12-Jan-2016, 11:04
Hi again, Sleach.
It seems MQA is not _just_ what was written above.
In addition to the above tricks, it actually does employ a novel technology: it resamples the data (or samples it anew if used during recording) focusing on timing of the samples, somewhat reminding of DSD. What happens is that it claims to retain more of the timing information, by convolving the signal first with a triangle function during sampling, and then interpolating during decoding part. As this adds some new noise, they claim that resulting data fits into a 900kbit/sec 96/24 stream. Some of the ideas were previously used in various papers on non-sinc kernels.
The claim to "lossless" reproduction is thus murky - it changes the signal irreversibly, but the "losses" are hard to quantify.
DSD signal even at standard sampling rate retains much better timing information, and thus MQA cannot compete with DSD128 and higher.
Moreover, DSD has a _very similar_ approach/tradeoff between sampling frequency and bit-depth to MQA.
It seems MQA is not _just_ what was written above.
In addition to the above tricks, it actually does employ a novel technology: it resamples the data (or samples it anew if used during recording) focusing on timing of the samples, somewhat reminding of DSD. What happens is that it claims to retain more of the timing information, by convolving the signal first with a triangle function during sampling, and then interpolating during decoding part. As this adds some new noise, they claim that resulting data fits into a 900kbit/sec 96/24 stream. Some of the ideas were previously used in various papers on non-sinc kernels.
The claim to "lossless" reproduction is thus murky - it changes the signal irreversibly, but the "losses" are hard to quantify.
DSD signal even at standard sampling rate retains much better timing information, and thus MQA cannot compete with DSD128 and higher.
Moreover, DSD has a _very similar_ approach/tradeoff between sampling frequency and bit-depth to MQA.
