This paper presents a neural network approach for modeling nonlinear memoryless communication channels. In particular, the paper studies the approximation of the nonlinear characteristics of traveling-wave tube (TWT) amplifiers used in satellite communications. The modeling is based upon multilayer neural networks, trained by the odd and even backpropagation (BP) algorithms. Simulation results demonstrate that neural network models fit the experimental data better than classical analytical TWT models. Index Terms-Neural networks, satellite communications, TWT amplifiers. I. INTRODUCTION S EVERAL nonlinear channels (e.g., satellite communication channels) [1] are equipped with memoryless nonlinear devices such as traveling-wave tube (TWT) amplifiers. These devices exhibit two kinds of nonlinearities, amplitude distortion (AM/AM conversion) and phase distortion (AM/PM conversion). Two equivalent frequency-independent representations have been proposed for these nonlinearities [1], [12]. These representations are amplitude-phase (A-P) and in-phase and quadrature (I-Q). Several analytical models such as Bessel and rational functions have been proposed for these nonlinear distortions. This paper adaptively models these nonlinear functions using the odd and even backpropagation (BP) algorithms [9]. These supervised learning procedures use the measured TWT input and output signals for iteratively adjusting the neural network weights. The simulation results indicate that the neural net approach performs better than classical approximation techniques. Furthermore, a change in the TWT characteristic can be tracked by the neural net since the net is adaptive. This yields a new model which has adapted to this change. The neural net can approximate all types of TWT's with the same basic structure (only the weights change). The neural architecture has few parameters (e.g., only 15 scalar parameters are needed to model the AM/AM conversion of an Intelsat TWT amplifier). Moreover, this architecture can be implemented rapidly in parallel for both the learning and generalization phases. This paper focuses Paper approved by P. H. Wittke, the Editor for Communication Theory of the IEEE Communications Society.

No abstract