Great topic, Brad.It would indeed be good if some actual designers could weigh in with details of what is really going on in some of the decent DACs. Unfortunately this may violate their sacred oaths of non-disclosureOne could also piece things together from patents, although in that case the art of the patent is to protect without actually giving away how to do something!
I would like to point out that the manufacturers of these various chips should support their product with the detailed information about which you justifiably inquire. Texas Instruments has an Engineer to Engineer site with a forum dedicated to high precision converters and another forum dedicated to high speed converters. High speed converters typically run in the MHz range, and high precision may be anything 12-bit or more (I'm a little fuzzy on their definition of precision).
My most recent design is not audio, but is built around a 14-bit DAC capable of 125 MHz. It has two current outputs in a balanced or complementary configuration. It is a beautifully symmetric design where each bit diverts a weighted current source to either the positive or inverted current output. In fact, the circuit performs better with balanced outputs because the current sources always have "somewhere to go." An interesting detail about this chip is that is has a pin where a resistor is used to set the maximum current scale from 2 mA to 20 mA. The most important detail is that the data sheet explains quite well about where the minimum distortion is achieved. The engineers at TI were even working on SPICE simulations for design engineers, despite the difficulties in modeling a chip with digital to analog conversions going on inside. Anyway, I was a little unclear about the distortion characteristics, so I wrote to them on the forum and received excellent support from TI on this subject. The chip wants to see no more than 1.25 V on its current outputs, but the ideal is 0.5 V or less.
I wanted to avoid adding a negative supply to my board, and I considered using a current mirror, but ended up with a precision dual op-amp I-V instead. I still managed to dial in the minimal distortion for this particular chip.
In any case, I suggest that it should not be black magic to extract the optimum performance from a DAC unless the manufacturer is failing at their job of supporting their product. You might find it difficult as a DIY type to get serious support, but the E2E forum at TI welcomes all skill levels, students, and non-commercial engineers equally.
Off topic, but: TI is intelligent, nay even wise, in allowing that democracy. A manufacturer who has vacillated in this regard, for example, is IR. The wise insight: you never know with whom you are dealing. If you filter via email extension (non- .coms excluded for example) or worse, you are doing two things at least: (1) P*ssing some potential future customer off, and (2) excluding someone who may not be in an official employee-employer relationship with an already-very-real customer, but who may have substantial influence on part selection.
Having said that, I sympathize with the people who are attempting to reduce the intrusions on the time of the folks who really are doing the creative work at the particular company. And as well, engineers love to talk about what they are doing, and often can't resist revealing things proprietary. Widlar used to hold court at some of the silicon valley watering holes and chat up a storm, but National knew he was usually a few steps ahead of what he was blabbing about, so didn't get too upset about it.
That sounds like a very clever DAC design. At 14 bits it's entering the realm of where self-heating becomes important, but at least one can determine the tradeoffs with that programming pin.
Brad
Having said that, I sympathize with the people who are attempting to reduce the intrusions on the time of the folks who really are doing the creative work at the particular company. And as well, engineers love to talk about what they are doing, and often can't resist revealing things proprietary. Widlar used to hold court at some of the silicon valley watering holes and chat up a storm, but National knew he was usually a few steps ahead of what he was blabbing about, so didn't get too upset about it.
That sounds like a very clever DAC design. At 14 bits it's entering the realm of where self-heating becomes important, but at least one can determine the tradeoffs with that programming pin.
Brad
Yes, now that I've worked with the THS5671, I really wish I could find a 24-bit version. It's just so elegant. I do wonder how difficult it would be to produce some sort of FPGA hybrid implementation of the same idea with more bit depth. Even a 19-bit or 20-bit version would probably be fine. Perhaps the scaling of the current sources, where each must produce a power of 2 different than its neighbor, reaches the limits of precision of such circuits.
From the other perspective, looking at some of the other DAC chips in this thread, I'm really disappointed to find that some only have a single current output (instead of a balanced pair).
Hi,
I find the support of TI usually quite well, but with regard to the Q of output impedances of their converteres they totally failed. At TI-Germany no one was able to tell me the data, and direct contact to TI-America resulted in the same. This Q was also asked by more than one person in the E2E-Forum and the results were disappointing. From ignoring the Q to "Sorry, we don't have the value of the output impedance available".
