부산대학교 도서관

The world relies on accurate and reliable navigation made possible by GNSS. There is significant interest in expanding the range of signals that can be used for navigation in case the traditional GNSS sources are unavailable. The proliferated communications satellite constellations are particularly interesting because of they offer unparalleled level of resilience. The Starlink satellites transmit a wideband signal at power levels much higher than GPS and provide worldwide coverage. This paper provides new experimental results that demonstrate positioning with better than 100m accuracy using only Starlink downlink signal.These are the first published results that take advantage of the full downlink signal. A low-gain antenna is used to capture signals from a 30-degree cone that includes many Starlinks at any time. Unlike previously published work, we process all the in-view satellites rather than pointing at just one. Not only does this approach solve the question of which satellite is active at any time, but it also increases the number of useful observations.A wideband digitizer is used to process the entire band that consists of six 250 MHz channels. All the channels are processed simultaneously since the receiver cannot tell which channel may be active at any time. Each channel is individually correlated with a fixed sequence (preamble) that is known to arrive at regular intervals. The key contribution of this paper is an algorithm that takes advantage of the regular arrival intervals to extract the change in the range between the satellite and the receiver. This range-rate is similar—but not identical—to the Doppler frequency offset that was used in previous work.The rate change observations are combined into traces that correspond to different satellites and channels. An algorithm to separate overlapping and/or intersecting traces allows us to use all the available data.Starting from an estimated initial location and estimated initial time, we form a model to correlate each trace to a satellite/channel. An optimizer is used to minimize the difference between the observations and model is predictions parametrized by (x,y,z,t). The initial results using a simple model are very promising.