کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Design and Implementation of a Neural Network-based Digital Predistortion Unit for RF Power Amplifiers

نویسندگان :

Mohammad Samiee Esfahani¹ Rasoul Dehghani² Sayed Vahid Mir-moghtadaei³

1- دانشگاه صنعتی اصفهان 2- دانشگاه صنعتی اصفهان 3- دانشگاه شهرکرد

کلمات کلیدی :

Digital Predistortion،Deep Learning،Power Amplifier،Nonlinear Modeling،Neural Networks

چکیده :

The impressive growth in the use of portable devices, such as laptops and mobile phones, has led to a significant expansion in the telecommunications industry in recent decades. This growth requires the development of high-efficiency systems with low power consumption, especially in the design of power amplifiers (PAs), which are power-hungry components. However, high-efficiency PAs usually operate closer to the PA's saturation, introducing strong non-linear distortions that degrade throughput and increase error rates. These effects are more apparent in modern wideband communication systems utilizing orthogonal frequency-division multiplexing (OFDM) and high-order QAM constellations, where the high peak-to-average power ratios (PAPR) require advanced linearization techniques. This work presents the design and implementation of a digital predistortion (DPD) unit for RF PAs using neural networks and deep learning. Experimental evaluations based on measured RF WebLab data demonstrate that the proposed approach improves PA performance, achieving more than 17.5 dB reduction in adjacent channel leakage and enhancing linearity and transmission quality.