Robust Noise Removal for 3D Radio Maps via Correntropy and Tensor Decomposition
Chunmei Li, Zheng Dou, Hang Jiang, Mingfang Li, Yongxin Cui, Haiyang Dong · 2025
Radio maps have become increasingly vital for various wireless communication applications. However, their generation and processing often face challenges from noise and outliers stemming from unstable spectrum sensors, signal interference, and other environmental factors. To address these issues, we propose robust correntropy-based tensor decomposition (RCTD), a robust noise removal method specifically designed for 3D radio maps, aimed at mitigating the adverse effects of non-Gaussian noise. RCTD utilizes a correntropy-based loss function to effectively decrease the impact of outliers compared to traditional loss functions like mean squared error. Additionally, it employs low-tubal-rank tensor decomposition to leverage the inherent low-rank characteristics of radio maps, offering improved computational efficiency over tensor nuclear norm approaches. To tackle the resulting optimization problem, we implement a half-quadratic optimization technique alongside an alternating minimization algorithm. Extensive numerical evaluations on both Gaussian mixture model noise and symmetric alpha-stable noise demonstrate that RCTD significantly outperforms state-of-the-art algorithms in the noise removal performance for 3D radio maps.