The schematic you showed had BJT switches. They have the same issue as the MOSFET switches: You need to make sure that the resistance of the switches is consistent across the signal swing. Ideally, they should have zero resistance but until we get superconductors that work at room temperature that won't be possible. To avoid distortion you at the very least need switches that show no dependency on the signal voltage.
The DAC I designed used MOSFET switches (transmission gates so an NMOS and a PMOS in parallel with their gates driven 180º out of phase). It was basically a resistor ladder from VCC (or Vref) to GND and the DAC code turned on the transmission gate switch at the appropriate tap. If all the transmission gates were the same size, the ones near the ends of the resistor ladder would have different resistance than those in the middle of the ladder. In the T-gates closer to VCC the NMOS would practically be off and the PMOS would be fully on when the T-gate was supposed to be on. And vice versa for the T-gates closer to GND. Near the middle of the resistor ladder there's much lower gate voltage available (-> lower overdrive voltage), so the MOSFETs aren't as fully on. All this contributes to nonlinearity.
A few ohm variation in switch resistance would severely limit the linearity in an R2R DAC with, say, 1k resistors. A few ohm on 1 kΩ doesn't even break ±0.1%.
With BJT switches I'd think you'd be limited by the Early voltage once you push the precision far enough.
Tom
The DAC I designed used MOSFET switches (transmission gates so an NMOS and a PMOS in parallel with their gates driven 180º out of phase). It was basically a resistor ladder from VCC (or Vref) to GND and the DAC code turned on the transmission gate switch at the appropriate tap. If all the transmission gates were the same size, the ones near the ends of the resistor ladder would have different resistance than those in the middle of the ladder. In the T-gates closer to VCC the NMOS would practically be off and the PMOS would be fully on when the T-gate was supposed to be on. And vice versa for the T-gates closer to GND. Near the middle of the resistor ladder there's much lower gate voltage available (-> lower overdrive voltage), so the MOSFETs aren't as fully on. All this contributes to nonlinearity.
A few ohm variation in switch resistance would severely limit the linearity in an R2R DAC with, say, 1k resistors. A few ohm on 1 kΩ doesn't even break ±0.1%.
With BJT switches I'd think you'd be limited by the Early voltage once you push the precision far enough.
Tom
Very thanks for much of informations i need to know to find good r2r dac. They are lots of r2r chinese diy pcb dacs looks good, better than soekris. So im thinking obout them. The total dac pcb is a diy pcb?
Hi Tom thanks for good inputs.
Yes the mismatch simply cant be avoided because all of the the devises are not in the same die and from the same base...
You probably designed a IC with direct implementation of the paerts in the same DIE?
.
I am considering something like this for BJTs
DSC3G03
or double
BFU520Y
and single
BFU530XR
.
it is a just the notion maybe is more practical to go with digital IC switch r2r driver, BUT in 3-state mode?
Bits are coming to EN, inputs are GND (0) or Vcc (1). That way it is possibile to invert MSB and to have -DATA and +DATA for balanced 2 R2R ladders?
I tried this aproach in 1bit DSD with 4 ICs of 8 driver with common EN and the sound was very very good.
Putting all informations together what diy pcb r2r dacs will be good? Better than soekris. Take a lool of this https://m.pl.aliexpress.com/item/10...f16755483220191143e013f!12000031490980525!rec
Looks very good. 0.01% resistors. Can connect better clock. No switches.
Looks very good. 0.01% resistors. Can connect better clock. No switches.
Good R2R DACs tend to be somewhat more costly than oversampling chips DACs. If you know what you're doing an AKM oversampling DAC operating in DSD256 mode can sound better than a lot of R2R DACs, and do so without any digital glare whatsoever. That said if you want a really good R2R discrete resistor DAC, best not to waste you money on cheap junk. Holo Spring 2 or Holo May is probably the way to go and maybe the last DAC you will need to buy. OTOH if you cheap-ish buy junk now you will soon be looking for other low-cost junk to upgrade to a sound you like better. Its a fairly common learning path around here to find that out for yourself.
