Research on Deep Learning-Based Single-Channel Blind Source Separation of Communication Signals

With the growing application of signal separation technology, traditional methods often perform poorly under low signal-to-noise ratio (SNR) conditions. This paper proposes an improved Conv-TasNet network model aimed at enhancing signal separation performance under low SNR conditions. Experimental results show that, compared to existing signal separation models, the proposed model demonstrates higher signal similarity coefficients and lower error rates across multiple SNR conditions, with particularly superior performance in low SNR environments. This study indicates that the proposed model can significantly improve signal separation results and holds potential for applications in complex noise environments.