Ultimate SQ DAC

Inspired by this thread https://www.diyaudio.com/forums/digit...-new-post.html
I've been giving a little thought for how to move beyond the 'Ozone' to a DAC able to handle more than 16bit inputs and up to 96kHz input rate. The target being 20bits and 120dB SNR (non-A weighted). The OP in that thread preferred a more marketable 'ultimate' DAC (with 192k and potentially 384k capability, along with DSD) - aims which to my way of seeing clearly conflict with a DAC having any pretentions to ultimacy.

The simplest solution - building on the digital part I already have - would be to add more TDA1545As in parallel and with a Cortex M4 direct the data to the respective chips. What's unclear though is how low the noise will go when the extra chips are added. From the DS, an A-weighted SNR is quoted of 101dB (typ) with 2mA, however this is a static noise (code = 0) and hence may well not translate to the noise achieved with bits toggling. When paralleling DACs the noise will only reduce relative to the signal if uncorrelated between them. Thus the only approach would be to suck it and see.

Paralleling many DACs achieves more bits of resolution however their accuracy will depend on how closely gain matched the DACs are. The 1545 can be tweaked for gain but it'll be susceptible to drift so this is another unknown.

The worrying aspect of using TDA1545 is its speed, and hence glitch. Already it sounds 'greyer' when running at 88k2 than in NOS so I'm inclined to have various other DACs up against it for evaluation. Here are my three top contenders for a next generation DAC -

Harris/Intersil HI5741 - 14bit, 100MSPS
National/TI DAC14135 - 14bit, 135MSPS
ADI AD9744 - 14bit, 210MSPS

These I've picked out because of their low glitch specification (5pV.s and below) and reasonable-looking MTPR figures. There appear to be no contenders at the 16bit level so this means oversampling is called for even on RBCD material. I've not yet looked in detail at their noise performance in an audio bandwidth, this might disqualify them quite early on.

Have I missed out any contenders ? (TDA1541A won't run fast enough to get to 20bits with oversampling so would need paralleling). Incidentally, I did also look at TI's DAC904 but although it has 14bit input its measurements look more like those of a 12bit part.

Update : over on WBF (which I can access when my ISP doesn't give me my banned IP) Elberoth is waxing lyrical about his new DAC acquisition - https://www.whatsbestforum.com/showthread.php?12023-Trinity-DAC&p=238265&viewfull=1#post238265.
Apart from Audio Note this Trinity DAC could be the most expensive commercial DAC - when I last checked the price was about 45,000 Euro but that was for a mono DAC. But it might have been the price for the older 'GTE Trinity' which TNT-Audio reviewed in 2006. From looking at the current website, it seems this version of the Trinity exists in a single box (as opposed to 3) Clearly an opportunity for confusion!

If you read Elberoth's account you'll begin to understand some tidbits to justify the stratospheric price tag - they're doing in-house testing on PCM1704s to get the best ones. He says they select for lowest THD. He also details some history of 'ultimate DACs' there with a reference to Krell using PCM64. I've not come across this part number before - a bit of Googling revealed no datasheet for it but a reference somewhere that it was a consumer version of the DAC729, an industrial hybrid DAC. That part does have a datasheet and they're available on Taobao for around 600rmb a pop.

I mention the Trinity because I'm not impressed by the value-for-money here. Running PCM1704 at such a high rate (8X OS, 352kHz) is only going to exacerbate the glitching. Running lots of DACs in parallel, even here where time-delayed does help average out the bit-weight errors but these aren't the limitation on the PCM1704's subjective sound quality. I also have another beef - time staggering DACs is not the same as oversampling. Simply put, you can't make a 64X OS DAC out of 8 time-staggered 8X OS DACs : that doesn't stack up. To have 64X OS each DAC does need to update at the 64X rate. But at least the glitching isn't going to be worse with this scheme than running at 8X OS and probably with the right math, the OOB images can be pushed up in frequency - however due to the previous DAC's output 'hanging around' too long (only updating at 8X, not 64X), it'll force the math to create a larger step size for the subsequent DAC to compensate. Meaning a longer settling time - more noise. Swings and roundabouts as to whether time staggering DACs does buy anything useful - but certainly marketing-wise its a great story for those who believe faster=better in DACs

I'm fascinated that Elberoth has such a good technical grasp of DAC design issues - perhaps from talking to Rob Watts who he mentions he's friendly with. However his treatment of the analog output filter is rather simplistic. He says:

For years the analog reconstruction filter was (and still is) a real headache for any DAC designer, since it always introduces distortions.

Its not an issue for all DAC designers - plenty of NOS DACs don't use a reconstruction filter at all. Perhaps because they don't like the sound - but then SQ isn't related to 'distortions' except in the very general sense of things not in the original signal.

The problem is not dissimilar to the passive crossover in the speaker. Many people reported huge gains in transparency by getting rid of a passive crossover and going active.

Here he's assuming that the filter in question is a passive one - yet who apart from Zanden uses a passive filter after the DAC chip? Maybe Naim? I wonder who these people who gained transparency from going active were - did they replace their passive XO filters with opamps at line-level? Certainly no transparency gain there! But were they to use a PLLXO and multiple amps, they'd gain transparency but not from the loss of their PHLXO, rather from making life easier for their amps. Bi-ampers experience the same without throwing out their PHLXOs.

So from what Elberoth says - that there's no reconstruction filter at all at the output it must mean that image frequencies above 150kHz are being sent unattenuated to the preamp or poweramp. Choice of downstream electronics is therefore critical. A lot of kit has gentle (1st order RC) filtering on their inputs to reduce RFI, however its too gentle to prevent intermod from these high ultrasonics.

Pick your poison - if you believe distortions due to presence of an output filter are the bugbear, then you'll go the Trinity route - a wideband output, because all DACs produce OOB stuff, there's no way around that. The distortions of a passive output filter are primarily in phase response as the THD can be made arbritrarily low with the right choice of passive components. AP has an all-passive filter they recommend for testing class D amps with - no sign that distortion from that is significant.

However the price you pay for no phase aberrations is increased intermod in the downstream electronics. I'd guess that valves are more resistant to this than SS - with SS I'd say you're going to hear the intermod (loss of timbral colours, poorer dynamic constrasts) more than you'll hear the phase distortion. That's why I go for the passive output filter.