Efficient solid state power amplifiers: power combining and highly accurate AM/AM and AM/PM behavioural models with application to linearisation

Abstract

Radio Frequency (RF) Power Amplifiers (PAs) are a major contributor to modern communication systems, both in terms of being an enabling technology as well as having the most impact on overall system availability, linearity and power consumption. In order to achieve the most optimum system outcome there needs to be an appropriate method for selecting the most suitable RF PA design approach, as well as being able to select the most appropriate RF PA output device, based on a range of varying requirements, specifications and technologies. The ability to perform these tasks quickly, with improved accuracy, using existing available device data, with minimal or no further device testing and from a range of existing and emerging technologies would provide RF PA designers with significant benefits. The investigations and research provided in this thesis consider a range of existing and emerging RF PA technologies and power combining methods and compares them via a new selection and design methodology developed in this thesis. The new methodology builds on modern design and statistical approaches including manufacturing options that enable an appropriate technology to be selected for Solid State Power Amplifier (SSPA) design. In addition to hard design specifications, the current thesis also considers less tangible specifications, such as graceful degradation, time tomarket and ease of use, as well as alternative design approaches, such as fuzzy logic approaches. With a suitable technology approach determined, a selection of a suitable RF output device(s) is considered. As the demand for new communication services continues to increase, requiring tighter specifications and reduced product delivery time scales, then the ability to accurately and quickly compare available RF PA devices from a range of device technologies or devices from different manufacturers, at both the system and component level, makes such a selection paramount. In this thesis, simplememoryless (AmplitudeModulation/AmplitudeModulation (AM/AM) only) and Quasi-Memoryless (QM) Behavioural Models (BMs) (AM/AM combined with Amplitude Modulation/Phase Modulation (AM/PM)) are reviewed, extended and improved upon, with up to 20 dB Normalised Mean Squared Error (NMSE) modelling improvement achieved over a range of technologies, allowing effective RF PA device selection using these newly developed simple and fast models. This thesis uses recent existing accurate and powerful semi-physical memoryless BMs, suited to RF PA devices, and develops and extends their use for QM modelling. The trade-off from the improvement in the overall accuracy is some further simple processing steps. Furthermore, this thesis also provides a comparison of other models, presented in the literature. The improved simple RF PA device models and extension techniques presented in this thesis show, via simulation and measurement, that the new models are suitable for use over a wide range. Lineariser improvements, linked to the accuracy improvements of the proposed models of this thesis, are also investigated, showing further benefits from this research. Physically based simple QM BMs are also used to model thermal and bias network memory effects, which are becomingmore relevant tomodern communication services that use wider bandwidths, enabling the impacts of RF PA device memory effects to be determined and compared. The feasibility of the developed models and improvements are also utilised in the simulation of a low cost RF PA lineariser. With the trend to smaller localised low cost and power RF mobile wireless repeater cells being away from larger more expensive and complex hardware, used to perform linearisation, this thesis presents a trade-off between complexity and linearisation performance and demonstrates, through modelling and simulation, that 8-10 dB improvement in linearisation performance is achievable with the use of the newly developed models.