Machine learning-based dynamic spectrum access for aircraft-to-aircraft communication under coexistence with legacy radio systems

Spectrum scarcity is viewed as one of the key obstacles in the area of wireless communications. The lack of available unlicensed resources is impairing the development of newer and more modern communications systems. This is the case of L-band Digital Aeronautical Communications System (LDACS), an innovative air communication system that aims to use the part of the frequency spectrum licensed by Distance Measuring Equipment (DME), a legacy radio navigation system. DME has a low channel utilization rate, leaving idle numerous resources that could be used by LDACS through the use of Dynamic Spectrum Access (DSA) in Cognitive Radio (CR). In order to avoid interferences and collisions while taking advantage of these idle resources, this thesis proposes a new LDACS Machine Learning (ML)-based Medium Access Control (MAC). It incorporates a Long Short Term Memory (LSTM) Recurrent Neural Network (RNN) in order to observe, learn, predict and avoid the DME licensed users. The results from this new MAC are analyzed and compared to two alternative approaches, showing the advantages of a ML-based approach.