Neural networks and early fast Doppler for prediction in meteor-burst communications systems.

Abstract

In meteor-burst communications systems, the channel is bursty with a continuously fluctuating signal-to-noise ratio. Adaptive data rate systems attempt to use the channel more optimally by varying the bit rate. Current adaptive rate systems use a method of closed-loop decision-feedback to control the transmitted data rate. It is proposed that an open-loop adaptive data rate system without a decision feedback path may be possible using implicit channel information carried in the first few milliseconds of the link establishment probe signal. The system would have primary application in low-cost half-duplex telemetry systems. It is shown that the key elements in such a system would be channel predictors. The development of these predictors is the focus of this research. Two novel methods of predicting channel parameters are developed. The first utilises early fast Doppler information that precedes many long duration, large signal-to-noise-ratio overdense trails. The presence of early fast Doppler at the trail commencement is used as a toggle to operate at a higher data rate. Factors influencing the use of early fast Doppler for this purpose are also presented. The second method uses artificial neural networks. Data measured during trail formation is processed and presented to the neural networks for prediction of trail parameters. Several successful neural networks are presented which predict trail type, underdense or overdense, and peak trail amplitude from the first 50ms of the trail's lifetime. This method allows better estimation of the developing trail. This fact can be used to implement a multi-rate open-loop adaptive data rate system.