The former can yield a global optimal solution of the
power minimization problem, while the latter can only yield
a local optimal solution due to conservative successive convex
approximations performed at each iteration, and a rigorous
analysis on the upper bounds of step sizes is also proposed
to guarantee their convergence.