1. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 1 EM timing analysis Warren Focke February 12, 2004

2. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 2 Introduction We performed a preliminary timing analysis in the EM VDG data to: –Measure the particle source rates. –Investigate deadtime effects and its impact on possible science analysis. Technical Overview –Analysis was done on the EBF files and used output info from ROOT for comparison (Xin). –Results were consistent with those obtained by the CAL group (Sasha). –Cross checks were done on cosmic ray files.

3. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 3 Timestamp Consistency EM has 2 timestamps: –Mini-GLT provides 17 ms (60Hz) granularity and ns precision. –SBC cycle counter provides 60 ns granularity and precision. Ratio between the two clock speeds calibrated at a level of 1 in 10 6. T max – T min ~ 2% of expected value (16.7 ms). GLT-SBC timestamps GLT = GLAST LAT Trigger SBC = Single Board Computer Deviation of edges from vertical is probably due to clock speed drift during run.

4. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 4 Lightcurve and Rates Observed average rates are 16.8 and 32.9 events/s. Rates are consistent with different VdG settings during the run. Effects will be discussed later in the context of the –Power spectrum and –Time difference between events.

5. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 5 Waiting Time Distribution (long) To avoid deadtime effects, only separations longer than 3 ms are shown. Because of the rate change during the run, two exponentials were required to fit the distribution. Incident rates (deadtime- corrected) deduced from the slopes of the exponentials are 17.2 and 34.4 events/s. Time differences between adjacent events. For a constant rate and Poisson statistics, this would be an exponential.

6. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 6 Waiting time Distribution (short) To display deadtime effects, only separations shorter than 3 ms are shown. 2.3 ms deadtime is due to lack of zero suppression in the calorimeter filling up the TEM buffers. Peaks at 1.6 and 1.7 ms have been reported previously by the CAL group. May be due to: –Double-triggering –Event pileup in data buffers 5% of total number of events are in the peaks. 1-8% (rate-dependent) of events are in peaks for other runs. Time differences between adjacent events.

7. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 7 Power Spectra Distribution of signal power as a function of frequency –A common estimate of the power spectrum of a signal is the squared modulus of its discrete Fourier transform. Characteristic power spectrum of: –A single sine wave is a delta function. –Signals with random variations have continuum power. –Poisson counting statistics introduce a frequency- independent "noise floor,” expected level of 2.0.

8. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 8 Power Spectrum (low frequency) Power at low frequencies is due to rate change during run. In the absence of varying incident rates or deadtime, this would show a constant level of 2.0, due to Poisson noise. Power spectrum for frequencies < 2 Hz.

9. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 9 Power Spectrum (high frequency) 600 Hz sine is caused by 1.7 ms peak in waiting-time distribution. Decrease at high frequencies –caused by interference with a similar oscillation from the smaller 1.6 ms peak. Deviation at low frequencies – caused by interference with deadtime effect. If not corrected, –will severely impact scientific utility of data –real effects can be smaller than this by a factor of 10-100 or more. Power spectrum for frequencies > 2 Hz.

10. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 10 Power Spectrum (filtered) High frequency power spectrum after removal of events in 1.6,1.7 ms peaks. Deviations from 2.0 are due to deadtime. This is better but it is still pretty far from ideal.

11. GLAST LAT ProjectI&T Meeting – Feb 12, 2003 W. Focke 11 Summary A preliminary timing analysis in the EM VDG data indicated –Photon rates were consistent with expected values –Measured a long deadtime (2.3 ms) possibly due to hardware limitations Future Work and Recommendations –Would like to understand cause of 1.7 ms peak –Recommend keeping SBC time, even after arrival of a real GLT, as a check on and backup to GLT time.