Either TI keeps the figures a well kept secret, or the support sucks in this regard.
As it seems all the Audio converters feature rather lowish values of output impedance with the sigma-deltas a bit higher than R2Rs. PCM1792/1794 could be 2k2||22pF and featuring protection diodes, the AD1865 1k7, the ESS-converters just a couple of hundreds of ohms.
jauu
Calvin
I find the support of TI usually quite well, but with regard to the Q of output impedances of their converteres they totally failed. At TI-Germany no one was able to tell me the data, and direct contact to TI-America resulted in the same. This Q was also asked by more than one person in the E2E-Forum and the results were disappointing. From ignoring the Q to "Sorry, we don't have the value of the output impedance available".
Either TI keeps the figures a well kept secret, or the support sucks in this regard.
As it seems all the Audio converters feature rather lowish values of output impedance with the sigma-deltas a bit higher than R2Rs. PCM1792/1794 could be 2k2||22pF and featuring protection diodes, the AD1865 1k7, the ESS-converters just a couple of hundreds of ohms.
jauu
Calvin
The PCM1704 is a Burr Brown part. Perhaps the engineering knowledge beyond the data sheet and the fab specs have been lost after Texas Instruments acquired Burr Brown.
Maybe someone could travel to Tucson hoping to interview ex-employees to create an oral history of these 'old' parts.
On the IV conversion link that tiefbass posted, there are some remarks about the PCM63 and some rough results based on an experiemental attempt to measure the output Z. It's rather pessimistic, including the difficulty of prying information out of the manufacturer and the design challenge of producing a low-enough input Z in the I-V (such that the code-driven errors arising are anywhere near as low as to not spoil the linearity, if indeed it is an R-2R network to the outside world). And yet, at the end of the piece the author remarks that a passive resistor termination still seems to sound the best!
Thanks for that Joachim. This was easier for me to access: http://myweb.tiscali.co.uk/g8hqp/audio/TDA1543IV.html (but that may well be my unfamiliarity with the site file system). It's probably the same reference anyway.
Note that Dave does describe the 1543 output as something much closer to a current generator in the classical sense --- a transistor collector loaded with a current source --- and remarks that it can work with a passive termination whereas other converters will tend to have excessive distortion.
Brad
Note that Dave does describe the 1543 output as something much closer to a current generator in the classical sense --- a transistor collector loaded with a current source --- and remarks that it can work with a passive termination whereas other converters will tend to have excessive distortion.
Brad
Looks to me as if the FETs are merely steering straightforward constant currents from the bipolars into or away from the R-2R network. To the extent that the output Z's of the bipolars are sufficiently high, relative to the R's of the network, the output current pin should have a relatively constant impedance with code changes.
Brad
Brad
The ES9018 gang are never satisfied. : )
What do they want (not much) :
1) Ground referenced split rails
2) Built-in Vref circuit
3) No cap in signal path (not even in current mode)
4) Same circuit for 2 to 8 channels
5) Input impedance < 1R
6) Output impedance < 100R
Do I have a design for that ?
You bet.
The usual disclaimer :
a) I have not built, so build at own risk.
b) I have not even simulated, so no idea of performance.
c) The circuit incorporates ideas from Calvin & Joachim, so no PCB from us unless they explicitly agree.
Critical note :
1) Thermal coupling of transistors within each red box is absolutely essential.
2) I would at least make sure they are all hfe matched to much better than 5%.
I do hope someone will build this or something similar and compare to a SEN / CEN.
And please let us know whether the complexity results in better performances sonically.
Patrick
.
What do they want (not much) :
1) Ground referenced split rails
2) Built-in Vref circuit
3) No cap in signal path (not even in current mode)
4) Same circuit for 2 to 8 channels
5) Input impedance < 1R
6) Output impedance < 100R
Do I have a design for that ?
You bet.
The usual disclaimer :
a) I have not built, so build at own risk.
b) I have not even simulated, so no idea of performance.
c) The circuit incorporates ideas from Calvin & Joachim, so no PCB from us unless they explicitly agree.
Critical note :
1) Thermal coupling of transistors within each red box is absolutely essential.
2) I would at least make sure they are all hfe matched to much better than 5%.
I do hope someone will build this or something similar and compare to a SEN / CEN.
And please let us know whether the complexity results in better performances sonically.
Patrick
.
