Need for higher resolution (not accuracy) with AO

[EricEvola]

Problem Description:

A system that I'm controlling requires a 0-10V input signal which is used to linearly control an output. Even with any 16-bit DAC, the step sizes in the voltage input signal do not produce the fine control I require to tune the output. For example when I increase the output voltage by the smallest amount the AO is capable of for the 10V range... 

(VoltageRange/DeviceResolution) = (10 V / 2^16) ~= 150 uV

... this causes too large of a change to the thing I'm controlling (electro magnet).

1. I don't care about accuracy because I can measure the output very accurately and fine-tune the input accordingly.
2. I don't care about speed of the system too much (<10 seconds for next set point).
3. I do care about cost (<$4,000 target).  

Initial Ideas:

* Bias or combine channels with smaller output range selected to get about double the step size.

* Using 24-bit DSA. Expensive, has unneeded components. Doesn't seem right for these non-dynamic control applications.

* Using another vendor's hardware (blasphemy I know!) with LabVIEW (phew!)

How do I best achieve smaller step sizes for AO in this scenario?

Your knowledge or previous experiences would be really really appreciated. Kudos provided.

[drjdpowell]

---

@EricEvola wrote:

Initial Ideas:

* Bias or combine channels with smaller output range selected to get about double the step size.

---

If you can somehow combine voltages of two AO channels (depends on the details of the hardware) then you can do "coarse" and "fine" gain, with the "coarse" set for 10 Volt range and the "fine" set at the minimum range.  If the lowest range is 1 Volt, for example, then you will get 10 times better resolution.  

[saratoga2]

If you really don't care about speed, you can increase the resolution of your 16 bit DAC almost infinitely just by putting a lowpass filter with a long time constant (several seconds) and then dithering or noise shaping the output.  The output will bounce back and forth between the levels  closest to the true voltage you want, and then filter will emit the average.  Since your A/D will be about 6 orders of magnitude higher frequency than your system response time, and you can get >2bits/octave from noise shaping, your theoretical voltage step size is somewhere in the attovolt range (many orders of magnitude below the thermal noise floor at room temp).  

[Eric-E]

Excellent, that's an idea that I should have but had not yet considered. I will look more into the specifics for this. If I end up implementing that as a solution I will post back here with results or follow ups. 

Thanks!!