The feasibility and utility of long-distance communication via Earth-orbiting satellites has been demonstrated during recent years and it is appropriate therefore to focus attention on the more important scientific studies and technical developments that will be needed if full use is to be made of this valuable mode of communication in the future. The early communication satellites (the Telstar and Relay series) were pioneers in a relatively unknown propagation environment. The satellites themselves were conceptually simple and the communication equipment consisted essentially of a frequency-changing transponder with an r. f. power output of a few watts and a bandwidth some tens of megahertz. Carrier frequencies in the range 2 to 6 GHz were employed; typically either 2 or 6 GHz was used for transmission and 4 GHz for reception at the Earth station. To obtain an adequate signal/noise ratio at the output of the Earth station receiver, frequency modulation was employed, the frequency deviations being greater than those used on terrestrial microwave links. Launcher limitations and other factors meant that the satellites had to be placed in inclined elliptical orbits (see figure 1) with maximum heights of only a few thousand miles. Nevertheless, these satellites demonstrated that some hundreds of frequency-division multiplex telephony circuits, or a television channel, could be achieved with generally satisfactory quality of transmission. It is to be noted, however, that the satellite transponders accommodated only one, or at the most two, r. f. carriers at any time, and that the transmission performance was at times marginal due to limitations of the satellite effective radiated power. Furthermore, these relatively low orbit satellites provided communication in periods of generally less than an hour at a time and required continuous tracking by the Earth station aerials, due to movement of the satellites relative to the Earth.
Comparison of domain adaptation and active learning techniques for quality of transmission estimation with small-sized training datasets [Invited]