Can multiple 16 bit DACs make 24 bit definition?

[BrianHG]

We aren't even sure what the OP want to do or why.
If he only needs 1khz output bandwidth, then running a super high quality 12bit dac really fast, with a super precision voltage reference,  toggling / dithering the dac between 2 values on 1 LSB bit with a simple RC filter on the analog output can still achieve a programmable 24bit DC level / reference.

I guess it's time to stop tossing spaghetti at the wall to see what sticks.

[Kokoriantz]

This how the modern DACs convert, how much of the values remain exact after passing through the washing machine.

[SiliconWizard]

The PCM1794 is one of the best 24-bit audio DACs out there, I've used one in a design.

But you apparently want to do better. Nice. I'm sure that will happen.

Now of course most modern audio (and not just audio btw, this is the case for most modern high-res) DACs are sigma-delta, and sigma-delta modulators require some significant processing to get the job done with proper results, but it's a very well known and mastered topic these days.

Anyway.

[ejeffrey]

Modern DACs have internal oversampling, interpolating filters that alter the sampled values at recording. The purpose of the DSP along by reading and converting different formats, is to interpolate the mid points between two samples by using windowed sinc function and repeat second time with calculated midpoints to get 4 times higher resolution. The DAC in such case should not alter the data as is the case with modern DACs.

You really don't need to worry about that.  You can  just treat the delta sigma DAC as a "regular" DAC and ignore the internal signal processing.  They are  very good approximations of an ideal DAC up to ~80% of the Nyquist frequency.

There are a couple of real limitations to delta sigma DACs.  The first and by far the most important is the latency.  The internal signal processing increases latency. That a problem for high speed control applications where the extra phase can make feedback loops unstable.   Minor issues are that that the noise floor isn't flat and that the transfer function rolls off before the Nyquist frequency due to the digital anti aliasing filter.  This means you can't use them for second/third Nyquist zone generation. Actually you can but you need to modify the modulator to have a bandpass filter rather than low pass.  This is relatively uncommon to do with either R2R or DS converters.

[Kokoriantz]

If you read carefully the explanation of the decoder, it works as I suggested in post 4. The upper 6 bits command the five level delta sigma value. That is if the signal level is low the 18 lower bits are multiplied by a low voltage level. This type of DAC can also be implemented using a PWM of lower 12bits and the voltage of the pulse is provided by the higher 12 bits. It works if you mesure single frequency signal. Where these good measuring solution fails is when you have a high level bass and very low level high frequency, the system cannot provide resolution for the high frequency in such case. In short it can give clean 0 +/-1 LSB but not max +/-1 LSB.
On "Ali" I see Chinese versions of PCM1704 sold for 70$ but only 12$ for 1702 20 bit new production, May be it worth to try out. Still I will dig in to make use of the 16bits cost nothing only $1 for 10pcs. 

[jonpaul]

the technique of high resolution converters began in 1942, see SIGSALY Quantizer flash converter and progressive or cascaded converters.

By 1970s we used rundown converters and then the first 16 bit Sony rundown / residue converters

We started 16..24 bit digital audio design in 1985 at the dawn of oversampling ADC/DAC.

Suggest that the OP research à bit deeper before reinventing the wheel.

Bon chance

Jon

[Doctorandus_P]

You can try and probably get it to "work", however, whether it will be an improvement is very doubtful.
You would have to start with a 16 bit DAC, which has the specifications of a 24bit DAC, and I doubt very much they exist. Result is that the the "main DAC" has so much imperfections that you just don't know what the second dac will be correcting for.

[Kokoriantz]

The best I got at this moment is by paralleling DACs the signal is added in current but the noise is added in power, that is to the √n. http://www.sengpielaudio.com/calculator-leveladding.htm
This means if I parallel 32 DACs of 98db S/N each I get 113db in total, this allows me to add 3 more bits. As I am transfering the delta and not the value, 3 bits becomes 16 new levels that will be added by the interpolator, may be it is enough.   

