MQA technical analysis

[audiventory]

By graphs above, looks like MQA work like 24 bit uncompressed format.

 

But MQA lossless or lossy?

 

I read about a probable "frequency-amplitude response correction", but real implementation of encoder/decoder is unknown.

[audiventory]

That DSP is claimed to improve the way the file sounds, by correcting temporal errors. That is their claim. So words such as "lossy" and "lossless" are a little inappropriate ... MQA is always "lossless", but they believe that delivers a better sound.

 

In my opinion, we have either lossless (digital without distortions) or sound enhancer (that modify digital sound).

 

Many people love vinyl sound. But it is not lossless.

 

What is temporal errors here?

[audiventory]

OK, it is a firmware bug in the Meridian DAC firmware. Played exactly same non-MQA track twice, bit-perfect. Only difference being that between the two playbacks I played MQA version of the track. It seems to leave the noisy MQA rendering process active for the subsequent non-MQA track playbacks. So do not trust much on the listening impressions or such with this DAC when comparing MQA and non-MQA tracks, results depend on which order you play the tracks. LED indications are correct, but the real behavior is not.

 

For both cases, the DAC's leaky upsampling filter is the same. Remember the source is 96 kHz non-MQA track which by definition doesn't have any content above 48 kHz frequency.

 

Before playing any MQA tracks:

[ATTACH=CONFIG]32598[/ATTACH]

 

After playing an MQA track:

[ATTACH=CONFIG]32599[/ATTACH]

 

Exactly same track, exactly same bits going to the DAC in both cases.

 

Oh the fun of doing this research work, debugging Meridian's firmware as we go.

 

May be there is especial initialization of the DAC is need?

 

Such behaviour of DAC with any playback software (in bit perfect mode, of course)?

 

Possibly it is damaged DAC.

[audiventory]

do we care about 30 kHz and up freqs.?

 

if so, why?

 

seems like it is easy to filter to prevent noise injection...(?)

 

It is not so rare case.

 

The ultrasound harmonics may cause audible noise by intermodulations.

 

Of course, need ultrasound filter.

[audiventory]

What should be removed are the digital image frequencies and modulator noise, but those are at frequencies way over 150 kHz spreading to MHz range. What shouldn't be removed is real audio content below 150 kHz or so.

 

What MQA doesn't help at all, are those things above 150 kHz. Instead it generates digital images/aliasing under 150 kHz.

 

Do you don't agree that 30 and 33 kHz harmonics can cause 3 kHz (33 - 30) harmonic in audible spectrum by intermodulations?

[audiventory]

Yes, that can and does happen. That doesn't mean it's impossible to design equipment where it doesn't.

 

If say exactly, impossibly create equipment that don't cause intermodulations. Because any active electronic device is nonlinear.

 

However, different equipment provide different level of intermodulations depending on form of input-output characteristic (level-gain curve).

[audiventory]

Yes, but only in shitty electronics not worth having.

 

Digital pre-fitration can help even for «shitty electronics»

 

I'm much more concerned about for example 330 + 333 kHz or 1030 + 1033 kHz image harmonics.

For example when the digital filter outputs 352.8 kHz rate followed by zero-order-hold. When you play 1 kHz tone, the DAC will output both 351.8 and 353.8 kHz tones and thus intermodulation product of 2 kHz. And this is directly correlated with the input signal. This is further emphasized by the fact that at 350 kHz range the THD of analog sections will be higher, and thus also harmonic distortion products of these frequencies will cause also intermodulation products.

 

I’m agree.

 

With music, the level of 1 kHz tones is also generally much higher than 30 and 33 kHz harmonics.

 

I'm heard not once about harmonics in 30 … 40 kHz range with significant level. Probably it is artefacts of studio apparatus.

[audiventory]

Although it can never substitute good analog reconstruction filters. But removing any content that originates from instruments is clear no-no.

 

Referring to my earlier link and other test I've been doing myself with percussive instruments, there's plenty of stuff going beyond that. For example violin creates wide harmonic spectra, as well as glockenspiel which I've been testing with.

 

For accurate reproduction, every harmonic instrument creates needs to be recorded and reproduced.

