Directional Modulations Using an Artificial Neural Network

Abstract

A directional modulation (DM) system can provide physical layer security by distorting modulations along the eavesdropper directions while maintaining the correct modulation formats at the desired user. One DM approach is to optimize the transmit signals from the transmit antenna array so that a high bit error rate (BER) can be enforced at the eavesdroppers. However, this requires running an optimization algorithm each time the locations of the users change resulting in high numerical computations. To overcome this problem, a multilayer perceptron network from the artificial neural networks (ANN) is used in this paper to obtain the optimized transmit signals. This paper considers only one desired user and one eavesdropper, and the ANN is trained for various orientations of the users. The impact of hyperparameters, e.g., the number of neurons, is studied. The use of shallow and deep networks is investigated. Excellent BER performance is obtained with low numerical computations.