MQA – What is Meridian hiding?

I have been doing a lot of reading these last few weeks. There is quite some discussion on fora and audio news sites on Meridians latest invention called MQA. But what is MQA exactly and what is Meridian not telling us? This is what we will investigate in this article.

This article is based on info found on fora and audio news site, but mostly on technical descriptions from patents. I have not had the pleasure to listen to MQA myself, so I can not confirm nor refute any claims made on the quality of the sound.

2 versions of MQA

Where do we start? I came across this very interesting screenshot of the system Meridian is using to demo MQA on the CES. Besides the fact that they obviously demo MQA by comparing it to 128kbps MP3, we see that MQA comes in 2 flavours: Download at 44.4/48kHz 24 bits or Streaming at 88.2/96kHz 16 bits. We also notice that the download version is substantially bigger than the streaming version, because it has a higher bitrate.

Actually the 2 versions contain exactly the same data, however the download version is also backward compatible with legacy systems. Using a complex system the 96/16 PCM data is losslessly compressed into the 48/24 PCM data (same goes for 88.2/16 into 44.1/24). The top 16 bits of this data is playable as regular ‘CD quality’. This all follows from the patent named “Doubly Compatible Lossless Audio Bandwidth Extension”.

I remember reading somewhere that development took 5 years, so I guess at that time it seemed necessary to provide a backward compatible download version. To me it now seems a waste of space and CPU power, since the format is inefficient and unnecessarily complex compared to direct lossless encoding. Also nowadays there is no need for the backward compatibility as most or all media players handle 96/16 files just fine. I prefer the streaming version as download version, so there is no need for the download version in my opinion.

Just a PCM stream

So MQA is just a 88.2 or 96kHz PCM stream with 16 bits resolution. According to Meridian and as mentioned in the patent this is enough to satisfy audiophile requirements. In Mr. Stuarts 2004 AES paper “Coding High Quality Digital Audio” he concludes (page 19) that to capture all hearable frequencies we need 58kHz sampling rate with a 14-bit representation with appropriate noise shaping. MQA clearly exceeds this minimum requirement.

However from more recent research it turns out that although humans can not hear frequencies above 20kHz, they are sensitive to timing of sounds to about 10 microseconds. So first you notice the arrival of a sound (quick change in air pressure, a very high frequency) and later on you actual hear what sound it is. To preserve this timing info in the audio signal 96kHz is therefor not enough, we actually need 192kHz. MQA does not seem to take this into account?

Since with very high frequencies it is all about the timing (and amplitude) and not about the actual frequency, Meridian found a way around this. The solution is described in the patent named “Digital Encapsulation of Audio Signals”. In the presentation of MQA Mr. Stuart states that the information of the 192kHz signal is carefully folded into the 96kHz signal. The truth is that this automatically happens if you down sample without a lowpass (or brick wall) filter at the Nyquist frequency (half of the signal frequency). For instance if you down sample from 88.2kHz to 44.1kHz signal, 23kHz will be mapped onto 21kHz, 24kHz onto 20kHz and so on.

With the frequencies of the above example this is considered to be a big problem. However the higher frequencies (above 76kHz) of the 192kHz signal (above 68kHz for a 176.4kHz signal) have far less energy than the lower frequencies (below 20kHz and within human range). So in this case it should be no problem. And to avoid contamination of the lowest 7kHz range (to which the human hearing system is most sensitive), Meridian deploys a small 6 tap FIR filter to attenuate the upper frequencies of the 192kHz signal, since they will be mapped on the lowest frequencies after folding.

Decoding of the 96kHz MQA stream to the original 192kHz signal is simply performed by calculating in-between samples by averaging the 2 neighbouring samples. Besides that a small correction filter is applied to compensate for the filter used at the encoding stage. The idea is that this up sampling process should happen in the DAC or just before the DAC.

So there you have it: MQA is just a 192kHz signal folded into a 96kHz 16 bits PCM stream. Is that all or are we missing something?

The missing link

I have seen claims of people that have heard MQA that it sounds better than a 192/24 version of the same song. Based on all information we gathered above, this is not possible. In the best case it should sound just as good. Also Meridians co-founder Mr. Stuart claims MQA provides a transparent path from the original analog sound at the microphone to the analog sound at your home. We may find a clue in the above mentioned patent on Digital Encapsulation of Audio Signals. The very last claim (#38) states that the filter response is determined in dependence on information received from an encoder. In the notes at the end of the patent is referred to Pacific Microsonics, the creators of HDCD.

So most likely information on the used ADC in the mastering stage is hidden in the PCM stream, probably encoded in the LSB (least significant bit) as done with HDCD. This information is then used in or just before the DAC to compensate for any distortion caused by the ADC. The information is possibly the impulse response of the (lowpass/brickwall) filter used. Meridian was already compensating for this kind of distortion using its apodizing filter. Also possible would be information on the jitter, especially when it is data correlated.

