Noise Reduction of Multi-carrier Symbols Using Generative Adversarial Network

The 6G radio access networks may be based on a multi-carrier system because it allows us to maximize spectrum utilization and effectively combat against frequency selective fading. In addition, Narrowband Internet of Things (NB-IoT) uses a multi-carrier modulation. 6G is expected to improve NB-IoT capabilities. However, it is vulnerable to noise and frequency offset. One of the key research challenges in 6G systems is to reduce noise in a multi-carrier system. Generative Adversarial Networks (GAN) is a useful machine learning technique to generate new data from the limited training data. It is possible to work for noise cancellation. In this paper, we propose a new noise reduction technique using GAN in a multi-carrier system and evaluate the performance in terms of Mean Squared Error (MSE). The GAN structure with two neural networks generator and discriminator for a multi-carrier system is trained by preamble symbols by solving minimax optimization problems. After having the trained GAN model, the trained generator of the GAN is regarded as a noise reduction filter and predicts the denoised data. Using computer simulation, we compare the performance between GAN and low pass filter and show us a better MSE performance of the proposed GAN noise reduction technique.