Automated Two-Step Power Amplifier Design with Pre-constructed Artificial Neural Network


Kouhalvandi L., Pirola M., Ozoguz S.

43rd International Conference on Telecommunications and Signal Processing (TSP), ELECTR NETWORK, 7 - 09 July 2020, pp.617-620 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume:
  • Doi Number: 10.1109/tsp49548.2020.9163468
  • Country: ELECTR NETWORK
  • Page Numbers: pp.617-620

Abstract

Power amplifier (PA) designs at high frequency are not straightforward and they depend on designers' experience by dealing with a high number of parameters to be set. To address PA design problems, we propose an automated bottom-up method based on an artificial neural network (ANN) to be employed in the optimization-oriented strategy. The proposed methodology starts with a PA based on lumped elements (LEs), then ANN is trained for characterizing the lumped element PA and finally a PA with distributed components, the natural environment at high frequency, is designed by using a bottom-up method and the constructed ANN. In this way, the resulting distributed element PA inherits the advantages of the lumped element design, i.e. offering higher and flatter gain performance. To validate our method, we design 10 W PAs in band frequency of 1 GHz to 2 GHz (L band). The automated design of PA with transmission lines (TLs) results in gain between 10-13 dB and power added efficiency larger than 50%. Our results demonstrate the robustness of the presented approach adopting ANN in designing PAs, automatically.