Frequency estimation using back-propagation neural networks for a frequency hopped spread spectrum receiver

This paper presents a novel scheme for estimation and detection of signal frequency in a frequency hopped spread spectrum (FHSS) system. The proposed signal detection scheme aims to reduce the number of channelized radiometers in the FHSS receiver. It is based upon the concept of back propagation neural networks (BPN). The system is first trained using a training signal and weights of different layers in BPN are configured by an iterative process. This scheme was implemented on a simulated FHSS system which had a hopping bandwidth of 26 MHz. The system hopped once for every eight symbols. Nine band pass filters were used to cover the entire hopping bandwidth. The proposed scheme aspires to reduce system complexity and intends to make it more robust and flexible. This system is designed for software based transceivers.