add progress14

progress14/AnalysisFramework.tex

@@ -0,0 +1,124 @@
+%\subsubsection{plan}
+%\begin{itemize}
+%\item two stage: alcapana then rootana
+%\item alcapana is online analysis as well as primary midas -> root conversion
+%\item rootana is the main offline analysis
+%\item Waveform Analysis
+%\begin{itemize}
+%\item Pulse Candidate Finder
+%\item Amplitude: MaxBin
+%\item Timing: Constant Fraction
+%\item Calibration: Timing offsets, Pedestals, Energy scales
+%\end{itemize}
+%\item Event Correlating:  Fast and Slow coincidence
+%\item Event Correlating:  Muon events
+%\end{itemize}
+
+\begin{figure}[htbp]
+  \centering
+    \includegraphics[width=0.7\textwidth]{figs/WaveformAnalysis.png}
+    \caption{Illustration of the waveform analysis techniques used in Rootana.}
+  \label{fig:waveform_reco_demo}
+\end{figure}
+
+
+%The analaysis of the Alcap data is split between two main stages, known as
+%Alcapana and Rootana.  
+The AlCap DAQ delivers data as compressed MIDAS files. These are then
+handled in two stages.  In the first stage (Alcapana), 
+some preliminary analysis is
+performed and the MIDAS files unpacked into a ROOT format.  These are then
+passed through the second stage (Rootana) for physics analysis.
+
+\subsubsection{Alcapana}
+The MIDAS files produced by the DAQ contain both the digitised
+waveform outputs
+and the run-time DAQ configuration.  Alcapana is the first point in the
+processing chain to look at this data and performs two roles: 1) it 
+provides some preliminary, semi-online analysis, 2) it converts the 
+MIDAS data into a ROOT format.  
+
+The analysis produces simple histograms of quantities such as amplitude,
+timing, and pulse island length.  It also creates persistency-style 
+overlays of each
+waveform in the event.  These histograms are placed in an output file which can
+then be loaded into a simple ROOT-based GUI where results can be displayed
+rapidly and help to understand the data quality.
+%Much of this was implemented during the run and by the end the alcapana
+%analysis and displays formed a strong and user-friendly tool-kit.
+
+Since each digitiser is run in a self-trigger mode within a MIDAS block, the
+output waveforms contain, in principle, a single pulse stored as a vector of
+ADC samples. The waveform digitisers stamp each pulse with a trigger time-stamp relative to the
+start of the MIDAS block.  Each of these time-stamped ADC vectors is referred
+to as a Pulse Island.  Occasionally a
+real pulse may be split over two Pulse Islands. Therefore, during the
+unpacking in Alcapana Pulse Islands are with adjacent time-stamps are
+checked to concatenated into a single pulse when necessary.
+The treatment of the $\mu$PC detector is slightly different as it is readout using a discriminator and TDC, and so does not use the same waveform format for raw data.
+
+\subsubsection{Rootana}
+Most of the physics analysis is performed in Rootana.  The framework is
+designed to be easily configured by dividing the analysis into
+modules which can be called and reconfigured at run-time through a
+configuration file.  Most processing chains begin with waveform analysis so
+the module for this delegates the work to runtime selectable `generators' which
+implement the actual waveform analysis.  Then the program searches for
+correlations or coincidences between different detectors in order 
+to produce the final physics results.
+
+\subsubsection*{Waveform Analysis}
+The general approach to waveform analysis
+uses the following methodology:
+\begin{description} 
+\item [1. Find Pulse Candidates:] Remove pulse islands which are just noise and
+divide pile-up pulses into two separate pulse islands.
+%
+%\item [2. Subtract Pedestal] 
+%
+\item [2. Amplitude and Time reconstruction:] Use the Max Bin method (sometimes
+Peak Sample method) to extract the amplitude followed by a Constant Fraction Timing
+process using a linear interpolation between the two bins closest to the
+peak-sample which crosses a given fraction of the pulse's amplitude.  See Fig.
+\ref{fig:waveform_reco_demo} for an illustration of these techniques.
+%
+\item [3. Apply Calibration Constants:] Account for cable delays and 
+similar effects in
+the timing and convert the amplitude to energy.  The constants 
+are obtained from a prior data pass which calculates the timing and
+pedestal data, or from a dedicated calibration datasets in the case of energy
+constants.
+\end{description}
+
+We are investigating alternative methods of waveform analysis using
+pulse-averaging in order to produce a template followed by fitting or 
+convolution of the
+original waveform.  Since the analysis shown below has not used these
+methods, 
+they will not be described here.
+
+\subsubsection*{Pulse Correlation}
+There are two types of pulse correlation used in the analysis: 
+1) correlation of
+pulses between the fast and slow filtered read-out channels of a
+single detector, 
+and 2) correlation of pulses between all detectors.  The first method 
+provides a
+cross check on the data quality, but results in a reduction of the overall
+dynamic range of the detector since the correlation is only meaningful for
+pulses with energies within the intersection of the individual 
+dynamic ranges of each channel.
+The second method is used when pulses are correlated between
+detectors. It 
+is implemented
+by restructuring the dataset into what are known as ``Muon Events''.  Individual
+pulses on the $\mu$Sc are used as a reference point and all pulses which occur
+within a certain time window (typically 15~\textmu s) are collected together into
+a single muon event.  Should two or more $\mu$Sc pulses occur within this window,
+the event is marked as a muon pile-up event.  It is then trivial to scan
+over each muon event, applying various cuts, in order to build a
+spectrum of interest.
+
+Since the bulk of the correlation work is done by inspecting the timing of
+pulses on different channels, the accuracy of the time offsets due to
+ cable delays is particularly important.