...
To intercalibrate gains of each given log end, we summed almost (see below) all runs available (to maximize the statistics) and fitted the profile histogram of R_i (signal for gain #i) vs R_i+1 (signal for gain #i+1) to get the slope. This profile is limited to the events where a) R_i does not saturate, b) R_i+1 < saturation value of R_i and c) the ratio R_i/R_i+1 is close to 1 (within a band parallel to and centered on the y=x line).
In addition, for the logs at the center of a module, we first ignored the runs corresponding to scan positions near their end to avoid direct illumination of the diodes due to beam and shower spread (in that case the scatter plots are a mess...). Despite these criteria, some weird features were observed, e.g. for the log end T2-L6-X5-S0 (i.e. tower 2, layer 6, column 5, side 0): in this file the 4 plots at the top are the R_i spectra (summed over all runs) in each energy range (R_0=LEX8, R_1=LEX1, R_2=HEX8, R_3=HEX1) while the 6 plots at the bottom are R_i vs R_i+1 (scatter plot plots on the left, superimposed on the linear fit fits - in red - of the profile histogram in redhistograms) and R_i/R_i+1 ratios (distribution distributions on the right). Two populations are clearly visible in the R_1/R_2 ratio plot (with a difference of ~7%). The "bad" population was found to arise from one specific scanning position. Since other similar cases were found, new runs were taken at the same scanning positions, but the effect remains and is still not understood. However, we removed it using the following run selection:
...
This procedure yielded a list of selected runs for each log end ratio and better results, see e.g. T2-L6-X5-S0. It was applied to all log ends and the results were stored in the following files: tower 1, X logs, tower 1, Y logs, tower 2, X logs, tower 2, Y logs, tower 3, X logs, tower 3, Y logs. In these files the 96 first pages (one per log end) show the same kind of plots as those described above, and the 3 last pages show the graphs and distributions of the slope, chisquare and crossing value of the linear fit of the 96x3 profile histograms (the crossing value is in MeV and the range of the plots is +-1% of the R_i+1 spectrum maximum). These results show that tower 2 is correctly calibrated. In tower 3, Y logs are correctly calibrated as well, while 2 outliers (T3-L2-X7-S1-R_1/R_2 slope, T3-L6-X4-S0-R_2/R_3 slope) are observed as underflows in the slope distributions of X logs, also visible in the crossing value graph as points with large errors; the individual plots show that the second outlier is due to poor statistics, while the first one should be looked at in much detail (actually we assumed that the calibration is fine here, like for all other logs in tower 3...). Finally in tower 1, X logs log calibration constants are reliably obtained (1st and 3rd slopes very close to 1, and 2nd slope R_1/R_2 ~1.087 with a RMS of ~0.03), except the R_2/R_3 slopes (not used in data analysis anyway...) in layer 0 due to poor statistics at high energy since showers start deeper in the CAL in the absence of TKR in front of the 1st CAL module; as for the Y logs in tower 1, calibration constants are also reliably obtained, except for T1-L7-X11-S0 which exhibits a strange behaviour (R_1/R_2 and R_2/R_3 slopes were forced to 1 in the absence of any explanation...).
...