[Kokoriantz]

The PCM1794 is an 18bit DAC resolution +6 modulators and costs $5 only with -140db noise.. It is possible to use two paralleled by fixing one, the most 5 bits to 1 so that the multilevel is fixed to max, loaded by the highest 18 bits + sign but the second has the 5 lowest bits + sign and all others to zero. By this I get 24 bits non modulated for $10.

[Doctorandus_P]

... may be it is enough.

Enough for what?
Your part numbers suggest you want to do something with audio.

[Kokoriantz]

Yes, it is audio. By interpolating twice, I get three new samples between two initials and placing them on a grid of minimum 16 levels is enough.

[SiliconWizard]

If it is audio, unless you are actually designing a *measurement device* (that would obviously need to have much better performance, as a lab instrument), then trying to do better than the PCM1794 (or others on the market) sounds completely pointless. Those are already very good and anything after the DAC, such as any amplifier, is going to completely ruin any attempt at getting better specs anyway. I don't get it. Or are you designing a lab instrument? In which case, that could make sense. If you are designing something to actually *listen* to audio, then it's, no offense, wanking.

[Kokoriantz]

This what your beloved DAC produce. I believe, if only the MSB+18 lowest bits are used keeping all others zero, There will be flatter steps. Even 2×19 bit for $5 is a bargain.

[KaneTW]

I'm convinced Kokoriantz is the treez of audio.

[Kokoriantz]

How to compete?

[ejeffrey]

This what your beloved DAC produce. I believe, if only the MSB+18 lowest bits are used keeping all others zero, There will be flatter steps. Even 2×19 bit for $5 is a bargain.

I'm not sure what you are trying to say here, nor have you provided enough context for the scope capture you showed for it to mean anything.  At the very least show the horizontal and vertical scales and tell what the two traces are!

It is hard to say but it looks like you don't have a lowpass filter on the DAC output, if so that is a pretty big problem.

[BrianHG]

This what your beloved DAC produce. I believe, if only the MSB+18 lowest bits are used keeping all others zero, There will be flatter steps. Even 2×19 bit for $5 is a bargain.

Show us your DAC schematic.
And we better see the DAC running in mono mode, differential output feeding an opamp filtered, then with a second opamp with the gain set to a +/- 15v output full scale (IE: at least +/-18v supply for the opamps), all mounted on a single PCB with proper ground planing and EMI filtering.

All I see is some inaccurate measurement where the scope probe is picking up ground ring and noise floor due to a misunderstanding how tiny a voltage the OP is trying to make and all it's surrounding consequences.

And if we see the DAC wired in a standard stereo mode, with all the gain in the measurement on the scope side, or you doctored the gain of the data for your other dac reference, then, I don't think you understand or read the DAC data sheet, or the consequences of the gain and bandwidth of using a scope's gain to measure the DAC's performance.

(I'll permit a 20vp-p output with a +/- 15v supply. as a valid test.  And your output should be wired directly into your scope with a heavy thick shielded RCA cable tp RCA-BNC adapter, no probe, with your scope bandwidth set to 20MHz.)

[Kokoriantz]

I copied this picture from Youtube, I noted it is 15khz before and after filtering in blue. The noise you see is not noise it is 64xfs, oversampled, single bit multi level. I was wrong about the step levels. If MSB and 18 lowest bits are only used, the output maybe is single bit single level as normal delta sigma should. Note also that at positive high level, the modulation is much lower than the negative side. This confirms the lower bit definition is variable, but not symmetrical to zero as I thought.
This DAC if used with external filter, it works only in mono. 
 Imagine I need to see my interpolator result bellow -90db. If max output is +/-10V, I will be looking in 0.3mv range, it is impossible to verify with scope+ preamp the result with delta sigma, whereas with 20bit direct converter I can see as bellow.