Volume 17, Number 3

Simulated Annealing-Salp Swarm Algorithm based Variational Autoencoder for Peakto-Average Power Ratio Reduction

  Authors

Prabhakar Narasappa Kota ¹, Pravin Balaso Chopade ¹, Bhagvat D Jadhav ², Shriram Sadashiv Kulkarni ³ and Pravin Marotrao Ghate ², ¹ M. E. S. Wadia College of Engineering, India, ² JSPM's Rajarshi Shahu College of Engineering, India, ³ Sinhgad Academy of Engineering, India

  Abstract

Orthogonal Frequency Division Multiplexing (OFDM) have achieved significant advancements in spectral effectiveness and data rates within wireless communication systems. However, it is accompanied by a critical challenge: the high Peak-to-Average Power Ratio (PAPR). This issue demands attention and effective solutions to ensure optimum performance and reliability of OFDM-based systems. Currently, deep Learning (DL) algorithms perform well on end-to-end wireless communication systems. This study introduces a novel approach to PAPR reduction in OFDM systems by integrating a Simulated AnnealingSalp Swarm Algorithm (SA-SSA) with a Variational AutoEncoder (VAE). The proposed method effectively mitigates peaks while preserving favorable spectral properties, thereby facilitating seamless PAPR migration. The SA-SSA - based VAE method is used to develop a peak-canceling signal method depending on the input signal which reduces the PAPR signal. Constellation mapping and remapping of symbols are considered in each subcarrier of the VAE method that minimizes the Bit Error Rate (BER) and PAPR in OFDM systems. To further improve the performance of VAE, proposed an SA-SSA algorithm that tuned the hyperparameters of the VAE method to select optimum hyperparameters of VAE for better performance. The performance of the developed method is analyzed with characteristics of BER, Symbol Error Rate (SER), and Complementary Cumulative Distribution Function (CCDF) under various subcarriers. The proposed method obtained less PAPR of 1.9 dB, 2.0 dB, 2.4 dB, 2.8 dB, and 3.2 dB for 64, 128, 256, 512, and 1024 subcarriers which is less when compared to existing methods like Hyperparameter Tuned Deep Learning based Stacked Sparse Autoencoder (HPT-SSAE and Conditionally Applied Neural Network (CANN).

  Keywords

Deep Learning, Orthogonal Frequency Division Multiplexing, Peak-to-Average Power Ratio, Simulated Annealing-Salp Swarm Algorithm, Variational AutoEncoder