Here's something I wrote in response to Jim Austin three years ago regarding high-frequency content and "post-Shannon" sampling:
Very well, let's suppose there is something that matters above 40 kHz. What does MQA do with it? Starting with a recording at 192 kHz or higher, some unknown processing is applied, then the signal is downsampled to 96 kHz using a rather weak anti-aliasing filter. We know this because looking at recordings with some distinct content above 48 kHz (and these are rare indeed), faint alias products are recognisable in the lower frequencies of the decoded MQA file. The attenuation appears to be around 50 dB, but this is a very rough estimate.
The 96 kHz signal then undergoes band splitting, the top half compressed and encoded into the low 8 bits of the final stream. This step actually seems to work quite well in that the decoded output is pretty close to the input, at least for typical music and within the target precision. However, as clever as it may be, this scheme is wholly unnecessary. Standard methods, such as FLAC, perform equally well. As Xivero have demonstrated, the efficiency of FLAC can be further improved by preprocessing the input to remove non-information-bearing noise in the lowest bits. Needless to say, this process is not entirely lossless with respect to the input, but then neither is MQA. The Xivero method is also superior in that the output is a fully compliant FLAC file playable on any existing device without firmware updates or additional software. Of course, there are no royalties for Bob either.
Then comes the so-called rendering stage. As revealed by my reverse engineering, this consists of nothing but textbook FIR upsampling followed by shaped dither, usually at 16 bits. That last part is especially interesting. The images of the low frequencies left by the leaky upsampling filters, which is where any useful content must reside, are to a large extent buried under random noise.
To recap, whatever smidgen of useful signal identified by MQA in the high frequencies has, by the time it reaches the DAC, been attenuated, aliased ("folded" in MQA newspeak) into the much stronger low frequencies, compressed, uncompressed, imaged ("unfolded") back to the high range along with the mirrored spectrum of the (still much stronger) low frequencies, and finally drowned in random dither noise. "Post-Shannon" or not, nothing can survive this mangling and still be recognisable, let alone useful. If I'm wrong, show me the maths.