Ladder DAC confusion

[Adam Frandsen]

I am in the process of trying to understand and maybe eventually build my own ladder DAC, but I am struggling a bit to understand all the details, specifically:

1. Most content I play and stream is 24bit, thus I would like to have the best representation of that signal
2. The highest precision resistors I can find are 0.001% from VPG, this is only enough for 16bit accuracy
3. A standard DAC chip uses sigma-delta/multi-bit (max 5 bits, as I understand)
4. 16 raw bits then seem superior, if I then use the same type of oversampling via XMOS as the DAC chips use… or?
5. Does the resulting higher error count, non linearity, noise… caused by using 0.001% resistors in a 24bit ladder make it a bad choice over a more precise 16bit ladder, or will it still convert 24bit sources more accurately?
6. I see many designs using CPLD’s and FPGA’s, is this necessary, programming these would be outside my capabilities for sure.
7. When soldering in those resistors, won’t the amount of solder etc. be enough to mess with such a high precision margin?

that is it for now. Thanks

 

[Newmarket]

Imo it's not practicable to attempt such precision with soldered resistors etc.
CPLDs are near extinct now (FPGAs dominate).

 

[Adam Frandsen]

Imo it's not practicable to attempt such precision with soldered resistors etc.
CPLDs are near extinct now (FPGAs dominate).

Okay - so what you're saying is forget trying to make a ladder DAC?

 

[abbey road d enfer]

If you want to experiment for your own enlightenment, why not? But if you want to do that hoping for superior performance, fuggetdaboudit.
The nice thing about ladder converters is that the accuracy is supposed to be about that of the worst pair of resistors, but is theoretical. You have to consider many other factors, switch accuracy, tempco...

 

[ccaudle]

I see many designs using CPLD’s and FPGA’s, is this necessary

You are going to build all the necessary decode logic by wiring TTL chips together?

 

[Newmarket]

Okay - so what you're saying is forget trying to make a ladder DAC?

I think Abbey got it right. Great to do for understanding. Applies to many things eg instrumentation amps esp if going for DC accuracy.

 

[Adam Frandsen]

If you want to experiment for your own enlightenment, why not? But if you want to do that hoping for superior performance, fuggetdaboudit.
The nice thing about ladder converters is that the accuracy is supposed to be about that of the worst pair of resistors, but is theoretical. You have to consider many other factors, switch accuracy, tempco...

Right - well yes, I guess I did have hopes of being able to make something great sounding - think I will postpone this project... Thanks

 

[thor.zmt]

I am in the process of trying to understand and maybe eventually build my own ladder DAC

What next? Your own DIY version of SpaceX Starship?

1. Most content I play and stream is 24bit, thus I would like to have the best representation of that signal

Really? Where do you find REAL 24 Bit content (not upconverted CD quality or 48k masters) that is worth listening to?

2. The highest precision resistors I can find are 0.001% from VPG, this is only enough for 16bit accuracy

Correct. Mind you, that alone is one of the lesser problems you will have. Things to consider si how to handle the zero crossing glitch, for example,

3. A standard DAC chip uses sigma-delta/multi-bit (max 5 bits, as I understand)

This is not quite correct. There are many ways of making a DAC. You can buy off the shelf 20 Bit DAC's with 1MSpS data rate that are classic "Multibit". They are not cheap, but you can get them. Of course, they are "industrial" types and need SPI input, not EIAJ or IIS and do not accept two's complement.

4. 16 raw bits then seem superior, if I then use the same type of oversampling via XMOS as the DAC chips use… or?

XMOS is a Transputer with 8-16 Cores and some DSP libraries. If you can program one of those, you can do a few really cool things.

5. Does the resulting higher error count, non linearity, noise… caused by using 0.001% resistors in a 24bit ladder make it a bad choice over a more precise 16bit ladder, or will it still convert 24bit sources more accurately?

Most discrete R2R DAC's perform poorly. For example:

Measurements & Review Cayin N6MK2 R01 - L7Audiolab

Measurements & Review of Cayin RU6 USB DAC/HPAmp - L7Audiolab

Mind you, you could do worse, like using a "Nu-Tube":

Measurements & Review of Cayin N3Pro Timbre DAP - L7Audiolab

You can build a system with a second "parallel" error correction to lower basic HD notably, but you have to be able to measure the error and programm the correction into ROM for the system. The Holo Audio units supposedly work that way:

Measurements of HoloAudio Spring3 Discrete R2R DAC - L7Audiolab

6. I see many designs using CPLD’s and FPGA’s, is this necessary, programming these would be outside my capabilities for sure.

This is generally needed because we need convert (at the minimum) the data format from Audio (EIAJ/IIS) to a format that can be used in a discrete DA.

7. When soldering in those resistors, won’t the amount of solder etc. be enough to mess with such a high precision margin?

I'd be more worried about the logic in use. The differences between two outputs will throw you way off.

Okay - so what you're saying is forget trying to make a ladder DAC?

NOT QUITE. But it is a level of challenge, one I declined for paid work.

If you want to "DIY" a DAC, you buy the Soekris board (it's not that good though).