 

We don’t know exactly where is edge of accurate reproduction. Spectrum is infinite. So band should be infinite too, theoretically.

 

All spectrum limitations is measurement tools matter. New tools - new edge.

 

Band 100 kHz of apparatus limit spectrum too. Despite harmonics may be above.

 

At the Mansr’s picture the harmonics looks like wrong products, not harmonics of instruments.

 

[ATTACH=CONFIG]32642[/ATTACH]

 

I suppose, these wrong harmonics have much higher level than harmonics of musical signal.

 

As rule, we can hear lower 16 … 20 kHz. We can test it.

 

But where proofs of the edge of reconstruction?

[audiventory]

It is where the harmonics are not measurable anymore from the noise floor.

 

For example for the 2L recordings, harmonics reach to about 56 kHz, that's why I concluded that 60 kHz bandwidth is enough for those. But if you extend it to ~90 kHz of 192 kHz sampling rate or ~80 kHz of 176.4 kHz sampling rate you at least have enough margin.

 

As a result, having the filter fc above the highest harmonic means that the filter doesn't have time domain implications on the signal either...

 

I suspect, you refer to noise floor -120 dB (modern DAC).

 

About 20-30 years ago noise floor -120 dB was fantastic.

 

I can suppose, that 22 ... 25 kHz harmonics was deep into noise floor that time.

 

So 25 kHz could be considered as enought for reconstruction.

 

Let's look to probable future. There may be released DAC with noise floor -200 dB.

 

It allow to see (don’t hear) not only 56...90 kHz harmonics, but 150 kHz too. It is not real figures, of course, but suggested as example only.

 

In this case for reconstruction need filter with cut 150 kHz.

 

Also we can look further: when noise floor will -300 dB, as example.

 

What is edge there? When we must stop?

[audiventory]

Thermal noise puts the limit somewhere around -140 dB at room temperature.

 

-140 dB open new details comparing current -120 dB.

 

But I made "Sci-Fi" assumption. There is opening of new details by lower noise floor matter.

[audiventory]

DSD1024 is coming soon (and already used by some DACs).

 

I'm agree. DSD1024 is nice thing for transferring 100 kHz band. May be more too, but I don't checked yet.

[audiventory]

I have a new nugget of information. MQA places a data stream in bit 8 (i.e. what would be the LSB in 16-bit) of the encoded file. Any claims that undecoded MQA might provide CD quality are thus blatantly false.

 

Last time, me seems, that even "lossless" term now have more wide meaning: "same level of noise"

 

May be I'm wrong. But it is my impression.

[audiventory]

Well, most recordings, as I name this, have a nice tinnitus (square wave) on 15khz... may would be nice to see what MQA does with them, while I have no MQA gear..

 

In the DXD high frequency noise should be removed before playback or coding. So there will not issues for coding, I suppose.

[audiventory]

For the next experiment, I replaced the top 15 bits of the MQA file with a 1 kHz sine wave (TPDF-dithered at 15 bits) and decoded the file. This is the resulting spectrum (input blue, output red):

 

[ATTACH=CONFIG]33014[/ATTACH]

 

Blue signal in 15 bit resolution?

[audiventory]

Yes.

 

If there 15 bit, why noise level -140 dB?

 

Must be about -90 ... -100 dB.

[audiventory]

You are confusing total noise with noise spectral density.

 

Spectrum 15 bit looks like 23...24 bit resolution.

[audiventory]

If you integrate across the frequency range, you get the level you're expecting. It's why doubling the sample rate lowers the noise floor equivalently to adding a bit. Same total noise spread across a wider frequency range.

 

Each expanding range 2 times give difference 6 dB for level (voltage) spectrum and 3 dB for power spectrum.

 

For sample rate 22 kHz (44 kHz sample rate) there -90 dB.

 

At the picture we see -140 dB. Difference is 50 dB=-90+140.

 

50 dB / 6 dB is about 8 times.

 

22 kHz * 8 times = 176 kHz band (352 kHz sample rate).

 

At the picture I see input band 22 ... 24 kHz, not 176 kHz.

