Generation of soft information in a frequency-hopping HF radio system using neural networks

Abstract

An attempt has been made to improve the performance of a digital frequency-hopping radio system on the HF band. The object was to generate soft information for the convolutional decoder in the system and thus improve its error correcting capability. The nature of the problem implied that a solution with neural networks would be of interest. Computer simulations demonstrated that a two-layered network with two neurons in the first layer and one output neuron gives the best results. The results show that a neural network and the chosen convolutional code together have an extensive error correcting capability. For example, one could transmit information essentially correctly, despite the fact that noise, causing a bit error rate of 1.5% on all the used frequencies, was added to the channel and an active jammer completely corrupted a fourth of the frequencies used.<>