A fast engineering approach to high efficiency power amplifier linearization for avionics applications
- Autores: Jui Teng Wang
- Directores de la Tesis: Pere Lluis Gilabert Pinal (dir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2020
- Idioma: español
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
This PhD thesis provides a fast engineering approach to the design of digital predistortion (DPD) linearizers from several perspectives: i) enhancing the off-line training performance of open-loop DPD, ii) providing robustness and reducing the computational complexity of the parameters identification subsystem and, iii) importing machine learning techniques to favor the automatic tuning of power amplifiers (PAs) and DPD linearizers with several free-parameters to maximize power efficiency while meeting the linearity specifications. One of the essential parts of unmanned aerial vehicles (UAV) is the avionics, being the radio control one of the earliest avionics present in the UAV. Unlike the control signal, for transferring user data (such as images, video, etc.) real-time from the drone to the ground station, large transmission rates are required. The PA is a key element in the transmitter chain to guarantee the data transmission (video, photo, etc.) over a long range from the ground station. The more linear output power, the better the coverage or alternatively, with the same coverage, better SNR allows the use of high-order modulation schemes and thus higher transmission rates are achieved. In the context of UAV wireless communications, the power consumption, size and weight of the payload is of significant importance. Therefore, the PA design has to take into account the compromise among bandwidth, output power, linearity and power efficiency (very critical in battery-supplied devices). The PA can be designed to maximize its power efficiency or its linearity, but not both. Therefore, a way to deal with this inherent trade-off is to design high efficient amplification topologies and let the PA linearizers take care of the linearity requirements. Among the linearizers, DPD linearization is the preferred solution to both academia and industry, for its high flexibility and linearization performance. In order to save as many computational and power resources as possible, the implementation of an open-loop DPD results a very attractive solution for UAV applications. This thesis contributes to the PA linearization, especially on off-line training for open-loop DPD, by presenting two different methods for reducing the design and operating costs of an open-loop DPD, based on the analysis of the DPD function. The first method focuses on the input domain analysis, proposing mesh-selecting (MeS) methods to accurately select the proper samples for a computationally efficient DPD parameter estimation. Focusing in the MeS method with better performance, the memory I-Q MeS method is combined with feature extraction dimensionality reduction technique to allow a computational complexity reduction in the identification subsystem by a factor of 65, in comparison to using the classical QR-LS solver and consecutive samples selection. In addition, the memory I-Q MeS method has been proved to be of crucial interest when training artificial neural networks (ANN) for DPD purposes, by significantly reducing the ANN training time. The second method involves the use of machine learning techniques in the DPD design procedure to enlarge the capacity of the DPD algorithm when considering a high number of free parameters to tune. On the one hand, the adaLIPO global optimization algorithm is used to find the best parameter configuration of a generalized memory polynomial behavioral model for DPD. On the other hand, a methodology to conduct a global optimization search is proposed to find the optimum values of a set of key circuit and system level parameters, that properly combined with DPD linearization and crest factor reduction techniques, can exploit at best dual-input PAs in terms of maximizing power efficiency along wide bandwidths while being compliant with the linearity specifications. The advantages of these proposed techniques have been validated through experimental tests and the obtained results are analyzed and discussed along this thesis.
