Artificial Intelligence for Unidentified Mode S Registers Decoding

Aeronautical surveillance is primarily based on cooperative sensors that perform an exchange of messages between aircraft and surveillance ground infrastructure. The aircraft electronic equipment in charge of the messages exchange is called transponder. In addition to basic surveillance information, the transponder also holds registers with avionic related data such as the selected altitude, the magnetic heading, the ground speed, etc. However, messages containing data from transponder registers do not always include an identification number, making the decoding of such messages impractical for an external observer.In this paper, a new method to identify a transponder register without prior knowledge of its identification number is presented, allowing an external observer to decode correctly registers passively received.The proposed method is based on an innovative approach using Artificial Intelligence and more specifically Natural Language Processing (NLP) techniques applied to surveillance messages. Each message is considered as a sentence and is evenly split into words. First, a tokenization layer identifies each word, then the embedding layer attaches to them a fixed number of properties, and finally a recurrent neural network recomposes the sentence and ensures the message identification.The implemented method achieves a very good performance, showing an accuracy in the registers identification of 0.999987, which is significantly higher than any other existing solutions.