Sound quality does not matter.
By this I mean it is not the most important thing - its the ability to get emotional satisfaction from music (musicality) that is primary.
And I get to "enjoy" lots of über DAC's at shows - and for me - and of course this is highly personal - they fail big time as regards musicality. I get more musical satisfaction from little old Mojo than any of these über DAC's do.
Moreover, I am absolutely convinced that Mojo (ignore Dave for the time being) will beat any other DAC at any price point when it comes down to enjoying music. Why am I so convinced? Because there are sound technical and objective reasons why this would be the case. Let us consider the three most important things that is important for a DAC/amplifier:
1. Timing - if you spend any time researching psycho-acoustics you will appreciate that timing is the most important parameter that the brain uses to process the data from the ears. It is used for location, timbre, and being able to perceive the starting and stopping of notes. So when I talk about timing problems, what do I mean? Now I am NOT talking about phase errors, or timing differences with frequency response - these are linear errors, and are inconsequential - the brain is used to dealing with these kinds of problems. What I am talking about is when the timing of transients varies either with amplitude (a transient will have a different time depending on whether it is small amplitude or a large amplitude signal) or when the timing of a transient depends upon when it is sampled by the ADC. So if a transient crosses through zero half way between samples, differing interpolation filters will not recreate the timing of the transient accurately, and this sampling dependent uncertainty is highly audible. So how big is the problem of timing? Looking at all the many listening tests I have done, my contention is that a DAC/amp must be accurate down to tens of nS in that surprisingly small timing errors are very audible.
So what DAC's have minimal timing problems? Let's look at the interpolation filter first, something that is in all DAC's (yes even NOS DAC's have an interpolation filter). Now the job of the interpolation filter is to reconstruct the timing of transients and to remove the HF images of the signal that extends to infinite frequencies that is due to the sampling process. In short, it is part of the process that takes the discontinuous digital data and converts it back into a continuous analogue signal. The better the interpolation filter, the more accurate this is done (closer to the original analogue signal in the ADC before it was sampled). Now the accuracy of the filter in terms of how well transient timing is reconstructed depends upon the filter tap length, and the type of filter. NOS filters are by far the worst, with timing errors of up to a hundred uS. A NOS filter (this is a bad term, all DAC's oversample) is actually a zero order hold interpolation filter, where the oversample rate is determined by the DAC that is used. Because the timing errors are so huge, they sound very soft (when you get timing errors, the brain can't deal with the data, so it can't actually perceive the starting and stopping of notes - and if you can't hear the starting of a note, it sounds soft or out of focus). Now some people like this; but it is clearly artificial and is categorically not transparent. The usual FIR filter is a half band interpolation filter, with a tens of taps. This filter is better than zero order hold (NOS) for timing, but still has massive errors, as there is considerable measurable sampling image errors. These filters return the original sample unchanged, and they are cheap to implement which is why they are used for 99% of the time. Apodizing filters can offer better performance, but they still have substantial timing errors.
So how can you reconstruct the timing of transients perfectly? If you look at sampling theory, it has been proven that a Whittaker-Shannon interpolation filter will return a bandwidth limited sampled signal absolutely perfectly; there will be no timing or amplitude errors at all. But to have an ideal interpolation filter you need a sinc impulse response and many many taps; to absolutely guarantee just 16 bit performance under all circumstances the coefficients need to be smaller than 16 bits - and this happens at about 1 million taps for a 16 FS filter. And every listening test I have ever done gives the same conclusion - more taps, better and more transparent sound quality. Even with Dave at 164,000 taps, we have not reached the subjective limit yet.
So onto the next timing problem - amplitude errors. Now R2R DAC's have slow FET's to switch the resistors in and out, and these themselves cause their own glitch issues. Moreover, they are very slow - you can only go to 16FS max - so that means the timing is limited to only 1.3 uS. DSD dac's have very large amplitude timing errors too - a small signal has a much bigger delay than a large signal. Currently, the only technique that has the smallest amplitude related timing error is pulse array, as it runs 5 bits at 104 MHz - many times faster than any other DAC topology.
So to solve the timing issues you absolutely need extremely fast DAC's, and very large tap lengths. Even Mojo easily beats all other non Chord DAC's in this regard.
2. Small signal resolution. Clearly small signals are vital - if you can't hear the tiny details, then it no longer becomes a believable performance. Small signals also are used for depth perception, something that I am personally very interested in. Go to a cathedral and listen to an organ at 100 m away; it sounds 100 m away - but reproduced audio is severely depth truncated. Now small signal linearity is measured using fundamental linearity tests - you measure the amplitude of a -60db, set that value as a reference, then measure at -120 dB say. It should be exactly -120.000 dB, but a real DAC won't be. Delta sigma or DSD will actually attenuate the level, R2R will have random errors due to resistor tolerance problems. When it comes to depth perception, there is something extremely strange - any small signal amplitude error (no matter how small) affects depth perception. That's why Dave has 350 dB performance noise shaping, as increasing the performance of the noise shaper gave much better depth performance. Now R2R DAC's have easily measurable small signal errors; DSD is only -120dB accurate noise shaping; conventional DAC noise shapers are about -140dB; Mojo is at -200 dB, and Dave takes the record at -350 dB. I have published FFT's showing Dave's noise shaper performance.
Now this issue is more complex than this; you also need simple analogue stages to further improve performance, you additionally you need a distortion performance that gets better as the signal get smaller.
3. Noise floor modulation. This is another can of worms, as DAC's have a multitude of problems from different areas for this. Noise floor modulation is subjectively very important; it is when the noise floor changes with signal level. All conventional DAC's have large amounts of noise floor modulation, and it is my contention that it is very audible - it adds a hardness or grain to the sound. Even the tiniest amount affects smoothness and refinement. With Mojo, we have no measurable noise floor modulation - and similarly Dave but with an even lower noise floor. I have not seen any other DAC come close to this performance.
This gives you a brief flavour of the issues involved in producing a truly transparent DAC; and for me it's only by having a truly transparent DAC that one can get musicality. But of course some people have very different tastes, and respond differently to music. Some people like a particular sound; I have done some listening sessions when I drew the complete opposite opinion to somebody else. Some people like distortion; others like an overly soft warm sound. Whatever floats your boat I guess.
Rob