MQA's 'impulse response' is indeed extremely short, and indeed uniquely so, just as MQA boasts (even rivaling air itself). In fact, MQA's intrinsic impulse response is uniquely so very short (as shown in the Stuart/Craven AES paper) that it extends only 5us to at most 8us (at its -10 dB point) into each sample interval, thus does not even span the 20us sample interval (48 kHz sample rate). But this technical feature, instead of truly being MQA's greatest technical asset, is actually the complete opposite, its worst technical liability, which causes its worst sonic defect. MQA's very short 'impulse response' means that its coefficient function, which is the guiding template for the playback reconstruction filter, cannot gather the information required for reconstruction, not even over the entire span of one sampling period. Thus, MQA's reconstructed time domain signal waveform cannot effectively even span one sampling period. This might not be a severe problem when connecting the dots for a low frequency repeating sine wave. But it is deadly when trying to reconstruct a single transient attack that does not repeat (that's what makes it transient) and that contains high frequencies (that's what makes it an attack).
Consider a transient attack's sharp and narrow single peak, from a Bob Dylan guitar pick, just as we heard it in MQA's own MQA vs. PCM A-B. Suppose the signal waveform of this sharp, narrow peak happens to be initially sampled unluckily (just as most high frequency peaks are likewise unluckily sampled). The sample dots will be way down on the flanks of this transient attack peak, and the high peak itself will have occurred somewhere near the center of the sampling interval, so its high peak amplitude will not be represented directly by any sample dots at all.
How does MQA, with its very short, nearly 'ideal' impulse response, reconstruct this unsampled transient attack peak to play it back correctly, in the time domain that MQA boasts of such superiority in? It doesn't. Instead, MQA, with its extremely short 'impulse response' that's even shorter than a sample interval, can only see the sample dot information very near each dot, thus only at and from each one dot itself. Thus, MQA can reconstruct, for its time domain playback, only two isolated low amplitude narrow spikes, representing the two sample dots that sampled the low skirts of this transient attack peak, but unluckily entirely missed the high amplitude sharp peak itself. What then does MQA do with this high amplitude sharp transient attack peak that was there in the original signal of the guitar pick? MQA can never even see this peak, so it never reconstructs it, thereby utterly erasing it (or at the very least eviscerating the true hard peak strength of this guitar pick's transient attack).
www.headphonesty.com
