R2R or single-IC for DAC?

The TDA1540 and TDA1541 are far more advanced than just a plain old R-2R DAC. They use current sources that are switched to the output or to ground and use an R-2R network for generating the bit currents for the lower bits, but have a dynamic element matching technique for the higher currents.
So 1540 and 1541 are hybrid of DEM and R2R? Hmm... I always thought these classics were pure DEM. Any advantage (or limitations/drawbacks) of using DEM for generating the bit currents for the lower bits as well?
 
In any case, I can't find a similar heavy-traffic thread on AKM dacs -- do any exist?
Don't know. ESS were all the rage for awhile. Great measurements, 'swiss army knife' extensive feature set. Seems like we all spent some time with them, know I did. Then came AKM's AK4499EQ which could compete on measurements, and which some people thought sounded better than ESS (pretty sure that's what AKM was aiming for with that particular chip). Anyway, if you have questions about AKM dac implementation, pretty sure you can get specific questions answered, and or be offered some general advice.
 
Don't know. ESS were all the rage for awhile. Great measurements, 'swiss army knife' extensive feature set. Seems like we all spent some time with them, know I did. Then came AKM's AK4499EQ which could compete on measurements, and which some people thought sounded better than ESS (pretty sure that's what AKM was aiming for with that particular chip). Anyway, if you have questions about AKM dac implementation, pretty sure you can get specific questions answered, and or be offered some general advice.
Likely, with ESS, it may have been initial cheap batch price, the 9038 being offered in many low-$ Ali/ebay kit DACs, its tweak-ability factor (as you suggest), etc. Not sure AKM was as charitable to Chinese kit manufs, with licenses and low costs, but they did offer their devices to Chinese plug-n-play manufs like Topping, SMSL, etc.
And then there was the AKM fire, and huge mark-up on AKM dac prices, based on hoarders selling their inventory. Not good for DIY.

Doesn't really matter now that BOTH manufs are releasing new models that are improved over prev. ones. I already mentioned new ESS models. But AKM has a new "two-chip solution with the AK4499EX and the AK4191".
https://asahi-kasei.eu/world-premiere-new-flagship-dac/

Back to the slow sound I noted about discrete R2R ... well, that speed of light thing may be part of the problem -- R's too far spaced. I have HiBy dap with a micro R2R pcb inside. This device sounds quick and punchy, not at all like Denafrips.
1665595550308.png
 
Would agree with much of that. Have an AK4499EX eval board here that is in the process of getting working. Have listened to a number of dacs here using electrostatic speakers and or planar headphones, including entry level Denafrips. Have a custom discrete resistor dac with SOA clocks, galvanic-isolated FIFO buffering, custom power supplies, etc. Its pretty darn good, but not better in every way than my old custom AK4499 dac running in DSD256 mode. Like the custom discrete resistor dac more at LF, and like the AK4499 in DSD256 better at HF. To me, its that we don't have all the answers yet if perceptual SQ is the goal. If great PSS measurements are the end goal then the problems would all seem to have been solved before now. Yet, I risk being attacked for holding such an opinion. Makes life complicated when the science isn't as settled as some people seem to need to believe. Just saying as much as I have might provoke some strong contrary opinions. Oh, well. Don't wish to submit to censorship nor to being canceled.
 
Well before the discrete R2R phenomenon began in the mainstream market (Denafrips, Holo, etc), I saw the Soekris PCBs here on diya ... and noticed a couple of things. (1) The PCB looks blingy and impressive -- all those R's nicely arranged on a PCB has a certain "artistic" / show-offy appeal. (2) Long ckt traces (compared to ASICs) and all those R's (= Johnson noise). Both of these observations STILL make me skeptical of the overall advantages of going discrete R2R .
About the prettiness and bling factor ... just have a look at the latest Gustard DAC, and kind of hypnotic attention it draws from audiophiles not as technically "aware" of real-word physics.
https://audiophilestyle.com/forums/topic/65962-gustard-r26-new-r
 
