MQA is Vaporware

[yamamoto2002]

Are MQA encoded music files able to be decoded properly when official MQA decoder is disappeared? I heard it is protected by DRM and there is no legal way to decode properly by the third party even after all related patents are expired

[yamamoto2002]

39 minutes ago, The Computer Audiophile said:

The decoder is proprietary and unavailable outside of licensed products. The upsampling can be done after the decoding. 

Oh that's unfortunate. Some higher frequency part of the music will be lost forever at some point of time. it is damnatio memoriae of 21st century

[yamamoto2002]

On 4/8/2023 at 12:35 AM, Jud said:

They’re really not able to be decoded ‘properly,’ in the sense of being faithful to the original, even with the official decoder, because encoding uses lossy compression and truncation of bits (to substitute bits that tell the decoder what to do); then that’s followed by filtering as part of the decoder that leaves aliasing/images. (I don’t know what the filtering looks like on the ADC side, but if MQA adheres to the same philosophy there it should behave similarly.) So once a file is encoded with the MQA process, some information is irretrievably lost. This is why I think the great pity of MQA while it lasted was not so much the bad productions, but the good carefully produced recordings that could have been yet better.

I don't know MQA algorithm. I understand it cannot be lossless compression because 2:1 or 4:1 (I read somewhere they compressed 176.4kHz PCM onto Compact Disc) fixed rate lossless compression is technically not possible, data with maximum Shannon entropy cannot be compressed further so all the lossless compression should be variable bit rate. Still, with some of the high frequency info can be stored and recovered by embedding a hint information onto least significant bit of PCM signal. While without proper decoding, the extra hint info becomes very non-musical noise and signal quality becomes worse than plain linear PCM and it is not good (at least in psychologically, may be it can be real audible noise issue in some corner cases). Similar technology HDCD, which squeezes extra 6dB ? of dynamic range using perceptual unevenness nature of linear PCM, is relatively benign because it is not DRM protected and it can be decoded properly forever.

[yamamoto2002]

Open source free software cannot implement MQA therefore I don't have interest to it, but, today I found this thread is VERY interesting😄, cannot stop reading.

[yamamoto2002]

There is an article about MQA appeared on local magazine 無線と実験 July 2023 pp.116～117. General info about MQA technology and the report of current company situations.

 

I think MQA is fleeting, passing, transient technology using traditional analytic method. There should be completely different audio codec in the near future, this time with deep learning autoencoder, much higher compression ratio and perceptually lossless, because it extract features (instrument, timings, pitch, expression, ...) and store / transfer,  and reconstructs sound from them and trained model parameters

[yamamoto2002]

1 hour ago, manisandher said:

Chord's filters are very sharp, but they don't attenuate fully before Nyquist. Something I just can't understand.

 

I think that low-pass filter is implemented using "half-band FIR filter" technique in order to significantly reduce computing cost. Very sharp brickwall filter can be designed using extremely large FIR taps thanks to reduced computing cost, but one drawback is attenuation level of Nyquist frequency is fixed to -6dB.

 

This is from Richard G. Lyons "Understanding Digital Signal Processing" 3rd ed. p.208

 

 

[yamamoto2002]

39 minutes ago, manisandher said:

Why have Chord chosen to start the attenuation at 22kHz, and not 21.5kHz, say?

As I said, it is impossible to change cutoff frequency with half-band filter, cutoff freq is fixed

 

39 minutes ago, manisandher said:

what possible advantage could such a sharp filter have over the Motu's everso-slightly-slower filter? To justify a >2000% price difference perhaps? Total madness.

 

I think it is rather technical demonstration to show what is possible with modern (～10 years old) FPGA DSP. Price is a bit steep for me  but there is a market for those edgy things

 

And when FIR filter tap is, say 262145, coefficient of the edge tap becomes smaller than PCM 19bit LSB and windowing becomes not necessary. In the past I tested with 65537 taps FIR and found 65537 is not sufficient to omit windowing