A novel approach to noise reduction in eLoran signal based on 1D-CycleGAN

Abstract

Enhanced Loran (eLoran) system is a ground-based navigation system that plays a vital role in both strategic and civilian timing applications. However, the eLoran signal in real-world use is often noisy and distorted, making it hard to decode accurately. This paper presents a novel eLoran signal noise reduction algorithm based on Cycle-Consistent Generative Adversarial Networks (CycleGAN). The algorithm capitalizes on the inherent relationship between signals and images, applying image-based denoising techniques to a one-dimensional signal. CycleGAN includes two generators and two discriminators. It utilizes both noisy and clean signals to achieve noise reduction. This paper also uses one-dimensional convolutional neural network (lDCNNs) and dense residual blocks to enhance the network's performance. Experimental results demonstrate that the proposed algorithm performs well on analog signals with a signal-to-noise ratio (SNR) above — 7 dB, enhancing the SNR to over 17.54 dB while maintaining an information accuracy rate of above 98.96%. On analog data sets, this algorithm improved the SNR by an average of 22.03 dB. Furthermore, when tested with measured eLoran signals, the model not only eliminated signal distortion but also increased the SNR of a measured signal from 7.26 to 22.37 dB, achieving an information accuracy rate of over 97.91%.