TY - GEN
T1 - A Bandwidth-Aware Figure of Merit for Behavioral Modeling of Power Amplifiers
AU - Pham, Thuy T.
AU - Pham, Dang Kièn Germain
AU - Mohellebi, Reda
AU - Almairac, Pierre
AU - Pedrosa, Carolina
AU - Desgreys, Patricia
N1 - Publisher Copyright:
©2026 European Union.
PY - 2026/1/1
Y1 - 2026/1/1
N2 - This paper introduces a bandwidth-aware figure of merit (FoM) that unifies four dimensions of model performance: average accuracy, prediction stability, bandwidth sensitivity, and model complexity. The FoM is formulated as a weighted sum of normalized indicators, enabling fair comparisons across both polynomial and neural-network-based models. A case study on measured 5G NR signals, spanning 10 bandwidths from 20 to 100 MHz, demonstrates the utility of the proposed FoM. Results show that the NARX neural network (NARXNN) achieves consistently better FoM scores than the generalized memory polynomial (GMP) and other neural baselines, reflecting superior robustness and efficiency under bandwidth scaling. Beyond bandwidth generalization, the framework is extensible to other dynamic conditions such as power scaling and carrier aggregation, making it a versatile benchmark for future RF modeling research.
AB - This paper introduces a bandwidth-aware figure of merit (FoM) that unifies four dimensions of model performance: average accuracy, prediction stability, bandwidth sensitivity, and model complexity. The FoM is formulated as a weighted sum of normalized indicators, enabling fair comparisons across both polynomial and neural-network-based models. A case study on measured 5G NR signals, spanning 10 bandwidths from 20 to 100 MHz, demonstrates the utility of the proposed FoM. Results show that the NARX neural network (NARXNN) achieves consistently better FoM scores than the generalized memory polynomial (GMP) and other neural baselines, reflecting superior robustness and efficiency under bandwidth scaling. Beyond bandwidth generalization, the framework is extensible to other dynamic conditions such as power scaling and carrier aggregation, making it a versatile benchmark for future RF modeling research.
KW - Figure of merit
KW - NARX neural networks
KW - bandwidth scalability
KW - behavioral modeling
KW - wideband power amplifier
UR - https://www.scopus.com/pages/publications/105036370997
U2 - 10.1109/LASCAS67804.2026.11457077
DO - 10.1109/LASCAS67804.2026.11457077
M3 - Conference contribution
AN - SCOPUS:105036370997
T3 - 2026 IEEE 17th Latin American Symposium on Circuits and Systems, LASCAS 2026 - Proceedings
BT - 2026 IEEE 17th Latin American Symposium on Circuits and Systems, LASCAS 2026 - Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 17th Latin American Symposium on Circuits and Systems, LASCAS 2026
Y2 - 24 February 2026 through 27 February 2026
ER -