Beyond the outright copy of the PCB design (and therefore circuit design) of someone else's work, I have to say there's a more pragmatic reason to avoid this R-2R DAC.
R-2R is simple, cheap, and I'm all for that - They're great for up to 10-12 bit resolution, and maybe slightly more (up to about 14bits).
Of course - they'll 'operate' up to 24 or 32 (or 128) bits - whatever you can build (and sell?) will work, but the resolution suffers , greatly, beyond about 10 bits for 0.1% SMT resistors, or beyond about 14 bits for well-matched, trimmed, 0.005%/ single-digit ppm temperature drift (typically thru-hole) resistors.
The problem is that when a digital input rolls over, for example, from 01111111 to 10000000, (and there are many, many, similar roll-overs in the digital domain), the tolerance on the MSB must be smaller than about 1/2 the LSB -otherwise you get non-monotonicity (relatively large local non-linearity) which is inherently bad for audio (not to mention the transients that need to be filtered out).
As you add bits (up to +/-24 bits in this product, which is actually 25 bits), the required tolerance is impossibly tight. Even with trimming (whether laser or network, etc.), the tempco of the resistors also results in non-linearities due to slight temperature changes and gradients (unavoidable with dynamically changing signals).
Of course, this assumes that one can hear non-linearities in the lowest levels of noise & high resolution - some can, others can't, and some others can but don't let it ruin a good time (music is usually supposed to be a good time?).
And if you choose to go R-2R, then at least use trimmed networks - they'll match better than even high-dollar discrete resistors (and WAY better than discrete SMT resistors).
Analog Devices' AD5791 is an R-2R based DAC, that i believe is used in some audio DAC's, and it benefits tremendously from highly-trimmed R-2R ladder in silicon - not discrete - which provides excellent TCR matching, among other benefits. But it's actually a 14 bit R-2R ladder with specially trimmed upper 6 bits (as 63 individual stages - not part of the ladder used in the lower 14 bits). This is an advantage that a discrete solution does not have.
However - this DAC is aimed largely at applications where DC performance & adjustability are key - not linearity or non-monotonicity in reproducing complex & subtle audio - things like programmable power supplies and test equipment.
In my opinion, although the AD5791 (and probably a couple other similar integrated & trimmed R-2R DACs) sounds wonderful - it still doesn't live up to the expectations of '20 bit' in terms of audio DAC performance - it's probably no better than 16 bit. But the area where the loss occurs just isn't glaringly obvious to most listeners - it only falls short in the finest resolved signal levels that honestly, most people can't discern - but will be happy to convince themselves that it's superior to their old set - they've invested their money and emotions in it - because they've bought into the poetry and zen-like qualities offered. "It simply MUST be better - I'm sure I can hear it!"
But my objection stands - don't bother paying for a R-2R DAC that claims to be any better than about 14bits resolution (and be very skeptical of claims of better than 12 bits...)
Almost any other DAC technology is better for audio.
-David
R-2R is simple, cheap, and I'm all for that - They're great for up to 10-12 bit resolution, and maybe slightly more (up to about 14bits).
Of course - they'll 'operate' up to 24 or 32 (or 128) bits - whatever you can build (and sell?) will work, but the resolution suffers , greatly, beyond about 10 bits for 0.1% SMT resistors, or beyond about 14 bits for well-matched, trimmed, 0.005%/ single-digit ppm temperature drift (typically thru-hole) resistors.
The problem is that when a digital input rolls over, for example, from 01111111 to 10000000, (and there are many, many, similar roll-overs in the digital domain), the tolerance on the MSB must be smaller than about 1/2 the LSB -otherwise you get non-monotonicity (relatively large local non-linearity) which is inherently bad for audio (not to mention the transients that need to be filtered out). As you add bits (up to +/-24 bits in this product, which is actually 25 bits), the required tolerance is impossibly tight. Even with trimming (whether laser or network, etc.), the tempco of the resistors also results in non-linearities due to slight temperature changes and gradients (unavoidable with dynamically changing signals).
Of course, this assumes that one can hear non-linearities in the lowest levels of noise & high resolution - some can, others can't, and some others can but don't let it ruin a good time (music is usually supposed to be a good time?). And if you choose to go R-2R, then at least use trimmed networks - they'll match better than even high-dollar discrete resistors (and WAY better than discrete SMT resistors).
Analog Devices' AD5791 is an R-2R based DAC, that i believe is used in some audio DAC's, and it benefits tremendously from highly-trimmed R-2R ladder in silicon - not discrete - which provides excellent TCR matching, among other benefits. But it's actually a 14 bit R-2R ladder with specially trimmed upper 6 bits (as 63 individual stages - not part of the ladder used in the lower 14 bits). This is an advantage that a discrete solution does not have. However - this DAC is aimed largely at applications where DC performance & adjustability are key - not linearity or non-monotonicity in reproducing complex & subtle audio - things like programmable power supplies and test equipment. In my opinion, although the AD5791 (and probably a couple other similar integrated & trimmed R-2R DACs) sounds wonderful - it still doesn't live up to the expectations of '20 bit' in terms of audio DAC performance - it's probably no better than 16 bit. But the area where the loss occurs just isn't glaringly obvious to most listeners - it only falls short in the finest resolved signal levels that honestly, most people can't discern - but will be happy to convince themselves that it's superior to their old set - they've invested their money and emotions in it - because they've bought into the poetry and zen-like qualities offered. "It simply MUST be better - I'm sure I can hear it!"
But my objection stands - don't bother paying for a R-2R DAC that claims to be any better than about 14bits resolution (and be very skeptical of claims of better than 12 bits...)
Almost any other DAC technology is better for audio.
-David