ringing of the filter to minimum.
Depends on the mastering engineer what tool he uses for that. Could be iZotope SRC and MBIT dither or something else like MQA.
Then they also use their filter (kernel) for upsampling the encoded data at playback stage.
If you prefer to have exactly same filter design paradigm at both ADC and DAC side, look into Ayreās QA9/QB9ā¦
I was not referring to Bob Stuartās āPerspectiveā but to the much more technical articleāreference 1 in this Perspective.
I agree with everything you said Jussi, but the answear to Viruskillers question depends on how you interprete the filter-kernel and is a bit of a tricky question (in particular as we donāt know what they are using)? (Please object if you disagree.)
What I mean is that Nyqvist sampling is about sampling a bandwidth limited signal (with respect to the used sampling rate). Hence if you are sampling a bandwidth limit signal you can view the sampling kernel as āideal diracā e.g. the amplitude in a point hence with zero width. But on the other hand if you are sampling a non-bandwidth limited signal (with respect to your sampling rate) the āideal brick wall filter kernelā has infinite width.
This is from the Stuart/Craven JAES paper:
The information loss can be avoided by using an integration kernel in the form of a triangle or dual ramp that spans two sample periods, as shown in Figure 11 (lower). By comparing the values of Sample 0 and Sample 1, both the area and the position of the transient can now be unambiguously determined.
These possibilities are extended in [23] wherein it is shown that by using a higher-order B-spline kernel,7 it is possible to determine separately the intensities and positions of two or more pulses even if they lie within the same sampling period!
Footnote 7 says āThe triangle being considered a B-Spline of order 1.ā The reference [23] is to Dragotti, P.L., Vetterli, M., Blu, T., āSampling Signals With Finite Rate of Innovationā, IEEE Trans.Sig. Proc. Vol. 50, No. 6, pp. 1417ā1428, (May 2007)
1 Like
I think this is a perfect example of what Jussi was upset about earlier. They quote something that is true but does not give any real support for MQA. So they bring in a reference to make it appear that they have scientific support for MQA, but these references does not have any strong relevance.
For example, while it is true that you can sample using triangle kernels and reconstruct the position of dirac pulses, this is irrelevant for Music. Consider that if you used the proposed reconstruction method you would get a sequence of dirac pulses independent of the input signal, and no one can hear nor want to listen to dirac pulses.
Speak for yourself! 
I interpret it in the way they say on the paperā¦
This is about conversion of bandwidth limited signal of sampling rate X to bandwidth limited signal of sampling rate Y. As long as you have discrete samples (IOW, you have some sampling rate) in digital domain you always have bandwidth limited signalā¦
P.S. Sure you can also run apodizing 1:1 conversion filters too, also in certain unnamed software player, in such case the rate changes from same to same and you have just new construction of samples.
Half of the paper is like politicians speaking, lot of words with close to zero information contentā¦
1 Like
In case no one else spotted it Bob Stuartās answers to āthe big oneā (aka the massive list of questions from CA readers) are in.
Give yourself some time - itās seriously long.
http://www.computeraudiophile.com/content/694-comprehensive-q-mqa-s-bob-stuart/
One thing of note that I took away (as a mere āaverage listenerā) is that you will be able to combine Room Correction. Q65.
It seemed pretty open and frank to my mind, but Iām sure much discussion will come of it, particularly from those that have an advanced understanding of the fieldā¦ā¦.
New (or working) URL:
http://www.computeraudiophile.com/content/694-comprehensive-q-mqa-s-bob-stuart/
(Thanks and apologies, have updated the link in mine too)
Room correction seems to be allowed only if it is built into AVR processor or a DAC that has also licensed MQA decoderā¦ So it needs to be specifically approved by MQA first. Nothing yet that would allow any room correction to sit between MQA decoder and the DAC. Except of course if you are fine with degraded performance undecoded ācompatibilityā stream.
I still find it disturbing they refuse to consider the difference between āauthorizedā DACs and degraded ācompatibilityā stream being equivalent of DRM. Exactly same as SACD layer vs RedBook layer on the same disc. To me, thatās DRMā¦
They also refused to answer any of the actually interesting critical questions.
What questions have you asked?
What questions has Bob Stuart refuse to answer???
I understand the Mytek Brooklyn firmware with MQA was just released. Any feedback from Brooklyn owners on sound quality?
I donāt understand the slurping in this TAS article. Isnāt MQA supposed to be the same as a studio master high-res file when itās āunpackedā? If so, how can he state:
How can two bit streams that are identical sound different, unless all we are talking about here is different after-the-fact mastering, which of course would sound different.