Stability Analysis of Closed-loop Algorithms for Cancelling Nonlinear Memory Effects

In a nonlinear memory system, using a nonlinear adaptive filter as a predistorter is an effective compensation technique. Predistorters based on closed-loop algorithms typically exhibit better linearity than those based on indirect learning algorithms. However, the steady state of a closed-loop algorithm may be affected by external noise or error signals. In this paper, the long-term stability of a closed-loop algorithm is analyzed using perturbation and iteration analyses. It is shown that because of accumulation of historical errors, it is difficult for typical closed-loop algorithms to converge when the number of iterations becomes large. To address this problem, a modified closed-loop algorithm is proposed. By reducing the effect of accumulated historical errors, the long-term stability of the algorithm is largely improved, as verified by numerical simulations.