Robust verification of look-up-table-based models for all-digital RF-transmitters

The efficient development of all-digital RF-transmitters (DRFTx) requires models that can capture the memory induced, nonlinear behavior of the circuitry. Broadband time-domain models work well for this application, although, gaining dependable model prediction errors from verification measurements is difficult. For the presented DRFTx, the model predicts output signals over the full bandwidth (DC-20 GHz) of the used active device. This makes it very sensitive to synchronization errors. This work illustrates that by comparing the estimated prediction errors of such models, calculated during model training, to measurements of the same DRFTx test setup when used as LTE transmitter. It is shown that the impact of residual synchronization errors on the computed verification error is strongly dependent on the method utilized to compute it. It is also discussed which limitations in the measurement setup cause errors and how they can be considered by the error analysis. Finally, it is shown that time-frequency methods offer an elegant way of calculating more robust verification errors for such systems.