The Value of Memory: Markov Chain versus Long Short-Term Memory for Electronic Intelligence
Sabine Apfeld, Alexander Charlish, Gerd Ascheid
Abstract
In this paper, we compare Markov chains and Long Short-Term Memory neural networks for the prediction of radar emissions and the identification of the radar's type. This comparison is especially interesting since Markov chains and Long Short-Term Memory networks are in contrast to each other in terms of memory. The two methods for prediction and identification are based on an emission model that understands the radar's emissions as a language with an inherent hierarchical structure. The evaluation is performed with the data of a simulated airborne multifunction radar that can make use of three resource management methods of varying complexity. It is shown that the Markov chain outperforms the Long Short-Term Memory in simple scenarios, while the neural network is better suited for more complex tasks. Moreover, its identification performance is much more robust with respect to corrupted data.