I have spent the last 2h writing the a detailed post on the design of the Trinity DAC for another forum, so I thought I will post it over here as well, as it contains many valuable information.
microstrip said:
Yes, the BurrBrown PCM1704's (now 15 years old, should we call them digital vintage
) are R2R types. It is something that puzzles me - how is it possible to get 24 bit 768 KHz with such excellent measured performance from a R2R switching DAC using current sources?
The word 'vintage' is a relative term. The PCM1704 is still in production. They have been eclipsed specs wise by the PCM1792 DACs, but many manufacturers still prefer it to the newer, low bit DACs and continuue to use them. Not many people realise, that even the top of the range PCM1792 is already a 10 years old design ...
Such an excellent measured performance from 1704 is possible thanks to the special time-staggered arrangement they are used in. The
DACs are not run in parallel, as in most designs (with inputs and outputs interconnected),
but run via a special delay line, where each DAC is fed with a slightly delayed signal. This creates additional interpolation points which are situated linearly between the original sampling points.
According to the manual,
the whole thing in the TRINITY DAC behaves like a quadratic series. Therefore, it means that two D/A converters are needed for 2 ‐ way analog oversampling, 4 for 4 ‐ way oversampling, 8 for 8 ‐ way oversampling, etc. The Trinity DAC uses 8 DACs per channel (16 DACs in total), meaning used arrangement gives an effective oversampling rate of 8x.
Since the Trinity DAC still uses the digital filter, the result is an increase in the effective oversampling rate from 8x to 64x (8x in the digital filter, multiplied by eight by the time-staggered DAC technique).
I know of no other DAC that would use 64x oversampling, as that would require and extremely fast DAC, able to run at 2.8224MHz (PCM1704 tops at 768kHz) and massive computing power. Krell, beeing a Krell, once did it in their Reference 64 DAC using some brute force computing with 4 FPGAs and countless EPROMs. They ended up with a DAC that was huge, required equaly huge and heat generating PSU to power all those FPGAs and eventually didn't live up to its promise, mainly due to the limitations of the DACs used (they had to settle on the only DAC of the era that was fast enough to operate at 64x sampling frequency, or 2.8224MHz - Burr-Brown PCM64; it was an 18-bit device that needed manual MSB trimming and had a fairly high level of glitch in its output, which in turn needed an extra de-glitch circuit in the signal path ...)
Wadia used to do something similar to what Trinity is doing in their 27 DAC - they had used 16x oversampling digital filter and four time-staggered DACs in parallel to get to a 64x rate (this is where I'm not sure how Trinity got their patent granted; most likely Wadia's and Trinity's solutions differ slightly).
MSB has 32x digital filter in their latest Platinium IV series of DACs, so does Chord, but that's it.
Such a high oversampling rate allows the Trinity to
work without the anlog filter. This is unique among tha DAC designs I'm familiar with. For years the analog reconstruction filter was (and still is) a real headache for any DAC designer, since it always introduces distortions. (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. This is also one of the reasons why crossoverless speakers sound so immediate and real.)
But that is not all. For frequencies 176kHz and above, the DAC works not only without the analog filter, but also without the digital one ! It basicly makes it a NOS (non oversampling) DAC, but without all the problems of NOS DACs and their unfiltered outputs !
But it is only one of the many technical features that make the DAC so special. The next one is
DAC selection.
Trinity buys the 1704 DACs from Burr-Brown (TI now) in the highest grade available (K), but it is not good enugh for them. They run a further, in house selection to pick up the best of the best. Not many people realise, that the Texas Instruments specifies the 1704 'K' THD as 0,0008% (typical) and 0,0015% (max). That also means, that among the DACs you will find the very few that are 0,0004% THD.
Trinity buys those DACs in spools of 2000pcs, then measure each and every DAC on a special test jig they designed that allows the DACs to be tested without soldering, and hand picks those that exhibit the lowest THD of the lot. So it is like making your own 'Double Crown' DAC selection (in Philips speak), only in house. (The question remains what they do with the remaining DACs; I doubt they throw them away - most likely they return those to the TI or sell them to other manufacturers).
That strict DAC selection is one of the reasons, the Trinity DAC is so expensive. It is also a major bottleneck for their manufacturing capacity. Dietmar told me that they are olny able to manufacture 3-4 DAC per MONTH. So unless they invest heavily into the selection part of their business, those DACs will always be available in a limited supply.
The third technical reason the Trinity is a superior DAC is the
attention they paid to the clocking. They use two fully customized, hermetic sealed Voltage Controlled Oven Controlled Xrystal Oscillators (VCOCXO), which deliver the 2 master clocks of 44.1kHz*512= 22.579200.0MHz and 48kHz*512= 24.576000.0MHz. They claim that the accuracy of those clocks is 1000x (!) higher than the clocks used by dCS (1ppb or 0.02Hz).
And even though it is debatable if accuracy of the clock has any relation to sound quality in audio (what is important is clock
stability), the phase noise figures must also be extremely low, as they quote jitter levels at just 28 fs (that is femtoseconds, or 0,028 picoseconds !) in 10Hz-10kHz range. That is absolute state of the art teritory.
The forth and reason the DAC performs so well is a well implemented
the USB input (with galvanic isolation). The USB interface allows the data to be clocked from a local, free running, ultra low phase noise clock and use its full potential. DACs that use outboard clocks arrangement (dCS, Esoteric) still have to send the fragile clock signal through the loom of cables, and then through a PLL loop inside the DAC - each of those steps adding a certain levels of jitter. I'm not even mentioning DACs that cannot sync to the transport and have to rely on a clock signal recovered from the SPDIF signal, as that is way too oldschool to be even discussed in the realm of state of the art DACs...
Trinity Solution is similar to what MSB does (ultra low phase noise in close proximity to the DAC chipsets) and is most correct from the technical point of view. And it shows.
