Foto del docente

Marco Crescentini

Professore associato

Dipartimento di Ingegneria dell'Energia Elettrica e dell'Informazione "Guglielmo Marconi"

Settore scientifico disciplinare: ING-INF/07 MISURE ELETTRICHE E ELETTRONICHE


TESI DISPONIBILE: FPGA Implementation of DPD Models Based on Machine Learning

Digital predistortion (DPD) is a signal processing method applied to wireless communication systems,
with the primary objective of enhancing the linearity of power amplifiers (PAs). PAs generally show
nonlinearity issues that introduce distortions into the amplified signals, especially when they are operated in an efficient operating point. These distortions result in spectral expansion, reduced signal
fidelity, and interference with neighboring channels. The purpose of DPD is to compensate for the
nonlinearities in the PA by applying an additional distortion to the input signal, so to obtain a linearized behavior of the whole signal chain.
The objective of the thesis project is to exploit machine learning approaches for DPD targeting beam-forming PA arrays. The DPD model and linearization procedure will be implemented using the AMD
Zynq UltraScale+ RFSoC ZCU111 Evaluation Kit (Figure 1). The board provides an FPGA for the
numerical implementation of the DPD algorithms, and it is equipped with eight 12-bit RF-ADC with
a sampling rate up to 4.096 GSa/s, and eight 14-bit RF-DAC with a sampling rate up to 6.554 GSa/s.

he preliminary part of the project concerns single-input single-output DPD models and the corresponding development in MATLAB for use with an off-the-shelf PA (Figure 2). Then, the algorithms
will be mapped into the FPGA. As a subsequent step, the DPD models should be extended to target
multiple-input multiple-output PA devices and beamforming PA arrays. This will necessitate more
complex DPD architectures and the exploitation of machine learning techniques. The final part of the
project concerns the experimental evaluation of the linearity performance as well as of the complexity and resource consumption of the proposed techniques. If implemented successfully, the project is
organized so to allow for the publication of the results in one scientific article or conference proceedings.

Pubblicato il: 03 ottobre 2023