Bandwidth of MQA
This question was posed by some online commentators. We are giving a precis because the question has the wrong frame of reference.
Questions:
a) MQA has publicly stated that the resolution of the baseband will depend on the spectral content of the music being encoded/decoded.
b) By referring to widely respected Caltech Professor James Boyk's landmark paper on the high-frequency content of various musical instruments, "There's Life Above 20 kiloHertz! A Survey of Musical Instrument Spectra to 102.4kHz", we can easily see that all musical instruments have harmonic content above 20kHz. The amount of content varies according to the instrument and how it is played . . .
b) The Ravel string quartet used for all of MQA's presentations have only a maximum of 0.04% of their music energy beyond 20kHz . . . and in fact some of MQA's promotional material points out that the Ravel string quartet chosen has section of pizzicato playing, further reducing the average musical energy above 20kHz . . . In contrast, much symphonic classical music and nearly all jazz, pop, rock music employs cymbals.
c) Compared to a violin in MQA's string quartet, a crash cymbal has 40% of its energy above 20kHz, with harmonics extend past the measurement limits of Prof. Boyk's equipment. This not a small difference. It is a factor of 1000x = 60dB. Yet we are given no information whatsoever how the high frequency content of the cymbals used in much music affect the storage space required for MQA's "folding" process. This oversight is especially curious as Prof. Boyk's paper was even used as a reference in the original Journal of the AES paper written by Stuart and Craven.
d) Instead can only speculate as to what happens. We know that when a Ravel string quartet is used as the source material that there are only 16 or 17 bits of resolution left for the baseband audio after the "folding" process is complete. When the high frequency energy is increased by a factor of 60dB, how much resolution is left for the baseband audio? Presumably less than 16, but how much less? 14 bits? 12 bits?
Answers:
a) Boyk's contribution has indeed been both inspiring and useful and served as a point of reference for over two decades.[27]
b) This question seems to suggest that we have deliberately chosen an example that is deficient in high-frequency content and that is not the case. The Ravel quartet was indeed used as the main example for the AES 2014 paper but system testing and demonstrations have used music of all genres—plus wideband recordings of harpsichords and cymbals, as shown in Figure 17.
c) Once again we see the questioner confusing measures; miles vs miles/hour. Boyk's paper tabled peak SPL and percentage of power above 20kHz for a variety of instruments. What actually matters is the peak spectral level in the music recordings we are trying to encode. That does not mean that high frequencies are unimportant, far from it, but the capability of an instrument in isolation doesn't necessarily inform us what the chosen microphones or mixing desk will do (footnote 8). In the next section we show some analysis of recordings and explain how the hierarchical variants of MQA preserve the signal.
d) Speculation is indeed just speculation. MQA is not embarrassed by cymbals or any other material. The encoded signal continues to fit into 24 bits and the noise introduced into the decoded signal remains below that of the original recording in all cases so far examined.
In very rare cases, active music energy will remain above the noise floor even at frequencies exceeding 48kHz. To reproduce accurately the signal waveform of such content, MQA can be operated at a higher transmission rate—see the table at the end of these answers.
However the ear's abilities as a spectral analyser extend only up to 18kHz and it has proved much more important to preserve the temporal structure (envelope) of the signal at high ultrasonic frequencies than it is to preserve the frequency structure.
MQA at the standard transmission rate (44 or 48kHz) does preserve the envelope and standard MQA continues to deliver an improvement in perceived instrumental detail even on a rogue recording such as the harpsichord (see Figure 17) where the original frequencies extend well beyond the 48kHz fold.
Footnote 8: We have been fully cognizant of the power spectrum of a cymbal crash, having introduced it (to the misery of many would-be lossless compression designers) as a test signal in the technical competition which selected MLP (Dolby TruHD) for DVD-Audio and BluRay. The waveform produced by a cymbal is chaotic and is statistically more like a wideband noise that tonal signal and is therefore difficult to compress.
Footnote 8: We have been fully cognizant of the power spectrum of a cymbal crash, having introduced it (to the misery of many would-be lossless compression designers) as a test signal in the technical competition which selected MLP (Dolby TruHD) for DVD-Audio and BluRay. The waveform produced by a cymbal is chaotic and is statistically more like a wideband noise that tonal signal and is therefore difficult to compress.
