It would appear that there is a particular audiophile DAC with a cult following that gets rave reviews and costs over $2000, and is based on a non-audio DAC chip.
Why would they do that? Well, I think it is so they can run it “NOS” (not New Old Stock, but “non-oversampled”) and add their own “proprietary” filtering – plus it’s different from what the hoi polloi uses so it must be better. But, it would appear that someone has found a glitch, literally.
I am no expert, but I think that because this chip is a non-audio DAC, the output comes directly from a R-2R ladder, or similar. Small capacitive charges are transferred whenever the ladder switches operate, and sometimes the switches don’t all operate at the same speed. This means there is a glitch at the output whenever the DAC value changes, and it is worst when all the switches operate simultaneously i.e. when the most significant bit changes – around the mid range in other words (hmm…). Presumably there are other significant glitches at multiples of 1/4 full scale and 1/8 full scale too.
Low pass filtering the output can reduce the amplitude of the glitch at the expense of increasing the settling time. There are better techniques using a further piece of circuitry (a sample-and-hold) but, apparently, for the designers this was regarded as unacceptable for some reason (why?), and at audio frequencies still wouldn’t be as good as a typical $1 audio DAC in a mobile phone.
The evidence is all in the DAC chip’s data sheet:
I don’t know whether the glitch energy scales with the the VREF (i.e. the full scale signal amplitude), but this glitch is huge compared to the smallest signals that we might generate with the DAC.
An owner of this product now thinks he is hearing a certain harshness in the sound, and seems to have found that when reproducing a sine wave at -90dBFS, the output of the $2000 DAC contains significant glitches at the zero crossings. It would be interesting to know if there are detectable glitches at 1/4 and 1/8 full scale, too. This could be the phenomenon shown in the data sheet, or a by-product of whatever mechanism is being used, unsuccessfully, to suppress the glitches – they are rumoured to be using a combination of two DAC chips. Scrutiny of other reviews and measurements of the device seems to reveal distortion and noise figures that suggest something strange is going on – apparently.
An aspect of integrated circuit DACs is that because they are very small and constructed on a single chip, they have fantastic performance relative to themselves i.e. they remain monotonic and linear at all times. However, their absolute gain and offset may drift slightly with temperature. These temperature coefficients vary from chip to chip and can even be positive for one chip and negative for another (this appears to be the case for this particular DAC chip according to the data sheet). This means that any attempt to blend the outputs of two DAC chips externally using a combination of scaling, offsetting, inverting, mixing and interleaving would be most unlikely to succeed down at the lowest levels.
If these suppositions are correct, then this product is a great example of where the basic engineering of a basic product appears to have been sacrificed in the interests of just making something ‘different’ and supposedly ‘simpler’ – although as usual it ends up being more complicated.
[Last edited 04/05/16]