Hi superleaf,
the circuit you have shown is a special kind of multi-loop feedback.
If two loops have (at least) one common path, both margins are to be measured/simulated - and the smallest one determines the margin of the circuit.
In your case (it looks as) both loops are independent of each other. Thus - because they are identical - they also have the same loop gain and the same margin. That means, it is sufficient to open only one loop in order to insert a test voltage.
However, this may create some problems because the opamp (I assume it is a vltage-in/voltage-out amp) has no resistive feedback to fix a suitable operating point.
I think, that was the reason for the problems you have encountered. Because of the capacitive feedback I would suggest to simulate the dc transfer curve in order to find the actual offset voltage (for one of the outputs).
Then, apply this offset as a dc input voltage. Now you can open the loop at the output node, inject a test voltage (of 1 volt ac) into the feedback path and determine the output ac voltage. The input without the dc correction voltage has to ge grounded during this test.
This gives you a good approximation for the loop gain response. It is an approximation only (with high frequency errors) because the opened loop does not load the opamp output anymore. But in case of a voltage output this is only a minor problem.
Good luck.