If you want great R2R sound quality, find something with one of the following DAC Chips:

PCM63/1702/1704
AD1862/1865

Modify it to allow 24 bit / up to 768kHz input, using an off the shelf XMOS or similar board and learn more about how it all works.

Then maybe you will be ready to take it to the next level.

Thor

 

[Philbo King]

I think you would need a laser-trimmed resistor array to get better than 14 to 16 bit accuracy.

 

[Adam Frandsen]

What next? Your own DIY version of SpaceX Starship?



Really? Where do you find REAL 24 Bit content (not upconverted CD quality or 48k masters) that is worth listening to?



Correct. Mind you, that alone is one of the lesser problems you will have. Things to consider si how to handle the zero crossing glitch, for example,



This is not quite correct. There are many ways of making a DAC. You can buy off the shelf 20 Bit DAC's with 1MSpS data rate that are classic "Multibit". They are not cheap, but you can get them. Of course, they are "industrial" types and need SPI input, not EIAJ or IIS and do not accept two's complement.



XMOS is a Transputer with 8-16 Cores and some DSP libraries. If you can program one of those, you can do a few really cool things.



Most discrete R2R DAC's perform poorly. For example:

Measurements & Review Cayin N6MK2 R01 - L7Audiolab

Measurements & Review of Cayin RU6 USB DAC/HPAmp - L7Audiolab

Mind you, you could do worse, like using a "Nu-Tube":

Measurements & Review of Cayin N3Pro Timbre DAP - L7Audiolab

You can build a system with a second "parallel" error correction to lower basic HD notably, but you have to be able to measure the error and programm the correction into ROM for the system. The Holo Audio units supposedly work that way:

Measurements of HoloAudio Spring3 Discrete R2R DAC - L7Audiolab



This is generally needed because we need convert (at the minimum) the data format from Audio (EIAJ/IIS) to a format that can be used in a discrete DA.



I'd be more worried about the logic in use. The differences between two outputs will throw you way off.



NOT QUITE. But it is a level of challenge, one I declined for paid work.

If you want to "DIY" a DAC, you buy the Soekris board (it's not that good though).

If you want great R2R sound quality, find something with one of the following DAC Chips:

PCM63/1702/1704
AD1862/1865

Modify it to allow 24 bit / up to 768kHz input, using an off the shelf XMOS or similar board and learn more about how it all works.

Then maybe you will be ready to take it to the next level.

Thor

I will likely proceed using the still in production DAC11001B (20-bit, and has a special anti glitch function) for PCM signals and a separate section based on either Sabre or AKM chips for native DSD decoding - two separate power sections as well. I will outsource the parts I cannot do myself on Fiverr. Any suggestions to the build and what I should ask the engineer on Fiverr to include and pay attention to is much appreciated

 

[living sounds]

I use Soekris DACs for everything. Build my own 32-channel converter out of their cards, and while there are a few problems with it, the biggest being that the developer doesn't seem to care about his customers (anymore), I find it to be a far superior sounding solution to modern DS converters.

A lot of work and thought went into developing modern discrete resistor based ladder DACs and it doesn't make any sense to attempt this on your own, unless you really know what you are doing and have a lot of time, IMHO...

The DACs I use have very low noise, custom digital antialiasing filters that don't destroy the signal (like the steep digital filters used in most modern converters), no highpass (DC offset) filter that negatively impacts the signal, THD is reasonably low at any amplitude (thanks to the sign-magnitude principle) and if you prevent the clock from readjusting constantly (like I do with a workaround because Soekris won't fix the issue) jitter isn't in the way of good sound. Linearity is near-perfect down to the noisefloor.

The Soekris boards needed a few well-placed additional power filtering caps to perform satisfactory, too.

 

[Wavelength]

I have done this with FPGA and previously with CPLD which sounded better. I had custom resistor packs for R2R network from Vishay and it sounded pretty good. You might be better off working with some of the available parts that are either not made anymore. There is tons of these from Philips, Analog and Ti. You would be surprised what some of the original NOS parts can sound like at 16bits.
BTW was an earlier adopter to XMOS which is basically an IO processor. XMOS is trying to do too much now so I have moved on to other platforms.
There are also a lot of DIY modules available on the net these days doing most of the heavy lifting. Put some nice tube buffer on the output or take the differential signal and convert that to single ended with a transformer or something to put your own twist on it. Designing a good power supply will yield better results as well.

 

[Wavelength]

BTW there is probably a lot more info over on DIYAudio than here on the subject.
Also take a look at Comm-True for USB solutions better setup than XMOS: https://www.comtrue-inc.com/

 

[Newmarket]

I have done this with FPGA and previously with CPLD which sounded better. I had custom resistor packs for R2R network from Vishay and it sounded pretty good. You might be better off working with some of the available parts that are either not made anymore. There is tons of these from Philips, Analog and Ti. You would be surprised what some of the original NOS parts can sound like at 16bits.
BTW was an earlier adopter to XMOS which is basically an IO processor. XMOS is trying to do too much now so I have moved on to other platforms.
There are also a lot of DIY modules available on the net these days doing most of the heavy lifting. Put some nice tube buffer on the output or take the differential signal and convert that to single ended with a transformer or something to put your own twist on it. Designing a good power supply will yield better results as well.

CPLD sounded better than the FPGA. Which devices ?