During the mission, two very different kinds of data are routinely acquired, processed on-board, transmitted to ground and here reduced, namely the ``housekeeping'' data needed for the spacecraft control and the scientific data proper. The former are obviously needed on a continuous time basis, while the latter can also be transmitted ``once in a while'', provided that a sufficiently large storage device is accommodated on the spacecraft and that the total amount of transmitted data at the end of the mission meets the scientific requirements.
For this reason, communication between the satellite and the Earth is done
via two sets of equipment. An X-band telemetry and telecommand link with
an omnidirectional coverage provides a permanent control of the spacecraft
via a continuous but relatively low data rate of 6 kbit/s.
A dedicated telemetry link, also in the X-band, provides the transmission of
the scientific data at a much higher rate of 3 Mbit/s only when a
ground station is ``visible'', and is complemented by a 100 Gbit
solid state memory ensuring the temporary storage between two consecutive
transmission periods.
Since only one ground station is likely to be affordable for the mission,
giving a mean visibility period of 8 hours a day, the resulting mean
science data rate is of 1 Mbit/s.
During the mission the satellite will therefore transmit to ground something
like 20 Tbyte of data, a huge amount of information.
A special CCD reading strategy was thus developed in order to identify the
CCD regions containing useful information, so as to observe as many objects
as possible, the achievable telemetry rate being fixed.
Such strategy is described in detail in Section 5.1.
Some level of data compression, by a factor in the 2-3 range, is then
performed to furtherly increase the total amount of scientific information
contained in the trasmitted data.
It is believed that such a compression rate could be easily achieved with
minimum impact on the data quality.
Still, the telemetry rate presently poses the most stringent constraint on
the number of objects that GAIA can observe, meaning that the nominal number
of 
observed objects could be made even larger if an higher telemetry
rate could be achieved.
A dedicated scientific data chain includes all the units required for the acquisition, the aforementioned discrimination, the compression and the storage of the data. The raw data are then transmitted to the ground without any further processing, and are readily made available to the astronomers for the data reduction proper.
The processing of GAIA raw data into consistent sets of astrophysical data is an extremely challenging task. It is not just the amount of data that is formidable, but even more so the intricate relationships between different pieces of information gathered with the various instruments throughout the mission. A highly automated, yet sophisticated data processing system will be required to take care of the bulk reductions. At the same time, a great deal of flexibility and interaction is needed to cope with special objects or astrophysical investigations, many of which cannot be foreseen at the software design stage. On the other hand, the delicate calibration of instrumental parameters and satellite attitude, necessary to interpret the data in terms of absolute astrometric and photometric quantities, must be protected from unintentional modification. It is envisaged that an object oriented database might provide a suitable environment for the GAIA data processing. [O'Mullane and Lindegren 1999] have accordingly developed a simplified model and have tested it using the Hipparcos Intermediate Astrometric Data contained in [ESA 1997a]. Basically, the data reduction starts with the so called great circle reduction, i.e. the processing of the data for each one-dimensional strip scanned by the satellite, which will allow to locate the objects relative to each other. Then the reduction process has to orient and phase the different great circles with respect to each other in what is called the sphere reduction. More specifically, this model follows a general scheme known as Global Iterative Solution, outlined in Chapter 23 of [ESA 1997a] and consisting of a cyclic sequence of three processes which are applied until convergence to four data sets, namely the CCD data, the calibration data, the sky data and the attitude data. The tested approach seems to be feasible and appropriate for the purpose of GAIA data processing.