Stereophile Series on MQA Technology

[Fokus]

There is not a shred of evidence that this so-called ringing, once it appears outside of the audible range, has any impact on perceived sound quality.

 

It is also true that, unless the original signal contained high-level energy at the filter's cut-off point, the ringing as displayed does not appear in the sampled music signal.

 

Impulse response testing is done in digital audio because it is the fastest and simplest method for revealing the nature of the underlying digital filter(s). That is all.

 

No, that is not all ...

 

Impulse response testing is done in consumer digital audio because it is a handy tool for instilling fear and uncertainty in the minds of a techno-illiterate audience.

 

 

[Fokus]

I actively dislike MQA entirely for objective reasons. There are no emotions involved.

 

As an engineer and a scientist I see it for what it is, and it is to be stopped before it causes the industry more damage.

 

This said, I am also getting bored with this all. I guess that is an emotion.

[Fokus]

6 hours ago, Archimago said:

No... This is not elegant. It's sadly rather ugly.

 

I always found the underlying principle elegant, or clever, as I expressed it on Day One:

 

Look at the actual signal, look at the actual noise floor. Determine how shallow an AA filter you can get away

with, so that any aliasing below 20kHz falls below the noise floor.

 

 

But anything MQA beyond that is an attrocity, and politically so totally wrong ... making the grand sum of it ugly, indeed.

 

[Fokus]

2 hours ago, Archimago said:

as far as we're aware, there are no strong DRM mechanisms in place (other than the hassles of using specific software, upgrading hardware and not accessing the full resolution for doing our own DSP).

 

I disagree. In my book the latter all mean a significant curtailing of my digital rights.

 

[Fokus]

Did anyone catch this gem?

 

Austin writes

 

One of the challenges levied against MQA by its more knowledgeable critics is that ... its sampling method—and the resulting, presumed(*) increase in aliasing—introduce randomness in precisely when those impulses occur..... I synchronized the MQA and non-MQA impulse responses: MQA in the left channel, non-MQA in the right. Over 30 seconds of impulses spaced 0.7ms apart, examined on a microsecond scale, I saw no random offsets—or offsets of any kind—in where MQA's impulses landed.

 

The stimulus file being perfect impulses generated in the 96kHz digital domain ...

 

And this guy is writing a technical investigation that should carry some authority, that is impartial?

 

Either he is lying, or he does not understand sampling. At all.

 

 

 

(* Oh, and 'presumed'??? really?? As if said critics were making it up?)

 

[Fokus]

20 minutes ago, mansr said:

It's anything but clear what that "test" entailed or what it was supposed to show. A file cannot have MQA in one channel only. Besides, there's no reason to expect a cumulative timing error, nor any variation with a repeated input.

 

From what I understand the test impulses were in 96k MQA file format, but they were still ideal dirac pulses, i.e. totally unfiltered, except for the origami split/join. As such the individual pulses are in sync with the sampling grid, and indeed there should be no variation in output timing.

 

But this is contrary to what Austin expected: " I saw no random offsets—or offsets of any kind—in where MQA's impulses landed ". Of course he didn't. He didn't test for it. But now he thinks, and wrote, that he has debunked one item of the criticism.

 

As for the channels: he recorded the MQA and PCM outputs of a DAC in to an ADC, and afterwards put one channel of each file side by side. That should be obvious.

 

[Fokus]

1 hour ago, Norton said:

If you apply DSP after decoding you lose  the MQA license (?) and get < CD quality?

 

You can only apply DSP properly after full decoding, but there is no decoder in existence that allows you to apply DSP to its output.

 

If you apply DSP before full decoding it will break the MQA code and will not allow you to decode.

 

The best one can do today is to take the unfolded digital output of Tidal or of a Node2 and apply DSP to that.

Even this was originally not allowed in the MQA paradigm.

[Fokus]

10 minutes ago, Norton said:

 once fed by Tidal MQA, the idea I then get  <  CD quality is patently absurd.

 

 

In MQA parlance the output of Tidal is not "fully decoded".

 

Once truly fully decoded you no longer have access to the data, hence no DSP.

 

[Fokus]

9 hours ago, PeterSt said:

If someone name Bob Stuart somewhere tells that there's B-spline interpolation "involved" he can be correct.

 

December 2014: BS in search of the minimal AA and AI filter.

 

"Reconstruction can be regarded as the dual of sampling
and approached in a similar way. Thus, it is not
recommended to present the samples as unfiltered Dirac
spikes to subsequent equipment. Even convolving each
spike with a rectangle of width one sample period (which
is equivalent to a zero-order hold) still generates
theoretically infinite slew rates at the transitions.

 

It thus seems that convolution with a triangle function is
the least that is needed to produce a signal that can be
handled satisfactorily. This is equivalent to linear
interpolation between sample values.

If sampling and reconstruction each use a triangular
kernel, then simplistically8 the total impulse response is
a 3rd-order B-spline, of total width four sampling periods.
That is a total width of 42 μs at a sample rate of 96 kHz
and a time from 10% of peak to the peak of 13.2 μs.

Unfortunately, that is not the end of the story, for we also
have to correct a frequency response droop from the 3rdorder
B-spline which, for 96-kHz sampling, amounts to
2.5dB at 20 kHz (or 3 dB if sampling at 88.2 kHz).
To meet a criterion such as 0.1dB for the maximum
acceptable 20-kHz droop we have generally used a
maximally-flat minimum-phase 3rd-order FIR digital
flattening-filter immediately prior to the triangle
convolution in the reconstruction.

The flattening filter increases the total length of the endto-
end impulse response by three sample periods, giving
a total length of seven sample periods. Inevitably, the
impulse response is then no longer a single pulse, there
being a negative downswing, a positive, and another
negative following, as shown in Figure 12 below."

 

It just shows how they initially dreamed up their weak filters.

That splines were used is pretty irrelevant. It could have been

anything, including second-order bagels.

 

[Fokus]

15 hours ago, Galileo365 said:

the mathematics behind that presume a “perfect” brick-wall filter ...something that at least currently cannot be built .... Fourier transforms work well at expressing attributes of the frequency domain but not so much the time domain,

 

Three times nonsense. As Mans advised: perhaps going back to school would be useful.

 

BTW, your monicker seems a tad pretentious ...