An effective source number detection method for single-channel signals based on signal reconstruction and deep learning at low SNR

Abstract

Determining the number of sources under low signal-to-noise ratio (SNR) and signal interference with the same frequency and modulation presents a significant challenge. To address this challenge, we propose a novel method for detecting the number of signal sources from single-channel that leverages signal reconstruction and deep learning. The method employs subspace projection based on the Hankel matrix to reconstruct the measured single-channel signals, effectively suppressing noise. Furthermore, we incorporate the correlation of information and the integrity of feature in the signal, by fusing the in-phase component, quadrature component, and frequency spectrum feature of the reconstructed complex signal into a one-dimensional feature suitable for convolutional neural network (CNN). To address the source number detection task, we design a one-dimensional CNN based on convolutional block attention module, transforming it into a classification problem. Finally, experimental measurements demonstrate the effectiveness of our proposed method, with an detection accuracy of 94% even at an SNR of −10 dB.