Observing faint targets with MIDI at the VLTI -- The MIDI AGN Large Programme experience

In order to put MIDI/VLTI observations of AGNs on a significant statistical basis, the number of objects had to be increased dramatically from the few prominent bright cases to over 20. For this, correlated fluxes as faint as ~ 150 mJy need to be observed, calibrated and their errors be estimated reliably. We have developed new data reduction methods for the coherent estimation of correlated fluxes with the Expert Work Station (EWS). They increase the signal/noise of the reduced correlated fluxes by decreasing the jitter in the group delay estimation. While correlation losses cannot be fully avoided for the weakest objects even with the improved routines, we have developed a method to simulate observations of weak targets and can now detect --- and correct for --- such losses. We have analyzed all sources of error that are relevant for the observations of weak targets. Apart from the photon-noise error, that is usually quoted, there is an additional error from the uncertainty in the calibration (i.e. the conversion factor). With the improved data reduction, calibration and error estimation, we can consistently and reproducibly observe fluxes as weak as ~ 150 mJy with an uncertainty of ~ 15 % under average conditions.