A High-Speed Surrogate Modeling Approach for Reconfigurable Multi-Band UWB LNAs in CMOS Technology
Résumé
Ultrawideband (UWB) technology has taken center stage in the wireless communication industry for its outstanding bandwidth and strong resistance to multipath fading. In UWB receiver designs, the Low-Noise Amplifier (LNA) is a crucial element that requires a delicate trade-off among sensitivity, gain, and linearity across a wide frequency range. Traditional LNA design flows rely on iterative high-fidelity simulations that are computationally intensive and time-consuming, particularly when exploring the large-dimensional design space of modern Complementary Metal-Oxide-Semiconductor (CMOS) technology. These existing optimization-based approaches often struggle with the complex non-linearity between geometric parameters and Radio Frequency (RF) performance, creating a trial-and-error bottleneck that slows time-to-market. In this paper, we propose a new design flow for a reconfigurable, Multi-Band (MB)-UWB LNA optimized using three deep learning surrogate models: a Deep Neural Network (DNN), a Recurrent Neural Network (RNN), and a one-dimensional Convolutional Neural Network (1D-CNN). By optimizing these surrogate models on high-fidelity Advanced Design System (ADS) simulation data, the design cycle is significantly shortened. The suggested LNA capitalizes on a reconfigurable current-reuse topology, thereby retaining frequency agility across the 3.5 GHz and 4.5 GHz bands. The experimental validation proves that the deep learning models provide high predictive accuracy, and the mean relative error of the predictions varies between 0.23% and 0.3%. Notably, the 1D-CNN is particularly effective at capturing local Metal-Oxide-Semiconductor Field-Effect Transistor (MOSFET)-sizing features, whereas all three models, on average, demonstrate a significant reduction in computational effort compared to classical simulation-based methods.
Citer ce document
Accès au document
Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter
Voir l'article sur le site de la revueAuteur(s)
Statistiques
Consultations : 1
Téléchargements : 0