Yep. The DAC/ADC I designed is part of the LMK04800 timing IC. The components on an IC aren't perfectly matched, but they're a lot closer matched than discrete components. That's why ICs use circuit architectures, such as R2R, where the absolute accuracy doesn't matter much but the matching does.
Tom
Point is that You could not be sure that R parts are with specified % from the Ali or similar sites. Greater probaility that they are not...
Another thing try to make the difference with R2R dacs.
- One that uses processor chip, Soekris, these from ALI etc...
- Other, with digital IC only without software running the processor.
Try not to mix one type with other
You can compare the sound of course.
In this terms switch is active element, not passive switch. and all types of DACs has "switches", that is what the conversion make happening...
Hollo dac is far oway the budget. And im a diyer so im searchinq, moding😉 .yes we dont sure that% is real. But it is pasible to change resistors and put better tolerance. Im searching something with procesor like soekris,,but with one frequency clock. These from allie cost much. Chinese can with co problem make good r2r dac with good % resistors. But neet somechow to chenk this,
This is the only serious performance oriented R2R dac I know of: https://www.lavryengineering.com/pdfs/lavry-da924-manual.pdf Read about the oven timing (to get stable temperature) and self calibration to understand the efforts Lavry went to to get decent performance. Its an interesting dead end for technology. Higher resolution from a resistor based "DAC" is possible but limited by how fast you can rotate the switches: https://www.edn.com/measurement-techniques-help-hit-the-1-ppm-mark/
I have not so good experience with R2R FPGA finished PCB-s. Because they usually not give the tolerance and type of the R2R resistors in ladder... 🙁
If You by from unknown stores You most certainly not get resistors with desired tolerance.
I was measured and amout of noise is not good at all. I by the FPGA/CPLD based R2R to compare to glue logic with 1% worst case Rs
(with trimming 2 MSBs) just to compare. And I was disappointed a bitt 🙁
Sound was even better with glue logic digital ICs even that tollerance was unacceptable 1% standard metal film resistors... 🙁
.
Master clocks and this what is before R2R dac is another story and probaly another topic...
If You by from unknown stores You most certainly not get resistors with desired tolerance.
I was measured and amout of noise is not good at all. I by the FPGA/CPLD based R2R to compare to glue logic with 1% worst case Rs
(with trimming 2 MSBs) just to compare. And I was disappointed a bitt 🙁
Sound was even better with glue logic digital ICs even that tollerance was unacceptable 1% standard metal film resistors... 🙁
.
Master clocks and this what is before R2R dac is another story and probaly another topic...
This is "computer" controlled and even "computer" drive r2r ladder network dac.
This design is matched with small dimmension of the base Raspberry PCB. Because of that there are no "room" and design is well adopted in these conditions...
.
It is interesting and small, but the sound will be sharp, edgy, non analog and with a fatigue. From my listening exp with tat kind of R2R dacs...
And no guarantee that You will end up with resistors type and precision like in the videos...
sorry...
"Spec:
Hand made and hand matched 0.005% resistors"
when I read something like this I am always a bit distant...
Last edited:
There is a patent that has been claimed to be the one Holo uses for error correction on its dacs. Please see attached.
True balanced 26BIT R2R + 6BIT DSD (32 steps FIR Filters)
High Precision Oven Controlled Crystals OCXO
Why are you hung-up on high-precision resistors? All you need are 0.1% with a good tempco and dynamically trim them with software. If I did the math correctly, 0.005% is about -86 dB and your DAC's noise floor will be worse than a CD. Cook up a 24-bit R2R with dynamically trimmed 0.1% resistors and save money. The hardware needed for system calibration is about the same price as a bag of 0.005% resistors but it can be reused with any DAC. You can keep everything tuned up by running the calibration routine whenever the DAC is not used for music.