The use of the trans-spectral coherence technique, to separate a signal from noise

Abstract

A new concept of the trans-spectral coherence has been developed. Consider a narrow-band signal A, with a center frequency f1, and a signal B, with a center frequency f2. Now, if a nonlinear filter could be found that converts the signal A into a signal C at f2, the trans-spectral coherence between A and B is the ordinary coherence between C and B. If the two signals A and B are derived from a common source through a nonlinear interaction, this kind of coherence is nonzero. Most speech signals, signals from musical instruments, and some signals from industrial noise sources, can be expected to possess nonzero trans-spectral coherence. In this paper, it has been conjectured that, if sound from a nonlinear source is mixed with random noise, the trans-spectral technique can be used to separate the signal from noise.