Aplicación de Deep Learning a la reducción del PAPR en OFDM

Abstract

Abstract: This work addresses the problem inherent to the high Peak to Average Power Rate (PAPR) in OFDM systems, a condition that induces signal distortions, reducing energy efficiency and scaling the operational complexity of the system. In response to this challenge, a novel solution is proposed that combines Deep Learning (DL) with the technique clipping, aimed at mitigating PAPR without sacrificing signal integrity. The research is developed through a structured methodology that begins with a Detailed literature review to identify DL applications that reduce PAPR in OFDM systems. Based on this analysis, a combined DL and clipping, carefully choosing data and test scenarios for evaluation. This innovative strategy allows the formulation of an effective technique for the minimization of PAPR, facilitating in turn the comparison with traditional methodologies. During the course of this titration work, a decrease in PAPR was observed. However, it is observed that the Bit Error Rate (BER) obtained is similar to the obtained by the conventional clipping method. Consequently, although significant advances in PAPR management and a competitive BER is maintained, parallel to that observed in clipping, the potential of incorporating network techniques is highlighted neurons within OFDM signal processing.