MQA - think origami paper folding skills. It’s not upsampling it’s unfolding!!! Very smart indeed.
1 Like
One extremely important feature I use in Roon that I cannot live without is volume leveling. I can’t stand reaching the volume knob or the remote to adjust volume for each track. No matter what Roon does with MQA, I will not be able to use that feature with MQA DACs. So a DAC supporting MQA is useless to me. That makes MQA useless to me…
4 Likes
Roon supporting MQA is contributing to its potential success. It is unfortunate. If MQA takes hold, the music companies will take hold of us and never let go. More expensive, more copy protected, less flexible. Not at first, but mark my words.
It’s like the drug dealer that gives the first taste for free.
Hmmh, “perceptual”? The actual number of bits varies depending on how much ultrasonics there are to encode. IIRC, I have MQA content where the resolution falls to 14 or 15 bits and it actually varies from track to track even within same album. The encoder needs to decide how many bits it needs to “steal” from the base band (0 - 22.05/24 kHz) to encode the upper (22.05/24 - 44.1/48 kHz) band.
So far those filters seem to be the same for all DACs…
There are 16 upsampling filters, and the source content indicates which one of the filters should be used. Those could certainly be run in software too. If you don’t mind using really leaky filters in first place…
1 Like
Nice try. dCS implemented their own filter.
Back too your software.
If this is the case, probably Roon will be the first one to take advantage of this. Let’s see how it works out in the end.
Even dCS implements their own filter, the main purpose here is to improve the impulse response called ‘de-blurring. By doing this, the filter construction will allow some ultrasonics noise to leak through plus aliasing distortion. It is a compromise one needs to take.
The ‘perceptual’ in term of resolution, in the case 17bit/88.2/96kHz MQA core is always referred to effective noise-shaped and dithering resolution. One good example is 16bit/44.1kHz CD such as Sony’s Super Bit Mapping (SBM) which can yield near 20bit effective resolution in the audio range with a penalty of increased ultrasonic noise. The actual resolution (determined by ADC) is much lower than this.
Having such upsampling filters is not a new thing, it has been around for more than a decade…
I also added my own custom filters to HQPlayer for upsampling MQA content. But they can be also used to upsample normal hires content.
As I said, it varies depending on properties of the content. It is not constant.
No, it can increase resolution in midrange, with increase of noise in top of the audible range. 0 - 22.05 kHz frequency range of a CD doesn’t contain ultrasonic frequencies, so the noise cannot be moved there… Total amount of noise/resolution in audible band doesn’t change, it is just distributed differently in the frequency-band.
1 Like
It is upsampling.
When turning a 4X rate file or 8X rate PCM file into an MQA file, the MQA process ignores/deletes any information in the file above 48khz. So the maximum true sample rate of MQA files is 96k.
After the first unfold, an MQA file is turned into a 4X or 8X file through upsampling. It can only get to those super hires rates from upsampling, since the MQA file doesn’t include the original information from the hi-res file that was over 48k.
You can call it origami, but it is still fake hi-res. Not essentially any different than if I took a 24/96 or 24.88 file and upsampled it to 192k or 176k.
It’s fine if you like the sound of the MQA filters. But please don’t support the lie that the files are actually true 4X or 8X master files.
2 Likes
As far as is known till know, all MQA DACs use the same set of filters. The filters aren’t unique to the DAC. Apparently the only exception to this is, as you said, dCS, who wrote their own filters for MQA together with MQA.
Acc’d to dCS, making their devices MQA compatible and writing their own filters cost them thousands of hours of programming time.
It is very doubtful many - or even any - other companies will do this. Pretty much everybody will use the standard set of MQA filters.
So much for the MQA claim that the unfolding on the DAC side is unique to the DAC.
MQA does the first stage called decoding, essential unpack undecoded 24/44.1k/48k to 17bit/88.2/96kHz (MQA core), the process of unpacking is lossy.
The second stage of ‘rendering’ is essentially up-sampling with de-blurring filters. The function of up-sampling here is to restore sampling rate to match the original recorded sampling rate. This is only done if the original recorded sample rate is greater than the MQA core sampling. In this case no additional information is retrieved beyond the MQA core sampling rate.
