1: \begin{abstract}

2: Saddle point problems arise from many wireless applications, and primal-dual

3: iterative algorithms are widely applied to find the saddle points.

4: In the existing literature, the convergence results of such algorithms

5: are established assuming the problem specific parameters remain unchanged

6: during the iterations. However, this assumption is unrealistic in

7: time varying wireless systems, as explicit message passing is usually

8: involved in the iterations and the channel state information (CSI)

9: may change in a time scale comparable to the algorithm update period.

10: This paper investigates the convergence behavior and the tracking

11: error of primal-dual iterative algorithms under time varying CSI.

12: The convergence results are established by studying the stability

13: of an equivalent virtual dynamic system derived in the paper, and

14: the Lyapunov theory is applied for the stability analysis. We show

15: that the average tracking error is proportional to the time variation

16: rate of the CSI. Based on these analyses, we also derive an adaptive

17: primal-dual algorithm by introducing a compensation term to reduce

18: the tracking error under the time varying CSI.\end{abstract}

19: