Date: Fri, 05 Apr 2002 13:00:44 +0200 From: Jean-Paul Berthias To: Pascal Willis Cc: ries@sol.csr.utexas.edu, bjh@cobra.jpl.nasa.gov, feissel@ensg.ign.fr, Carey Noll , frank.lemoine@gsfc.nasa.gov, nerem@colorado.edu, nzelensk@geodesy.gsfc.nasa.gov, krchoi@csr.utexas.edu, "Tavernier, Gilles" , yeb@cobra.jpl.nasa.gov, Christian Jayles , werner.gurtner@aiub.unibe.ch, Jean-Jacques Valette , "Granier, Jean-Pierre" , Alfred Piuzzi , Sabine Houry , Flavien Mercier Subject: Re: Please change DORIS/SPOT preprocessing and DORIS/CDDIS files Dear colleagues, For many years we have been looking at how to provide DORIS data to users in a "rawer" format, without loosing the advantages of the current format. We have to admit that we did not make much progress in this field, as the various objectives contradict each other. What happened to the 2.0 format is a good example of this situation. The core of the move toward the 2.0 format was to remove any modification to the raw 2 GHz Doppler measurement, besides a constant shift and a change of units. There was also a few marginal changes that included adding the channel number to accomodate new instruments. But the key idea was to give an absolute status to the measurements themselves. Pascal soon pointed out that the result was somewhat incoherent, because the time tags and the Doppler are in different time scales. Time tags are corrected (they are in TAI), Dopplers are not. This has no impact on traditional single satellite Doppler processing, but it is a limitation for direct multi-satellite solutions. In practice, when you solve for frequency biases you obtain a combination of the beacon and satellite offsets. Thus you cannot constrain the offsets to be the same for different satellites at nearly identical times. There exist a way out, using the fact that the long term trends in the frequency offsets observed over the master beacons is due to the on-board USO. This is what we do, but this requires multi-step processing ... A simple and elegant solution to this problem was to correct the time-tags and the Dopplers using the same transformation, that is removing in a coherent fashion the long term trend of the on-board USO (timing polynomial) both from the time-tags and the Doppler measurements. This is what we have done to the Topex and Jason data which are available to the Jason POD group for evaluation. Pascal is now asking us to implement this procedure to all the CDDIS DORIS data. We have not done it so far because this approach destroys the absolute nature of the measurement that we were trying to introduce with the 2.0 format. More precisely, the timing polynomial is not uniquely defined: for each interval, the processing of the DORIS "pseudo-ranges" collected over the (limited set of) master beacons will provide a different polynomial. Thus, the Doppler data corrected for this long term drift become dependant upon the time window used to compute the polynomial. This is not a problem for Topex and Jason POD activities which are conducted on a fixed cycle basis. But for the user community as a whole, this can be misleading. For example, when converting cycle files to daily files, the "end of cycle" days will contain a form a discontinuity induced by the use of two timing polynomials. In addition, until now, we had more or less aligned the Spot data processing on the Topex "rythm", but with the addition of Envisat there will exist a new rythm ... And the workload associated with Topex, Jason and Envisat is likely to induce additional delays in the handling of the Spot data. There has been some work on faster data delivery, with the production of daily data files. But, if the timing correction is applied, daily files and cycles files will not contain the same value for the measured Doppler ... In my view, this is the result of a simple fact: the only true solution is to process raw pseudo-range and Doppler data, and due the use of a globally unsynchronized network, this cannot be easily done in a single step (or, at least, I should say that all our attempts have failed !) even though it is feasible in theory. The time-tagging of the DORIS data requires an additional input, a modelization of the long term behavior of the on-board clock, which has to be balanced with the short- and medium-term behavior of the USO. There is a basic inobservability that has to be dealt with carefully. In addition the arc length required to obtain a proper estimation of the timing polynomial is significantly longer than one day. Thus, a multi-step iterative process which solves alternatively for orbits and for timing is the best thing that we have found. But, obviously, this requires more work than standard orbit determination. In order to enable outside users to run simple one-step processing we have provided preprocessed data. However, now that users want to get more out of DORIS, this approach has reached its limits. It is probably about time for others to become involved in the full DORIS processing, with all its complexity. It is not just a problem of format ... This should certainly be discussed at the workshop in June. As for Pascal's request, it has implications that need to be addressed before implementing a new strategy at CDDIS. I am waiting for your inputs on this subject. Best regards, Jean-Paul