If i is a CVX variable or expression, that constraint is non-convex.
However, you may be able to model it (actually with >=, not >) in CVX by using binary variables, using CVX’s MIDCP capability. If i ;is real, that’s just abs(i). So use the Big M model for abs in Logics and integer-programming representations - YALMIP, which you need to use because you are using abs in a non-convex way (inequality is going the wrong direction to be convex).
If iis complex, you might be out of luck using CVX or any convex optimization tool. But if someone knows how to do it, please post. Well, maybe there’s a messy way of linearizing it per component.