Could you recommend a 2-channel ADC for an electronic load?

Question

Dear Precision ADC Forum, 
 I'm designing an electronic load for my lab. The load voltage is sensed with a standard resistor divider, and the load current is sensed across a shunt resistor so that the voltage across it can be sensed in single-ended fashion. I have LT6230-based amplifiers for each sensed value to adapt them to the inputs of an ADC. My system is limited to a maximum of 100 ksps. It would be easiest for to power the ADC's with 5V, but I can generate other voltages if needed. 
 There are so many different two-channel ADC's available that I'm feeling overwhelmed. I'm not sure if a simultaneous sampling device is worthwhile, and I don't know how many bits are truly needed. I also don't know if a differential input ADC is worthwhile, even though I would use it as single-ended (GND-referenced). 
 If it helps, I would like to use a 16-bit DAC, because the maximum load voltage is 430VDC and the maximum current is 30.0A. As I understand, this would allow me to program with precisions of: 
 430V / 2^16 = 6.6 mV 
 30A / 2^16 = 0.45 mA 
 Many thanks, 
 Chris

@skowalik · Answer

Chris, 
 Given you have already decided on an amplifier and signal conditioning topology I would first recommend the AD7902 as a start. You will have to generate a 2.5V rail for the ADC core but otherwise you can run the signal from 0 ->5V assuming you have a 5V reference. The part is specified at 1MSPS but you can easily run the converters at 1/10th speed with no issues. What's nice about this part is that it gives you the flexibility to decide later if you want to run the two converters independently or simultaneous sample. 
 If you were interested in a direct interface part you may want to look at the LTC2353-16. The downside (potentially) of this option is you'll have to generate bipolar supply rails for the input but in the long run this might be the better solution for you. 
 Sean

@skowalik · Answer

Chris, 
 With respect to the physical implementation of a differential signal path the generic answer is that to the extent you keep the path differential to your sensor, the topology provides natural immunity to common mode disturbances and thus has advantages over a single-ended data path. The benefit is of course limited by the common mode rejection specification of the components in your signal chain. If you are able to directly connect to the LTC2353 you can expect >120dB of rejection up to about 1KHz, > ~90dB out to about 100KHZ. 
 If you then take advantage of the correct SoftSpan setting for you sensor output you can maximize the system SNR. So for example if we consider your voltage measurement and assume you're using a simple resistive divider network to attenuate your 430V into the 10.24V unipolar range then the system SINAD (excluding the divider noise) is high enough to provide and ENOB = 14.5 bits. Adding oversampling can increase your ENOB further and allow you to get to a full "noise-free" 16 bits if that is what you require which I think is possible with an OSR of 8. So assuming you run at 100KSPS/8 = 12.5KSPS (6.25KHZ BW). 
 Accuracy should not be effected by the input topology and will be a function of the underlying components chosen to implement the signal chain. 
 Sean