Spatial averaging of time-frequency distributions

This paper presents a novel approach based on time-frequency distributions (TFDs) for separating signals received by a multiple antenna array. This approach provides a significant improvement in performance over the recently introduced spatial time-frequency distributions, specifically for signals with close time-frequency signatures. In this approach, spatial averaging of the time-frequency distributions of the sensor data is performed to eliminate the interactions of the sources signals in the time-frequency domain, and as such restore the realness property and the diagonal structure of the source TFDs, which are necessary for source separation. It is shown that the proposed approach yields improved performance over both cases of no spatial averaging and averaging using timefrequency smoothing kernels.