Word Embeddings for Radar Emissions: A Comparison Between Word2vec and fastText

To analyse and store emissions of agile radars, an alternative representation to the traditional entries in a mode database is required. Such a representation is provided by the application of word embeddings in combination with the hierarchical emission model of a radar language. This paper gives a comparison of the embeddings word2vec and fastText. The use of subword information in combination with fastText is beneficial for the representation space and the approach is highly adaptable to different datasets and levels of abstraction. With the capability to represent unknown words of known symbols this method is explicitly useful for signals of agile radars. This comparison is the foundation for further research in this area.