Multi-Level Mean-Shift Clustering for Single-Channel Radio Frequency Signal Separation
Emerging wireless communication applications have led to a crowded radio frequency (RF) spectrum. Therefore, it is desired to develop signal separation techniques that can extract different RF signals from their mixtures. Existing signal separation approaches typically require multiple observations of the signal mixtures and depend on statistical independence among the signals. In this paper, we consider separating multiple RF wireless signals from their single-channel superposition. These RF signals are transmitted in their corresponding high-frequency pass bands with diverse power spectrum densities, bandwidths, and time durations. We propose a signal separation approach that exploits the mean-shift clustering algorithm with multiple levels of cluster sizes to identify RF signals with different bandwidths in the spectrogram of the superposed signal. We demonstrate the effectiveness of our approach by separating RF signals using real datasets.
