Memory Effect Modeling of Wideband Wireless Transmitters Using Neural Networks

In this paper a three-layer real-valued time-delayed neural network (RVTDNN) is employed to simulate the memory effects of a wideband wireless transmitter. The RVTDNN is trained at first using a Matlab program, and then it is implemented in Agilent Advanced Design System software. Different training algorithms have been applied to the neural network to extract its weights and biases, and it is found that the Levenberg-Marquardt (LM) algorithm exhibits the best performance. A look-up-table based memoryless predistorter is cascaded to the RVTDNN model to validate the capability of the RVTDNN model in simulating the memory effects of the transmitter. The validation results demonstrate that the identified RVTDNN model can accurately mimic the memory effects of a wideband wireless transmitter prototype, which is based on a 60-watt push-pull GaAs FET power amplifier, under a two-carrier 3GPP-FDD excitation signal.