Improved robustness and accelerated power amplifier identification with adaptive Wiener models in the complex domain

Identification of microwave power amplifiers including their memory effects, can be done computationally inexpensive with adaptive Wiener models which consist of a transversal filter followed by a static non-linearity that is parametrised using a polynomial basis. For identification of both blocks by simultaneously running gradient type algorithms, the resulting combined adaptive scheme suffers from low learning speed and vulnerability to instability. If the identification of the two blocks is done one after the other, the stability region is well defined but the achievable error does not even get near the minimum mean square error (MMSE). As a remedy, we propose a repetition of such a consecutive identification. By this, the algorithm allows to reach the MMSE, while stability is preserved. Based on recent results found for the real-valued case, we motivate the use of specific time-variant step-sizes enabling robust and fast simultaneous adaptation in the complex domain. By Monte Carlo simulations, we illustrate the improved performance that is obtained by the proposed algorithms.