Ice Top DOM Feature Extraction Paul Evenson David
- Slides: 17
Ice. Top DOM Feature Extraction Paul Evenson David Seckel University of Delaware Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Ice. Top Science Goals • In. Ice Veto – Retain all hits for readout • In. Ice Calibration – Trigger for vertical and horizontal showers • Cosmic ray composition – Shower reconstruction over large dynamic range: 300 Te. V – 1 Ee. V Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Ice. Top Raw Data • Single Muons 1200/second – Horizontal showers + background • Soft Component (>30 Me. V) 1000/second – Vertical showers • Waveforms 100/second – Showers: dcore > 300 m, muon id within core – Calibration + monitoring Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Ice. Top Data Return Strategy • Single Muons – Characterize, identify as muon, return short report • Soft Component – Characterize, identify as soft, check for local coincidence, return short report • Waveform – Feature recognition fails – Scaled selection of minimum bias and event triggers – Compress and return complete waveform Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Ice. Top Feature Recognition • Constraints – – 100 kilobyte/sec uplink rate 5000 raw events/second 200 microseconds per event 6600 cpu clock cycles • Strategy – Use Altera code to characterize event and compress waveforms Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Altera FPGA • Good at addition, subtraction, multiplication • Bad at long division • Therefore: Work toward feature recognition and characterization based on fitting with orthogonal functions – One pass does everything – Expansions always truncate Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Eight Function Orthogonal Fitter (Thanks to Andrew Mc. Dermott, Holger Leich and Gerald Przybylski) Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Eight Function Fitter Operation • For ease of simulation, the input waveform in the example is just encoded in a ROM • Outputs are likewise just put in a FIFO • Fits eight orthogonal functions • Uses about 1/3 of the “real estate” of the current chip in the DOM • Runs in about 80 microseconds at 25 MHz clock rate Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Eight Function Fitter Operation • First Pass (128 clock cycles): – – Clock waveform into (recirculating) Waveform FIFO Dot product waveform with first fit function to get coefficient 1 Enter first fit function into Fit FIFO Multiply waveform by (negative) normalization constant and enter into the Quality FIFO • Passes 2 to 8 (fit function “N”) – Dot product of waveform with fit function “N” to get coefficient N – Multiply fit function N-1 (in Fit FIFO) by coefficient N-1 and add to the contents of the Quality FIFO – Enter fit function N into Fit FIFO • Pass 9 – Sum absolute values of Quality FIFO (This could be done in Pass 8, but this seems to be overall much faster. Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Which Orthogonal Functions? Chebyshevs are sort of OK Chebyshevs Discrete functions: 128 samples Reconstruction Powers of t Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Basis Function Based on Data (Hey, it works!) Trial 1 Use mean m Reconstruction Mean m + powers of t Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg. for u 1
Tailored Functions: PMT model • Three parameter PMT pulse: g, t 0 – model: – expand: – data: – orthoganalize: Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Data + PMT Model Data based trial set Reconstruction m 0 + dm/da Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Trim noise + Offset parameter Best trial set Reconstruction Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
Summary • Ice. Top goals require efficient use of bandwidth between DOM and Hub • FPGA development effort at UD • Feature extraction – orthogonal functions – use data + model to make orthogonal basis • Estimated bandwidth 10 B per hit • Need to validate on actual DOM Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
STOP This slide intentionally left blank. Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg.
RMS pnt by pnt residual This slide was not shown at Berkeley. It illustrates the average improvement in fitting over the whole data set of 81 events. RMS noise is at about 5 m. V Black –Chebyshev Blue - Data 1+ powers Red - Best Paul Evenson & David Seckel 11 March 2002 Ice. Cube Collaboration Mtg. Best trial (expanded view)
- Shelley evenson
- Edge detection
- Feature dataset vs feature class
- Isolated feature combined feature effects
- Tat sat temperature
- Clear ice vs rime ice
- Down in the meadow where the green grass grows there sat
- David ausubel theory
- Biografia ausubel
- Paul david meesters
- Paul david qc
- Xml dom
- Unidade de ensino superior dom bosco
- Java xml dom
- Dai nos senhor esses dons essa luz
- Porque o dom de profecia é superior
- Ucla dom dra
- Domphp