GPSdancer performs routine analysis of GPS observation data from Continuously Operating Reference Stations (CORS). The main outputs of this analysis are highly accurate time series for the station position coordinates, receiver clock corrections and troposphere parameters.

The analysis consists of *rigorous global network solutions*, similar to the analysis performed at the Analysis Centres of the International GNSS Service (IGS). In such a solution, the (local) output products of all included CORS stations are computed simultaneously with global output products such as precise orbits and clock corrections for the GPS satellites, and Earth rotation parameters. This offers the highest possible GPS analysis accuracy for the CORS products. Global products can only be estimated accurately from a global network of stations.

A typical global analysis centre (such as the IGS Analysis Centres) can only process a few hundred CORS stations in a single solution. Such centres typically focus on the quality of the global output products (orbits, satellite clocks, Earth rotation).

Due to substantial differences in geometric dilution of precision between a CORS station on the ground, and a GPS satellite in space, the absolute accuracy of the CORS products in a global analysis is much better than the absolute accuracy of the global products that come from the same solution. For example, the baselines between the ~400 IGS CORS stations (forming the *station polyhedron*) are routinely computed at a noise of about 2.8 mm, while the satellite orbits that come from the same global solutions are only accurate to a few cm.

Regional analysis centres use these orbits, satellite clocks and Earth rotation parameters as fixed inputs to their own analysis: they simply cannot estimate such *global* products from their *regional* network of CORS stations. Some of the errors in the global products (notably, in the Earth rotation parameters) inevitably end up in the CORS output products of a regional analysis. Typical PPP accuracy levels for regional CORS stations is therefore limited to about 1 cm in the horizontal direction and 2 cm in the vertical direction, which is almost an order of magnitude worse than the accuracy of the station polyhedron from a true global network solution.

In short: the high accuracy of ~2.8 mm is currently only available for only 1% of the world's CORS stations, forming a coarse polyhedron with baselines in the order of 1000 km. All other CORS stations must live with PPP errors at the cm level.

The key difference between GPSdancer and a global analysis center is that it is implemented as a distributed process on the internet, in which many computers can work together in a single, *global* solution process. There is no practical limit to the number of CORS stations that can be included: all tens of thousands of regional CORS stations can in principle become part of routine, *accurate* polyhedron realizations, without suffering from errors in fixed, external global products. The accuracy of the CORS polyhedron in GPSdancer is expected to remain similar to that of the IGS (< 3mm), regardless of the number of participating stations.

The GPSdancer analysis will be routinely monitored against the IGS solutions, by including a subset of existing IGS stations in the GPSdancer analysis (...we call this the *backbone network*). From these comparisons, we can monitor the absolute accuracy of the ITRF realizations by GPSdancer, while the mutual accuracy of the statiion coordinates of thousands of further CORS stations can be monitored from their mutual baseline noise. For the first time ever, CORS stations all over the world will be able to know their ITRF coordinates with a verifiable accuracy level of a few mm, and at a very short processing latency.

The global network analysis consists of a continuous series of batch least squares estmation processes, in which a large number of global and local unknowns is solved simultaneously, from a batch of GPS tracking observations collected over e.g. the most recent 48 hours. Almost as soon as a process has completed, the next process starts, using a data batch that has a large overlap with the previous data batch (see Figure below). Because the precise orbit is estimated from its dynamic models, the orbit can also predicted over a few hours into the future with good accuracy (cm level). The key processing characteristics, and their target values for the GPSdancer analysis, are:

Data batch length | 48 | hours |

Observation sample interval | 30 | seconds |

Process execution interval | 30 | minutes |

Prediction interval | 2 | hours |

Number of satellites in the solutions | 30 | - |

Number of CORS stations in the solution | UNLIMITED | - |

In various further articles on this website you can read how the distributed solution of GPSdancer works, which model parameters are estimated, how the network of collaborating GPSdancer instances is constructed, and much more. Please follow the links in the text, or use the menu options above to navigate around.

The GPS Dancer project started in 2007 as a voluntary project of a working group of the International Association of Geodesy.

The GPS Dancer system was named after its "square dance" exchange algorithm. Of course, it also wants to to make the GPS reference frame denser.

The Dancer on-line network became immune against internet connection problems by leaving the US marines, and becoming a pirate.