| With the development of communication technology,many different structures and design theories of power amplifiers(PAs)have been proposed to either improve output efficiency or expand operation frequency bandwidth to meet different market demands.However,when the new theory and architecture need to be implemented into the actual circuit design and layout design,manual calculation and debugging become extremely complicated.Therefore,the optimization algorithms and optimization tools are embedded into the current electronic design automation(EDA)commercial software for circuit design engineers to use.However,PA design,as an active,multi-dimensional,and nonlinear complex problem,especially for PAs with multi-device complex structures such as Doherty PA,both the carrier PA and peak PA need to be designed,tuned and optimized,simultaneously,which means the matching network parameters need to be debugged repeatedly,thus,modern commercial EDA tools can no longer provide effective assistance to engineers.In addition,layout design and optimization are the most difficult steps in PA design.At present,there is no effective layout optimization method,and manual debugging relies heavily on the experience of engineers.Therefore,in order to shorten the design cycle and improve the design efficiency of PA,this thesis aims to study the design optimization scheme of high efficiency PA based on machine learning algorithm,and verify its feasibility,so as to establish the foundation for realizing the intelligent design of PA module.The main research contents of this paper include the following three aspects:First of all,aiming at the difficulty of optimizing the layout of PA,this thesis proposes an improved bayesian optimization algorithm to optimize the layout,and obtain the best parameters of the matching network in a small number of sample points,which can improve the efficiency of PA layout optimization.The optimization algorithm is designed and verified,and the layout of single device class AB PA is completed within 4 hours.The output power is higher than 40.93 d Bm in the 2GHz to3 GHz bandwidth,the PAE is higher than 61.73%.Compared with the EDA optimization tool,the layout optimization can be completed faster in the same time by using proposed improved bayesian optimization algorithm.The relevant results were published in IEEE Microwave and Wireless Components Letters.Secondly,aiming at the problem that the gradient algorithm in EDA software may fall into local optimal solution,this thesis presents a matching network optimization scheme based on the particle swarm optimization algorithm,and verifies it.In order to further improve the optimization efficiency,a cooperative optimization scheme based on the joint particle swarm optimization and bayesian optimization algorithm is proposed and extended to the design of Class AB broadband PA.The automatic optimization design process of PA based on Chebyshev low-pass topology is realized.In the 2GHz to 3GHz frequency band,a PA with a PAE higher than 65.46%,an output power higher than 40.23 d Bm and a gain higher than 10.23 d B is measured,the relevant achievements were published in IEEE MTT-S International Wireless Symposium(IWS)and IEEE Conference on Microwave and Millimeter Wave Technology(ICMMT),and the extended results were submitted in IEEE Transactions on Microwave Theory and Technologies.Finally,by comprehensively considering the relationship between carrier PA,peak PA and compensation line in traditional Doherty PA,this thesis achieves the automatic design of traditional Doherty PA based on the automatic design scheme of Class AB broadband PA and manual design experience,which saves more than 70% of the time compared with the direct optimization method of current commercial EDA software.As a result,a Doherty PA with 200 MHz bandwidth is designed and measured,the saturation output power is greater than 42.7d Bm,the saturation efficiency is higher than61.08%,and the 6d B back-off efficiency is higher than 50.8%.The relevant achievements are submitted in IEEE Microwave and Wireless Components Letters. |
