Disclaimer: I've only built one "true" sigma-delta DAC, and it was first-order because I was running it on a 25MHz FPGA and I only needed telephone-quality audio out of it.  So rather than spending my clients money to have fun, I stuck with the basics.

There doesn't _need_ to be much difference between the analog/digital sigma-delta modulator you might build for an ADC vs. the all-digital one you'd make for a DAC.

Both have a modulating element that takes a continuous (in the case of the ADC) or many-valued (in the case of the DAC) signal from a loop filter and, at each sample instant of the loop\*, turns it into a 0 or a 1.  Both translate that 0 or 1 to match the full scale of expected input values (with a carefully buffered digital output in the case of an ADC, and just a multiply or change of attitude in the case of the DAC).  Both have a loop filter that conditions the output of the summing junction to perform the noise shaping that makes a sigma-delta modulator worthwhile.

Here's a chart that sums up the differences between the elements -- all of which are present in either type of sigma-delta device:

| function | ADC | DAC |
|----------|-----|-----|
| modulating element | analog comparator | compare operation  |
| modulator output   | digital output    | multiply operation |
| summing junction   | analog circuit    | subtraction operation |
| loop filter        | analog filter     | digital filter |

----

\* Note that the sampling rate of that binary sampler needs to be significantly higher than the "normal" sample rate of the desired signal.