 

1. What is analyzis software shown in the picture?

 

2. What is window applied?

[audiventory]

Do you know what "integration" is?

 

I suppose, the integration can't decrease level noise to 40 dB.

 

What is level (in dB) of the signal (by oscillogramm) in LFSU scale?

[audiventory]

Suppose a noise floor of -140 dB at 48 kHz sample rate. Multiply that by the 24 kHz bandwidth (i.e. integrate the constant level) and convert back to dB scale:

 

10 * log10(10^(-140/10) * 24000) = -96 dB

 

 

With the rectangle window we get exactly the level expected according the usual 6 dB per bit formula. The others lower the level around 6 dB, so the -140 dB level seen above with 15-bit dither is precisely where it should be.

 

1. Why you suppose noise floor -140 db?

 

2. -140/10 - whats here -140 and 10?

[audiventory]

Do I really need to explain to you how the dB unit works?

 

Why -140 dB? Why not -110? Why not -200?

 

If you have time, could you show how you get the formula?

[audiventory]

As I already said, it was an example. The specific value has no significance.

 

 

 

I thought I already did, but lets take it step by step, again using -140 dB/Hz as the example noise floor:

 

1. Convert dB to linear units: -140 dB = 10 ^ (-140 / 10) = 1e-14

2. Multiply by the bandwidth in Hz: 1e-14 * 24000 = 2.4e-10

3. Convert linear to dB: 10 * log10(2.4e-10) = -96.2 dB

 

A noise floor of -140 dB/Hz over a 24 kHz bandwidth thus corresponds to a total noise level of -96.2 dB.

 

When noted noise floor as 6 dB * [bit number] there meant formula dB by level:

 

Level_dB=20*log10([absolute level]/[reference level]).

 

You used formula by power dB

 

Power_dB=10*log10([absolute power]/[reference power]).

 

If 10 replace to 20, where need, we get -52 dB in goal 3.

[audiventory]

The 20 comes from the power being the square of the amplitude. Aren't you supposed to know all this?

 

Yes. However, you use absolute values. At spectrum analyzer level dB are showed.

 

Me seems, need begin from other end.

 

Need calculate rounding error energy and distribute it across band.

 

As result you get tone with amplitude lower 0 dB (0 dB minus energy of rounding error).

 

The rounding error energy distributed by full band.

 

Wider band - lower energy per Hz.

 

However, analizer use FFT. So need distribute noise by [FFT length]/2 points, not per Hz.

[audiventory]

The calculation is reversible. It doesn't really matter if you work in amplitude or power as long as to stick to one.

 

You are right.

 

With an FFT you get frequency bins covering a fixed range each. The output is typically normalised to get values that are not dependent on the transform size.

 

At spectrum we view power in each point (bin?) of FFT. More points, lesser energy in each one, because sum of quantization errors for all points is constant.

 

Lesser energy for each point - lesser noise floor.

 

Example:

 

Let suggest, total energy spectrum: 100 = 90 (signal) - 10 (errors)

 

Signal take 1 point. 10 (errors) distributed by rest points.

 

If rest 255 point: energy per point is 10/255.

 

If rest points is 1023: energy per point is 10/1023.

 

Etc.

 

It in the each point we will see lesser noise for more FFT length.

 

Can we now please get back to MQA and away from elementary calculus?

 

Ok. We can don't discuss more about the analyzer. But its results of measurements so far from my experience and a bit theory what I know.

I suspect, need check the analyzer’s scaling before using.

[audiventory]

That's why the output is usually scaled according to the FFT length.

 

Yes, scaled. But I wrote about spectrum energy distribution. It is relative. Signal (90% energy) anyway take 1 point and 10% energy (errors) distributed by rest points.

[audiventory]

Only if the signal falls exactly in the centre of one of the bins. This is one reason windowing is used.

 

If it (we talk about pure sine only) fall between two points it distributed between these points and noise distributed between rest again.

 

Of course, if we have too low point number (16 as example) there more noise will mixed with the sine in its points.

 

I'd like, check the above mentioned source and decoded signals in an other analyzer, that show more traditional results.

 

I try check my results different ways for decreasing of error probability.