So we are saying the same thing. For 4x and 8X files it is an upsampled 2X file.
This is interesting mass test of what random listeners prefer: http://archimago.blogspot.ca/2017/09/mqa-core-vs-hi-res-blind-test-part-i.html?m=1
Very detailed.
The ‘2x file’ is always an MQA core at 88.2/96k from undecoded 44.1/48k where information is retrieved (unfold). The rest 4x (176.4/192k) and 8x (352.8/384k) is just up-sampled.
My understanding is MQA splits into two bandpass, the first bandpass (LF) from 0-22.05k/24k and the second (HF) is 22.05k/24k-44.1k/48k. Most of bits are used to code in first bandpass which is the audio range while fewer bits are used to code in the second bandpass which contains very few information in the ultrasonic range. The approximate bit distribution is estimated around ~14 to 16 bit at first bandpass and ~8 bit at 22.05k/24k-44.1k/48k. It uses bit allocation algorithms; the amount of bit code in the first and second bandpass will vary according to amount of musical contents but the combined bit distribution in both bandpass will never exceed 24 bit. This is why undecoded MQA is always ‘24 bit data’. In the area where there’s no information in the second bandpass, then all the bit can be allocate in the first bandpass, which MQA claims that it can deliver up to ‘24 bit resolution’.
When both bandpass are combined, you get bandwidth from 0-44.1k/48k, then the output sample have to double to 88.2k/96k (MQA core). See the diagram below:
Added: to make things more easier to understand in the diagram, the first bandpass (LF) split at ‘17 bit’ (0-16bit from LSB) the second bandpass (HF) splits at ‘18 bit’ (18-23bit from MSB). Combined bits in both bandpass is 24 bit.
As for the sampling frequencies at MQA core (88.2k/96kHz) it is probably derived like this. Since both bandpass have a same bandwidth of 22.05k/24k:
A. First bandpass 0-22.05k/24k, bandwidth is 0-22.05k/24k, only need to sample at 44.1k/48k
B. Second bandpass 22.05k/24k-44.1k/48k, bandwidth is also 22.05k/24k, supposedly the sample should starts at 22.05k/24k to 44.1k/48k, to prevent aliasing should be at least 88.2k/96k right? I suspect it uses Non Shannon-Nyquist sampling, using instead 44.1k/48kHz to sample, there will be aliasing but the information at this bandpass is very Low, so I guessed aliasing is minimised.
Combined both bandpass you will get bandwidth from 0-44.1k/48k with multipled sample of 88.2k/96k…
This is getting interesting, any comments are welcomed…
1 Like
I am the PO of this thread. Thank you all for you valuable info concerning MQA, even that some of the discussions areba little bit off-topic.
Indeed interesting reading!
kr
Kim
Theoretically MQA can behave like a lossless 24 bit/44.1k/48k PCM, provided there’s no information in the ultrasonic range (22.05k/24k to 44.1k/48k) where all the remaining bit in the second bandpass can be allocated to the first bandpass; thus achieve the claim ‘24bit resolution’
However, Hi-Res contents will contain some information above 22.05k/24k, so the MQA cannot achieve ‘24 bit resolution’ as claimed. However, if the content is just 24 bit/44.1k/48kHz, then MQA will behave like one after get decoded (resolution only).
I can conclude MQA is a partial lossless encoding system, smart enough to give lossless at very ‘unlikely condition’ but in practice it is always fall short.
I haven’t come to any conclusions yet. I probably won’t until Roon reveals its hand. At the moment our choices are purchasing (or ripping) high definition files, streaming at CD quality or MQA. If MQA doesn’t measure up subjectively I think it will die a death. But it will still be one of a number of choices either way. The gloom, doom and despondency I read here and elsewhere is unsupported by any real world evidence to me. People in the industry are hedging their bets but that doesn’t mean there is a conspiracy to take our freedom away.
We should just take your word for it or do you have reasons?
Personally I really like MQA.
2 Likes
Nobody is ever going to force anyone to use MQA. What product, streaming or otherwise, studio or otherwise, is going to do this?