NES2Net : A Number Estimation and Signal Separation Network for Single-Channel Blind Signal Separation

Single-channel blind signal separation (SCBSS) is a critical challenge in signal processing, aiming to recover original signal components from a single mixed signal. This technology has broad applications across various scenarios. However, existing methods often require excessive prior information on source signals, limiting their effectiveness in practical applications. To address this problem, we introduce a deep learning network called Number Estimation and Signal Separation Network (NES
²
Net). It is designed for efficient SCBSS where only the mixed signal is known at the receiving end. Our approach includes a Number Estimation Module (NEM) and a Signal Separation Module (SSM). NEM estimates the number of source signals in the mix signal and passes this number to SSM. SSM progressively separates the mixed signal by iteratively applying a U-Net structure. Experimental results and comparative studies demonstrate the superiority of our algorithm.