One of the questions to investigate is not just about R2R ladder but "how" the datas are managed before passive network.
(With persumption that Rs are from very small tolerances.)
.
1. By the cassic digital ICs, glue logic
2. CLPD, FPGA highly integrated ICs.
.
I have a R2r ladder dac with discrete logic
I made one piece of R2R discrete dac, glue logic without CPLD or FPGA, 32bit word. (4 PCB to try Balanced vesion with inverted DATA line.)
Compared to the FPGA driven ladder network, somehow prefer more glue logic DIP8 version.
I deliberatly use 1% Rs just to se what happening. Version vith processor uses 0.1% 0.05% I dont know.
Somehow the sound was better for me?
Measurements showed as i am remebered well, not so small amount of digital noise with booth pieces.
Also I spoted that balanced versions not showing an improvement but giving a slight more noise...
So I paralleled the boards to try, and was better. I must say that it was with the same value of supply voltage. onlu one voltage
of +5V for DIP8 version and +3.3V for processor version.
I dint try to make ballanced with one branch +5V-0V, and other 0V-(-5V). :)
.
Also I can report that trimming of the MSB or 2 first bits, (@-60db) is usefull but, not fully satisfied becouse it changes over the change of temperature with classic 1% resistors and multiturn small pots.
 
Looks like pure R-2R, no sign-magnitude. I wonder if anything special is done to keep glitches around the zero crossing under control.
I don't know whether the glitches would be visible on his listing scope screen shots. I have not seen any on my own scope testing.

If the glitches are audible, then none are heard - the output is very clean and quiet. In fact given it only has 2nd order 24k lp filter and of course is NOS, I was pleasantly surprised.

It needs true I2S (Philips spec 2s complement) and I drive it with a DIR9001 spdif interface.
 
The thermal noise of the lower sections gets halved by each higher section. The thermal noise of the whole R-2R network is simply the thermal noise of its output resistance R.
BTW, what is Your OP about the value of the resistors used in the network? In terms of current to the switching digital logic IC?
I tried 1mA from +5V to "0" state, 5Kohm...
 
I don't know whether the glitches would be visible on his listing scope screen shots. I have not seen any on my own scope testing.

If the glitches are audible, then none are heard - the output is very clean and quiet. In fact given it only has 2nd order 24k lp filter and of course is NOS, I was pleasantly surprised.

It needs true I2S (Philips spec 2s complement) and I drive it with a DIR9001 spdif interface.

I welcome anyone with a different ladder dac doing a comparison.
 
I don't know whether the glitches would be visible on his listing scope screen shots. I have not seen any on my own scope testing.

If the glitches are audible, then none are heard - the output is very clean and quiet. In fact given it only has 2nd order 24k lp filter and of course is NOS, I was pleasantly surprised.

It needs true I2S (Philips spec 2s complement) and I drive it with a DIR9001 spdif interface.
I2S is in the input probably because most og the interfaces exporting this format. But I think that input format of DAC is Left Justified, with MSB inverted.
That is what I did in mine and it is working. It is not huge opperation just deserves few more chips for delaying the LE line for one BCK period and one inverter after shift register serial-to-parallel management in the MSB position.
 
I don't know whether the glitches would be visible on his listing scope screen shots. I have not seen any on my own scope testing.

If the glitches are audible, then none are heard - the output is very clean and quiet. In fact given it only has 2nd order 24k lp filter and of course is NOS, I was pleasantly surprised.

It needs true I2S (Philips spec 2s complement) and I drive it with a DIR9001 spdif interface.
If there were a glitch problem, it would be most noticable during soft passages. Not during absolute silence, but during very soft music.
 
BTW, what is Your OP about the value of the resistors used in the network? In terms of current to the switching digital logic IC?
I tried 1mA from +5V to "0" state, 5Kohm...
I don't know.

I expect that if R is chosen too low, the difference between the high and the low output resistances of the logic registers driving the resistors will dominate the distortion, especially at low levels (you can correct for the average logic register output resistance, but not for the variations). If R is too high, the undesired filtering caused by parasitic capacitances and the resistors will dominate the distortion. There must be an optimum somewhere, but where?