MLP-based radar detectors for Swerling 1 targets

This paper deals with the application of Multilayer Perceptrons to radar detection. The dependence of the neural detector performance on the network size and on the signal-to-noise ratio selected for training is considered. Multilayer Perceptrons with different numbers of neurons in the hidden layer have been trained using different values of the signal-to-noise ratio to minimize the mean square error using the error back-propagation algorithm. Results show that the higher the number of hidden neurons, the closer the neural detector to the Neyman-Pearson optimum detector and the lower the dependence of the Multilayer Perceptron performance on the signal-to-noise ratio selected for training. Due to its practical interest, the very low probability of false alarm values has been considered. To estimate the probability of a false alarm, importance sampling techniques have been used in order to reduce the computational cost of maintaining a low relative error.