24-bit R2R DAC using miltiple 16/18/20-bit R2R chips

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hi,

I want tu bulid 24-bit non-oversampling DAC with passive I/V (resistor) and tube output stage.
Ofcourse first choice was PCM1704 but as you probably know this chip can't handle relatively high load resistances - with 10 ohm or lower you get best results =>very low output swing about 25mV. This is not suprise because this chip is designed to work with zero load impedance like input of OPAMP which I want to omit.

There are other R2R DACs which can handle higher resistances and provide more output current. So we got AD1865 AD1862 PCM63 TDA1541 etc.
Now, heve you ever saw a solution using multiple lower bit DACs to achieve higher one? I'm not thinking of pararelling chips to make ~17bit from 2x16bit chips but about making true 24-bit DAC in which lower significant bits are converted by one chip(s) and higher ones by another.

Anyone?

Thanks,
Marek
 
Marek,

What you want is not possible with chips designed for 16 bit. The accuracy of the MSB should be less than half of the LSB absolute value - in your case you want 24-bit accuracy from a chip designed for 16-bit accuracy.
Paralleling DAC chips is a different matter, because of statistical averaging of the errors. You gain 1/2 or 1 more bit depth (I am not sure) by paralleling 2 chips. So you need a lot of DAC chips to get 24-bit depth from 16-bit chips. But then you get the additional benefit of high output current capability...
 
Marek:

There's an approach which meets all of your criteria and still utilizes the PCM1704. Before I explain, you should realize the even the PCM1704 does not provide anything near a 24-bit THD levels. At best the PCM1704 enables a significantly lowered quantization noise floor, but you only obtain using either a true 24-bit data source, or a 16-bit source with an oversampling digital filter.

Okay, with that said, you could still use a 10 ohm i/v resistor with the PCM1704 and have a tube output. With a fullscale output current of 1mA from the PCM1704 the peak signal output will only be 10mV. To obtain a standard 2VRMS output level, you will need to amplify that 10mV about 280 times, or 49dB. Such a gain is less than is typically provided by most two-stage tube phono stages. You can now see where I'm heading.

The necessary tube output stage would be comprised of no more than two stages. Just utilize about any moving-magnet based tube phono design or diy kit and omit the RIAA network. Or, better yet, replace the RIAA network with a analog image-rejection filter. In fact, you would likely have too much gain and would need to pad down the signal level between stages. Even a single tube stage could suffice. An 12AX7 or 12AT7 triode would have enough gain if you used an active current-source, or a large choke of you would prefer, to load the anode and a cathode-follower to buffer the output.
 
Last edited:
Tubes are just not good to do I/V conversion. There is this popular belief that there can magically improve the antiquue DACs like the crappy TDA1541, but reality is that at the best a tube stage can attain only 80-83db Signal/Noise - Dynamic Range. That is equal or less than 14 bit of REAL resolution.
Trying to connect a 24 bit DAC to a tube stage is pointless.
 
16 bit played back on a 24 bit DAC is still only 16 bit playback, however with increased precision.
To make use of additional DAC bits capability requires interpolation and upsampling.

Eric.

I got many hi-res tracks 96/192kHz/24bit and even 384KHz/24bit (dxd files). I use lossless files through USB->I2S converter in asynchronous mode as a digital source. I do not use classic CD transport anymore.

Marek,

What you want is not possible with chips designed for 16 bit. The accuracy of the MSB should be less than half of the LSB absolute value - in your case you want 24-bit accuracy from a chip designed for 16-bit accuracy.
Paralleling DAC chips is a different matter, because of statistical averaging of the errors. You gain 1/2 or 1 more bit depth (I am not sure) by paralleling 2 chips. So you need a lot of DAC chips to get 24-bit depth from 16-bit chips. But then you get the additional benefit of high output current capability...

I was thinking about this solution (usign 16-bit chips):
-DAC chip feeded with 16 most significant bits loaded with eg. 100ohm resistor
-DAC chip feeded with remaining 8 bits loaded with 100ohm/65536 resistor

and then sum both output voltages (but I don't know how)


Marek:

There's an approach which meets all of your criteria and still utilizes the PCM1704. Before I explain, you should realize the even the PCM1704 does not provide anything near a 24-bit THD levels. At best the PCM1704 enables a significantly lowered quantization noise floor, but you only obtain using either a true 24-bit data source, or a 16-bit source with an oversampling digital filter.

Okay, with that said, you could still use a 10 ohm i/v resistor with the PCM1704 and have a tube output. With a fullscale output current of 1mA from the PCM1704 the peak signal output will only be 10mV. To obtain a standard 2VRMS output level, you will need to amplify that 10mV about 280 times, or 49dB. Such a gain is less than is typically provided by most two-stage tube phono stages. You can now see where I'm heading.

