Evaluation of Dual Input Models for RF Power Amplifier Arrays Under Antenna Crosstalk

Abstract

The current demand for higher spectral efficiency and data rates in 5G and in emerging 6G technologies has resulted in wide use of multi-antenna architectures in wireless communication systems. These architectures, with tightly packed antenna elements introduce antenna crosstalk to the transmitter chain. The resulting antenna crosstalk introduces a secondary input into the power amplifier (PA) from the PA output end. This reflected coupled signal elevates the nonlinear behaviour of PA which ultimately degrades the overall system performance. Accurate behavioural modelling is essential in these scenarios to employ linearization techniques such as digital predistortion (DPD). This thesis implements and evaluates several behavioural models derived from memory polynomial (MP) framework proposed from literature related to prior works. One single-input model and four dual-input models are implemented using measured multi-antenna datasets with distinct PA characteristics. Each model is then analysed in terms of performance, complexity, their ability to capture nonlinearities and significance of nonlinearities due to antenna crosstalk. Results demonstrated that the models which considered independent memory depths for dual inputs and capture higher nonlinear orders, achieve the best performance while simplified dual input models with same memory and moderate nonlinear contribution provide an optimal balance between the performance and complexity. Such models with both independent memory depths and higher order nonlinearity consideration, were able to reach normalised mean square error (NMSE) of -44.62 dB while simplified models were able to reach NMSE of -42.10 dB. Another major observation was that, the first order β term represented the linear contribution of the coupled input to the PA output and does not capture all nonlinear effects induced by antenna coupling. As higher order terms such as higher order β terms, and other PA coefficients of γ and δ are incorporated, the dual input models become increasingly capable of representing nonlinear behaviours between intended and coupled inputs, resulting in improved modelling accuracy. The findings of this study can be used to provide practical insights of behavioural modelling approaches for linearization techniques like DPD. As future work, the models evaluated in this study can be applied for DPD frameworks and also the study can be further extended by including novel approaches of PA behavioural modelling as well.