Wireless Signal Denoising Using Conditional Generative Adversarial Networks

Abstract

Wireless signal strength plays a critical role in wireless security. For example, we can intentionally reduce transmission power at a transmitter to prevent eavesdropping. Later the receiver will employ signal denoising techniques to enhance the signal-to-noise ratio. Also, signals are deteriorated by noise and interference during transmission. Therefore, wireless signal enhancement or denoising is a critical challenge. This paper tackles this challenge and investigates an adversarial learning-based approach for wireless signal denoising, which will correspondingly enhance signal strength. Specifically, we design a conditional generative adversarial network at the receiver to establish an adversarial game between a generator and a discriminator. The generator receives the noisy signal and aims to generate the denoised signal, while the discriminator aims to force the denoised signal to match the noisy signal exactly. Unlike traditional signal denoising methods that estimate the noise or interference in the noisy signals, our proposed method estimates and learns the features of real noise-free signals, which is more adaptive to dynamic wireless communication environments. We conduct simulations on signals with four different modulations to evaluate the performance. The results demonstrate that our method can generate denoised signals effectively and outperforms other traditional methods.