The necessary tube output stage would be comprised of no more than two stages. Just utilize about any moving-magnet based tube phono design or diy kit and omit the RIAA network. Or, better yet, replace the RIAA network with a analog image-rejection filter. In fact, you would likely have too much gain and would need to pad down the signal level between stages. Even a single tube stage could suffice. An 12AX7 or 12AT7 triode would have enough gain if you used an active current-source, or a large choke of you would prefer, to load the anode and a cathode-follower to buffer the output.

Yeah, I know that there is no audio DAC which provide real 24-bit linearity, but still there are sonic benefits of using 24-bit data with non-ideal linearity 24-bit DAC.

About 280x tube stage gain - that was my point.

Tubes are just not good to do I/V conversion. There is this popular belief that there can magically improve the antiquue DACs like the crappy TDA1541, but reality is that at the best a tube stage can attain only 80-83db Signal/Noise - Dynamic Range. That is equal or less than 14 bit of REAL resolution.
Trying to connect a 24 bit DAC to a tube stage is pointless.

From my experience tubes are great as output stage of DAC. I currently use SRPP circuit with 6N30P tubes with ESS Sabre32 DAC with impresive results but now I want to try one of top ladder DAC in nonos mode (only for higher sampling freq tracks due to known treble roll-off)


Another solution is PCM1704 with step-up trafo. I got LL1931 which has extremly low primary DCR ( 0,9ohm in 1:16 config.) so loading secondaries with 3k will results with effective 10ohm seen by DAC chip.
 
Hi,

the named DACs are current output DACs which all feature similar or lower current amplitudes than the PCM1704 with its 2.2mApp@1kOhm. AD1865=2mApp@1.7kOhm, PCM63=4mApp@670Ohm and the TDA1541=1.2mApp@?kOhm. Lower current amplitude means higher IV-resistor-values for a given voltage amplitude, meaning higher noise figures.
All those DACs can´t handle high load impedances because of the restricted voltage compliance. Its just that DACs like the AD1865 feature additional OP amps on-chip, so that You can easily configure a voltage output (IV-converter stage) just by providing for external resistors.
But what´s the sense in using Tubes as output buffer and putting OP-amps into the signal path at the same?
Another issue is the number of Bits. Every doubling of chips results in just 1 Bit more resolution. To achieve 24Bit resolution with a 16Bit DAC chip asks for massive parallelization (8Bit=256). The second solution would be to work with two DACs using different internal reference voltages and to multiplex between the two pipelines.
So it´d be best to stay with the PCM1704 even though its quite costly.
You also might give the NOS-idea a second thought. How do You manage different clock rates and how do You switch bandwidth limits of the output aliasing filters? NOS has serious disadvantages technically as well as sonically (A piano player hits the fingerboard with his fingers...he dosen´t plug the strings with the blade of a saw). The PCM1704 is capable of clocking at up to 768kHz using a appropriate up-/oversampling filter. At such high clock frequencies one can safely omit with the output aliasing filter which accounts for much of the bad reputation of digital sound.
Lastly You might rethink about using tubes as they limit the capabilities of a DAC stage. It´d be hard if not impossible at all to achieve better than 12-14bit linearity and dynamic range.

jauu
Calvin
 
Yes those DACs has similar output current BUT diffenent tollerance to load resistance. eg AD1865 (current output!) perform very well with 100ohm resistor, with PCM1704 it is unacceptable.
About aliasing filter. When using high sampling freq there is really no need to use of aliasing filter. I'm using tube amplifier so filtering is done on OPTs.
 
Last edited:
Yeah, I know that there is no audio DAC which provide real 24-bit linearity, but still there are sonic benefits of using 24-bit data with non-ideal linearity 24-bit DAC.

About 280x tube stage gain - that was my point.

What was your point? I thought you had dismissed the PCM1704 for use with tubes due to it's need for an 10 ohm, or thereabouts, i/v resistor to attain the best sound. I'm now a bit confused about what you are seeking to accomplish.
 
Last edited:
I want to make nonos 24-bit R2R DAC with passive I/V and tube output stage.
PCM1704 seems unreal because of very low voltage swing with 10ohm resistor - tube stage should have extremely high gain (noise problems etc).
So that is my point and reason for serching another solution.
 
I want to make nonos 24-bit R2R DAC with passive I/V and tube output stage.
PCM1704 seems unreal because of very low voltage swing with 10ohm resistor - tube stage should have extremely high gain (noise problems etc).
So that is my point and reason for serching another solution.

