1: \begin{abstract}

2: The convergence properties of the Iterative water-filling (IWF)

3: based algorithms (\cite{yu02}, \cite{scutari08a}, \cite{wang08})

4: have been derived in the {ideal situation} where the transmitters in

5: the network are able to obtain the \emph{exact} value of the

6: interference plus noise (IPN) experienced at the corresponding

7: receivers \emph{in each iteration} of the algorithm. However, these

8: algorithms are not robust because they diverge when there is {\it

9: time-varying} estimation error of the IPN, a situation that arises

10: in real communication system. In this correspondence, we propose an

11: algorithm that possesses convergence guarantees in the presence of

12: various forms of such time-varying error. Moreover, we also show by

13: simulation that in scenarios where the interference is strong, the

14: conventional IWF diverges while our proposed algorithm still

15: converges.

16: \end{abstract}

17: