Blind Source Separation Schemes for Mono-sensor and Multi-sensor Systems with Application to Signal Detection

In this paper, blind source separation (BSS) techniques based on time-frequency (t-f) distributions are proposed for multi-sensor and mono-sensor scenarios. The proposed schemes use t-f filtering and high-resolution t-f distributions to extract source signals that have very close components in the t-f domain. Through numerical simulations, the performance of the proposed schemes is compared with the existing algorithms. Results show that the proposed method outperforms the traditional BSS methods. In addition, the proposed BSS is applied to detect a presence of a non-stationary signal in a scenario, when noise power is unknown. Detection performance is compared with the existing detection methods through numerical simulations that show the proposed method performs better than the existing methods.