But that is my point - Tube phono stages are used to amplify far smaller signal levels, on the order of 20dB lower than would be provided by a PCM1704 through a 10 ohm resistor. As has been pointed out by Calvin above, tube noise is going to swamp any benefits of 24-bit conversion anyhow. Even if you could obtain a 2VRMS signal from a PCM1704 and i/v resistor the tube noise would still swamp converter resolution much beyond 15-16 bits.
 
Last edited:
I did say that already, but he won't belive. Blind faith is something special...

Yes, you did. I had inadvertently neglected to note your prior comment, SoNic.

At this point, I'm at a loss to even understand what it is he hopes to accomplish. I've suggested a rather simple solution which effectively addresses his entire list of objectives, yet he rejects it. Puzzling.
 
Hi,

I´m also a bit confused as to what Marek wants. NonOs would only make sense if the input signal is already clocked at high speed, say >100kHz. To not OS means the need to change the aliasing filter´s bandwidth limit depending on the clock of the incoming signal. I´d only omit with the analog post filter if input signal is sampled at more than 100kHz, otherwise there´d be too much artefacts left in the audible range.
One reason for OS is to supply the DAC-chip with a single constant highfreqency-clocked signal to allow for a fixed-frequency aliasing-filter or to omit with the analog post filtering at all. It´d also be good to post filter at an early stage as possible, before the stepped signal reaches an amplifier input. Otherwise the very steep flanks of the steps may overload the amplifying stage and generate distortion unless the amplifying stage is much faster.
The reason for the BB/TI DACs only capable to allow for small voltage levels at their DAC-outputs are protection diodes which start conduction at a few hundred mVs. The DAC allow for higher voltage levels (i.e. higher impedance levels) at their outputs but then THD figures rise quickly.
I don´t know if the ADIs feature output protection devices or not, but voltage output level is restricted to the voltage compliance which is typically around +-1V (depending on the devices used for the current switches, either MOS or bipolar). If the ADIs feature output protection diodes, then the same applies to them with regard to Distortion as applies to the BB/TIs. Unfortunately te Datasheet doesn´t inform us about the output configurtation, nor the behaviour of the Iout outputs.
In any case, the word length of 18Bit of the AD1865 asks for 64 parallelled Devices to achieve 24Bit.
In any case, if I wanted to accomplish something like NonOS with a 24/192 capable DAC-chip I´d opt at least for linear Upsampling to 176-192kHz (wordlength may remain anything between 16 and 24Bit) to have a clean clear situation at the DACs output/I/Vs input. And I´d use of course a 24Bit capable chip. Anything else is just asking for trouble.

jauu
Calvin
 
Ken, Sonic,

I did read your post again and you are absolutely right. S/N of finest tube preamp is about -90dB so it is equal to about 16bit resolution.


Calvin,
Yes I want to use NONOS DAC for high sample rate tracks ONLY. Red book will be upsampled by software resampler to get rid of aliasing
And yes the reason for low tollerance of load resistance are protection diodes. From experience AD R2R DACs tollerate much higher resistances than BBs.

I don't want to parallel multiple DACs feeded with the same word length (eg 18bit) but I wonder if I can use eg two 16-bit chips, one feeded with MSBs(16bit) and second by LSBs (8bit) and load them with weighted resistances.
 
Last edited:
Marek,

This will be my final attempt at advising you on this matter. In theory, one could create such a composite 24-bit converter if the DAC chip handling the LSBs could accept an external reference voltage reduced to produce a full-scale output equal to the magnitude of a single bit step of the DAC handling the MSBs. That's in theory. In practice, however, I doubt that the overlap could be set accurately enough to not have a significant non-linearity or discontinuity between the two DAC chips, effectively negating the purpose for creating a 24-bit converter in the first place. The noise floor of the DAC handling the LSB would, unfortunately, NOT be similarly reduced by the reduction in full-scale output level. You would likely have to use an M-DAC, which usually has other issues complicating it's use as an audio DAC. You would also need to implement some non-trivial control logic to coordinate the processing of the two DAC chips. Such a design might present an interesting challenge for an experienced converter designer, but would undoubtedly result in a failed diy project.

Or, you could simply use a PCM1704 + low value i/v resistor + high-gain tube output stage, and have music. It's your choice.
 
Last edited:
I was thinking about using a 2x PCM63 current-out DACs. First, which carries 16 MSBs loaded with eg 100ohm (max acceptable value for this chip) and second which carries remaining LSBs loaded with 100ohm/2^16=0,0015ohm. This is proof that this solution is completely unpractical.
Glue logic for this solution is trivial - just 16-bit D flip-flop DATA line delay for first chip, BCLK, LRCLK the same for both chips
Then both voltages should be summed.

Yeah, I will try PCM1704 with resistor or step up trafo followed by quiet tube stage.

Regards
Marek
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.