Spatiotemporal Density-Based Clustering for Dynamic Spectrum Sensing

Dynamic spectrum access is one promising model for managing spectrum congestion and ensuring primary users, such as essential radar systems, unimpeded access to spectral resources. However, this requires the secondary user to identify the temporal and spectral resources consumed by primary users. Thus, in a congested radar environment, the secondary user must be capable of resolving multiple emitter waveforms in the presence of channel noise, waveform ambiguities, and pulse-on-pulse artifacts. We propose a new kernel density estimator-based clustering technique which uses the time of arrival of radar pulses in addition to other features, such as angle of arrival, center frequency, and pulse width, to identify patterns in radar pulse trains with a wide range of possible pulse repetition frequencies, which is a weakness of many density-based clustering techniques, in the presence of measurement error and outliers.