Reference

Tutorial: Music Origami

We refer to the process by which MQA reduces high sampling rate signals to Fs = 44.1kHz or 48kHz as 'music origami', after the Japanese art of paper folding. Figures 14 to 16 show how it works for a 192kHz source.

Although the process is depicted here in two dimensions, it is actually a three-dimensional construct.

The first 'fold' (Figure 21) reduces the transmission rate from 192kHz to 96kHz, and the second (Figure 22) from 96kHz to 48kHz. The folding process is not filtering and the inherent sample rate and bit-depth remain. In the transport, as we fold, each resulting sample conveys more information. In the MQA lexicon this first folding process is known as 'encapsulation' and ideally it is done at a frequency above the point 'P'; where this is not the case we have techniques and options (described later).

The process is hierarchically scalable, so if the source were, for example, a 352.8kHz file then we use three folds to reach the final transmission rate of 44.1kHz. Similarly, if the source were only 96kHz, then we start with the lossless process of Figure 22. MQA is also hierarchically scalable so that each type of fold can be used one octave higher to enable double-speed transmission options (see later).

The second fold, illustrated by Figures 14 and 15, is able to be lossless because, as illustrated by the noise floor of the original analogue signal (blue trace), much of the available dynamic range with 24-bit encoding is occupied by random noise.

So the second fold process uses buried data techniques to reversibly hide the signal information from above 24kHz within the noise floor below 24kHz.

Principally because of the combination of environmental noise and microphone self-noise (plus tape noise with analogue masters), very few recordings achieve let alone exceed 16-bit dynamic range. Add to this the fact that we can hear signals within noise only to about 10dB below the noise level (see olive curve in Figure 21) and it follows that bits 19 to 24 carry no useful information.

What these diagrams don't convey is the range of novel processes that MQA uses 'under the hood' to achieve this folding and the lossless unfolding subsequently applied by the decoder.

For example, novel sampling kernels are used which suit environmental and music signal statistics, which are adapted for human listeners and provide tight time resolution, while fractional-bit lossless coding increases resolution.[2]

MQA also exploits its end-to-end architecture to enable subtractive dither techniques in the decoder to ensure that, unlike other formats, there is zero noise modulation. Unlike lossy codecs, MQA never ever intrudes upon the music signal, plus it maintains precise and constant characteristics throughout an entire musical work.

Playback:
If an MQA decoder is present then these two folding processes are undone to restore, in this case, a 24-bit/192kHz output with the end-to-end frequency and impulse responses already shown in Figure 1. Furthermore, the decoder reconstructs the signals fed to the D/A converter with bit-accuracy, giving the identical result heard in the mastering studio; it authenticates and indicates this result.

If no decoder is present then the file replays at 44.1kHz or 48kHz, providing backwards compatibility.

If the transmission path can only pass 16 (rather than 24) bits, then an MQA decoder can reconstruct the baseband (area A in Figure 10) along with a lossy version of B (and C), which sounds very close to the high-res original and much better than CD replay.

Similarly, if the replay device is not capable of the highest sampling rate, the unfolding process can be stopped part-way through and the decoder optimises the temporal and frequency response.