AD5934(EVAL-CN0349) output frequency problem

The chart is probably a good starting point, but do your own research of the matter. As your low-end frequency becomes lower the “DFT” artifacts start appearing in the data and the lower the frequency the greater the systematic error they cause (generally speaking). It is a continuum, and where those errors cross the acceptable threshold depends on your application. Personally I find the first line in the chart should be 100 KHz – 10 KHz, other ranges are OK, but probably not optimal for your specific application. You can figure out those ranges by observing the error similarly to what you have measured so far.

If you need your microcontroller and the measurement circuit to share the power supply, the ADuM1250 are not needed at all – useful only if your measurement circuit must be galvanicly decoupled from the rest of the system, for example, if you measure the impedance of a human and want to make sure one does not get electrocuted in the unlikely event your power supply fails. You might need to be a bit more careful with your design to avoid ground loops between the microcontroller and the measurement circuit, so that the latter does not pick up excessive digital noise, but that alone hardly justifies the use of the decouples such as ADuM1250.

The DS1077 was just a functional suggestion; there are similar parts that work with 3.3V: DS1085L for example. Using too low MCLK frequency does not make much sense unless you need really low-frequency excitation voltage – otherwise the digital approximation of the sine wave is too crude. Just get rid of U4 and U5 and feed the MCLK input from your microcontroller PWM (for 2 KHz excitation you probably still need something like 1-4 MHz to keep the excitation sine wave reasonably accurate).

The chart is probably a good starting point, but do your own research of the matter. As your low-end frequency becomes lower the “DFT” artifacts start appearing in the data and the lower the frequency the greater the systematic error they cause (generally speaking). It is a continuum, and where those errors cross the acceptable threshold depends on your application. Personally I find the first line in the chart should be 100 KHz – 10 KHz, other ranges are OK, but probably not optimal for your specific application. You can figure out those ranges by observing the error similarly to what you have measured so far.

If you need your microcontroller and the measurement circuit to share the power supply, the ADuM1250 are not needed at all – useful only if your measurement circuit must be galvanicly decoupled from the rest of the system, for example, if you measure the impedance of a human and want to make sure one does not get electrocuted in the unlikely event your power supply fails. You might need to be a bit more careful with your design to avoid ground loops between the microcontroller and the measurement circuit, so that the latter does not pick up excessive digital noise, but that alone hardly justifies the use of the decouples such as ADuM1250.

The DS1077 was just a functional suggestion; there are similar parts that work with 3.3V: DS1085L for example. Using too low MCLK frequency does not make much sense unless you need really low-frequency excitation voltage – otherwise the digital approximation of the sine wave is too crude. Just get rid of U4 and U5 and feed the MCLK input from your microcontroller PWM (for 2 KHz excitation you probably still need something like 1-4 MHz to keep the excitation sine wave reasonably accurate).