Machine learning techniques for frequency sharing in a cognitive radar

A Cognitive Radar, working in a frequency dense environment, has to perform effective wideband operations, spanning several frequency channels, by working in parallel with other radar and/or communication systems. The cognitive operation is possible by modeling the channel behavior and predicting future channel occupancy. The model of the electromagnetic environment is based on the observation of the spectrum occupancy during a number of time slots and on suitable machine learning to acquire the characteristics of the channel occupancy. The learning operation is paramount, as the prediction about channel occupancy is possible only after understanding the behavior of the concurrent emitters present in the scenario. This paper describes the concept of machine learning techniques, based on emitter pattern classification and matching. implemented on a number of real cases of emitter behavior. In particular, we are defining these techniques by considering four real cases of emitter behavior, namely fixed, sequential, periodical and random channel acquisition. We show that, in the above examined cases, our machine learning techniques can provide good emitter matching, even in presence of a consistent number of concurrent transmitters.