In Fig. 3, the convergence behavior of a random channel
realization is demonstrated. The left subfigure shows the
deviation of our distributed solution from the optimal centralized
solution. Similar to [12], the oscillation in primal
variables is dampened and diminishes as iteration goes due
to regularization. Therefore convergence can be reached when
the difference of total transmit power in successive iterations is
sufficiently small. The convergence is also visualized by the
right subfigure showing the diminishing Lyapunov function
(23) as iteration goes, while (23) has been modified properly
to include MUI-related terms. Note that (22) and (32)
are sufficient conditions for convergence. While in certain
channel realizations convergence can be observed with larger
stepsizes, (22) and (32) ensure convergence uniformly for all
cases. Hence the effectiveness and stability of our proposal is
demonstrated.