Multi-scale Feature Networks for Signal Noise Reduction

The modern wireless communication, the complexity and variability of background noise render traditional denoising methods inadequate for practical applications. To address this issue, a method to denoise at the base of a multi-scale characteristic network is proposed. This method, without any prior information, initially utilizes the nonlinear expressive power and feature extraction capabilities of deep neural networks to estimate background noise from multiple scales in a high-dimensional space. Subsequently, following the residual concept, it accomplishes denoising of the signal in this high-dimensional space and finally restores the signal to a low-dimensional space. Simulation results demonstrate the superiority of our denoising method over existing deep neural network-based denoising methods.