TY - GEN
T1 - Mixture of Experts Neural Network for Modeling of Power Amplifiers
AU - Fischer-Bühner, Arne
AU - Brihuega, Alberto
AU - Anttila, Lauri
AU - Gomony, Manil Dev
AU - Valkama, Mikko
N1 - Funding Information:
This work has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No. 860921. The work was also supported by the Academy of Finland under the grants #319994, #338224, and #332361
Publisher Copyright:
© 2022 IEEE.
JUFOID=70577
PY - 2022
Y1 - 2022
N2 - A new Mixture of Experts Neural Network (ME-NN) approach is described and proposed for modeling of nonlinear RF power amplifiers (PAs). The proposed ME-NN is compared with various piece-wise polynomial models and the time-delay neural network (TDNN) regarding their ability to scale in terms of modeling accuracy and parameter count. To this end, measurements with GaN Doherty PA at 1.8 GHz and a load modulated balanced (LMBA) PA operating at 2.1 GHz with strong nonlinear behavior and dynamics are employed, assessing the potential benefits of ME-NN over the existing models. Implementation-related advantages of the proposed ME-NN over TDNNs at increasing network sizes are furthermore discussed. The measurement results show that the ME-NN approach offers increased modeling accuracy, particularly in the LMBA PA case, compared to the existing reference methods.
AB - A new Mixture of Experts Neural Network (ME-NN) approach is described and proposed for modeling of nonlinear RF power amplifiers (PAs). The proposed ME-NN is compared with various piece-wise polynomial models and the time-delay neural network (TDNN) regarding their ability to scale in terms of modeling accuracy and parameter count. To this end, measurements with GaN Doherty PA at 1.8 GHz and a load modulated balanced (LMBA) PA operating at 2.1 GHz with strong nonlinear behavior and dynamics are employed, assessing the potential benefits of ME-NN over the existing models. Implementation-related advantages of the proposed ME-NN over TDNNs at increasing network sizes are furthermore discussed. The measurement results show that the ME-NN approach offers increased modeling accuracy, particularly in the LMBA PA case, compared to the existing reference methods.
KW - 5G and beyond
KW - behavioral modeling
KW - digital predistortion
KW - neural network
KW - nonlinear distortion
KW - power amplifier
U2 - 10.1109/IMS37962.2022.9865530
DO - 10.1109/IMS37962.2022.9865530
M3 - Conference contribution
AN - SCOPUS:85137975750
SN - 9781665496148
T3 - IEEE MTT-S International Microwave Symposium Digest
SP - 510
EP - 513
BT - 2022 IEEE/MTT-S International Microwave Symposium, IMS 2022
PB - IEEE
T2 - IEEE/MTT-S International Microwave Symposium
Y2 - 19 June 2022 through 24 June 2022
ER -