Strangely enough Meridian does not communicate the above information, although it seems the only explanation to achieve better than 192/24 sound quality. It also seems to be the unique selling point of MQA, certainly to audiophiles. But somehow the marketing department thought they should present it as the next generation MP3. And maybe that is just the way it might work to sell it to the mainstream audience.

This entry was posted in Audio Processing, Meridian and tagged . Bookmark the permalink.

6 Responses to MQA – What is Meridian hiding?

  1. Hi. Very pleased to have discovered your blog.

    Reading about this MQA business, I still don’t quite follow why we need this higher resolution timing, and how the folding around Nyquist achieves it.

    I presume that we know that a correctly band-limited sampled PCM’ed signal has near infinite timing resolution – something which some people don’t realise – but it cannot convey any information about the timing of a frequency component outside its bandwidth and, as you mention above, there is this suggestion that our ears can register the arrival time of high frequency components without being able to hear them in the conventional sense. Do you know if this is actually a scientifically-accepted idea, or is it simply a supposition to explain why, anecdotally, people prefer higher bandwidth sampling?

    If it’s an accepted theory, are we saying that the leading edge of a snapping twig sound is detected in our ears using a different sensor than the conventionally-known ones that only go up to 20 kHz? And however it works, that this extra sense detects the timing of sharp transients that contain components at, say 100kHz? If so, how can a conventional low sample rate signal convey the high frequency information to the ear? I presume it can’t.

    So, intuitively, is the idea to distort the shape of the time domain signal in order to give some sense of the inter-sample timing? I could imagine overlaying a high sample rate waveform over the top of a low sample rate version and, by eye, tweaking the low resolution waveform to convey something of what was going on in the high sample rate version, perhaps emphasising or de-emphasising edges here and there. Is this, effectively what the “folding” achieves?

    But even doing that, we still wouldn’t be stimulating the ear’s ‘high frequency sensor’ would we?

    Thanks in advance if you can clarify any of this stuff for me!

  2. Mr Apodizer says:

    Hi. Thanks for your comment.

    As mentioned I tried to sum up the ideas behind MQA, based on things I read on the web. Clearly this is all based on publications of Mr. Stuart and his colleagues (Peter Craven and others). As far as I could determine there is no (or at least no solid) scientifically evidence for these ideas. I agree with you that the leading edge theory might not be the answer why people prefer higher bandwidth sampling.

    As I explained in my other post on MQA, I think it is important to avoid using any (brick wall or lowpass) filters, and thus avoiding pre and post echoes. Avoiding them is better than trying to remove them with an apodizing filter, which will add a (larger) post echo itself. In my experience the post echo is just as bad as the pre echo, maybe even worse, so better have none. The higher frequency makes it easier to sample without a filter, because otherwise to much energy from higher frequencies would fold around the Nyquist frequency and map on a frequency in the human hearable range.

  3. Alex says:

    Geat explanation. I don’t really see any advantages that MQA actually offers, and consider this to be hype and marketing. Also with the right equipment, existing 44 khz PCM sounds fine to me, provided the recording has been (re)-mastered and encoded with care!

    • Mr Apodizer says:

      In my opinion the biggest advantage MQA offers is a countermeasure against the brick wall (or other lowpass) filter that is used in the recording stage. This is done in a more advanced way than the apodizing upsample filter does.

  4. Alex says:

    MQA is LOSSY therefore is destroys the mastered PCM. Why can’t people just use FLAC with 50% lossless compression? This reminds me of the speaker hype with quadraphonic etc (see Steve Martin).

    • Mr Apodizer says:

      Actually the PCM master in itself is also lossy, because it is a finite representation of an infinite phenomenon (an analog signal). MQA reduces the master further back to 88.2/96kHz 16 bits, so that is indeed lossy compared to the original master.

      However, as long as you retain the essence of the signal, there is no problem. For instance, keeping everything in 24 bits is ridiculous. No system will properly reproduce a 24 bits signal. Good systems will do 18 bits (108dB S/N), very good systems 19 bits (114dB S/N) and the best systems will touch 20 bits (120dB S/N). Also, the higher the sample rate, the fewer bits you need. If this would not be true, DSD would never work. Add to this the fact that correct dithering of the signal will increase the perceived resolution, 16 bits at 88.2/96kHz might indeed be enough to store all information our systems can reproduce.

      The real problem of MQA is the DRM that is used to encode this 88.2/96kHz 16 bits signal in a backwards compatible 44.1/48kHz 24 bits signal. Only with a licensed decoder the original signal can be reconstructed. Besides this, the encoded signal compressed with FLAC is 50% larger than the original PCM signal compressed with FLAC. So much for the best way to distribute and store the files.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s