I need to sense a differential voltage of 2000mV and the source is very noisy.
I think to use a Intrumentation amplifier, but I have two limits. The first is the supply voltage, I have 12V or less and 0V, not negative voltage. And the second one, is that I need to read this voltage with a DSP(3.3V).
I though to add a 1.65V offset voltage, so I will be in the middle of the voltage range.
A differential amp will only help reduce noise if the noise is common-mode.
The circuit you posted will not work near was well as a dedicated IC instrumentation amp due to the difficulty of matching resistors in a discrete circuit. Take a look at the AD732. It's a rail-rail device with good specs. The output can be offset by connecting the desired 1.65V offset voltage to the REF input on the amp.
If you can't get a rail-rail device then you would need to generate a negative voltage and use the AD620. You could use a small switching regulator or switched-cap converter to generate the negative voltage but you would need to be very careful about layout, grounding, and decoupling to keep the switch noise from the converter out of your signal.
Edit: What is the common-mode voltage of the differential signal?
200V?? A standard IA can't begin to handle that voltage, they can only go to the supply voltage at best. You need a high voltage difference amp such as this.
What is the lowest signal frequency you need to see?
Below is the sim of the classic 3 opamp instrumentation amp using a quad LM324 (similar to the dual LM358).
Rgain can be left out for a gain of 1.
R9 and R10 generate the output offset voltage (about 1.66V here).
Note that for best common-mode rejection all the 10k resistors should be on a single matched resistive network.
- - - Updated - - -
Below is the circuit with an added buffer for the Ref input to improve the common-mode rejection.