\section{Conclusions}

In the past, many MAGIC analyses have been conducted using different signal extractors. 
We developped and tested the most important signal and time extraction algorithms in the standard MAGIC software 
framework MARS. Our findings are that using a right signal extractor is important since some of the investigated ones
 differ considerably in quality and can severly degrade the subsequent analyses. On the other hand, we have found that 
advanced signal recontruction algorithms open a new window to lower analysis energy threshold and permit to use the 
time information of shower analyses. 
\par
In order to give a guideline for future usage of the tested signal extractors, we consider the following 
requirements to be of most importance:

\begin{itemize}
\item The calibration (including the F-Factor method) has to run stably and yield reliable results for all pixels.
\item The extracted signal should be as linear as possible over the whole dynamic range, including especially the 
low-gain range.
\item The combined resolution and bias should result in a lowest possible image cleaning threshold.
\item The extracted time should yield the best possible resolution.
\end{itemize}

Following these requirements, we recommend to exclude in the future the following signal extraction algorithms:

\begin{itemize}
\item All fixed window extractors using a window size of up to 6~FADC slices, 
including the fixed window peak search algorithm.
\item All sliding window extractors using a window size of up to 4~FADC slices.
\item The amplitude extracting spline.
\end{itemize}

For a conservative and stable analysis, we recommend to use (except for the December~2004 and January~2005 data):

\begin{itemize}
\item The sliding window, using an extraction window size of 6--8~FADC slices for the high-gain and 8~FADC slices for the 
low-gain channel.
\end{itemize}

For the most demanding analyses, especially at low energies and using the timing information, we recommend:

\begin{itemize}
\item The spline algorithm, integrating from 0.5~FADC slices before the pulse maximum to 1.5~FADC slices after the 
pulse maximum and computing the position of the half-maximum at the rising edge of the pulse.
\item The digital filter fitting the pulse over 4~or 6~FADC slices in the high-gain region and 6~FADC slices in the 
low-gain region.
\end{itemize}

Unfortunately, part of our recent data, taken in December~2004 and January~2005 had a severe problem with the pulse location
within the recorded FADC slices. In the recorded samples, the low-gain pulse is situated so far to the right that a part of 
it reaches out of the recorded window. This poses severe problems to all extractors which integrate the entire low-gain pulse. 
We have seen that the spline extractor and the digital filter over 4~FADC slices are still capable to reconstruct the low-gain 
pulse properly for this partly corrupt data sample, although the linearity of the reconstructed signal might still be affected above 
signals of about 300~photo-electrons per pixel.

\par

Special caution has to be made if the F-Factor method is applied for calibration with signal extractors which have an 
intensity-dependent resolution. This applies especially to the spline algorithms and the digital filter over a window size 
of 4~FADC slices.

%%% Local Variables: 
%%% mode: latex
%%% TeX-master: "MAGIC_signal_reco"
%%% End: