From: Ed Fomalont Subject: Note from Ed Fomalont Dear Brian, You are getting to the stage where you need good input from people who have met all kings of astronomical data. I am committed to debugging the VLBA and this will occupy most of my time, so I`m sorry that I can't spend much time on the aips++ design. But, I want to say a few things about what you should be thinking about for the design. The most important point, in my opinion, is that to design a good, flexible system, you need only worry about how to APPLY calibration. This is basically the TELESCOPEMODEL object. This means taking measured data and transforming them to obtain something called calibrated data. These algorithms are general simple matrix transforms of the input data. How to determine the parameters in the APPLY calibration matrices is called SOLVE. The complexities involved in solving from parameters from a set of data is enormous. But, I don`t think you have to worry about this. You have to make sure, that the structure of aips++ can handle the APPLY calibration process and transfer sufficient information (data and calibration parameters) to SOLVE, which the gives better calibration parameters. But, what and how SOLVE does is unimportant at this stage of design. One important assumption is that APPLY calibration need only handle data at one particular time interval. If this is true, then this must say something about the fundamental ordering of the data base. Clearly the APPLY parameters will change with time and the information (tables? algorithms?) must have to be repeated as often as necessary. And the proper interpolation methods must be used. On polarization calibration. At most you observe four independent polarization data points for any baseline and time. The general calibration procedure is a 4x4 matrix which transforms these four inputs points to four output points. It doesn't matter what type of polarization state is measured (RR, LL, RL, LR or linears or some Westerbork style) or what output is desired (I, Q, U, V or RR, LL, RL, LR), I think that a 4x4 matrix will do it. Of course each matrix element will depend on the style of polarization measurement and calibration. And SOLVING for the parameters for a specific telescope is tough, egs. VLBI polarization mapping. BUT, in determining the properties of aips++, I think you only have to convince yourself that a 4x4 matrix is all that you need. How to fill in the matrix is not important as long as you know how to send the appropriate data (now over a long period of time) and a description of the parameters to SOLVE and let it do its job, somehow, and return better estimates of the parameters which you stick in the proper matrix. Some complications: It is possible to measure many more than four input values which are then not all independent. The APPLY would then be a sort of least square solution which obtained the four desired independent polarization parameters. On the other hand, only one or two polarization parmeters can be measured. The output then cannot determine the four polarization states and simplifying assumptions (ie some of the 4x4 matrix elements will be zero) must be made. BUT, this can all be buried in some Matrix (nxM) where n is the number of input polarization measurements and M is the number of output (M