BurstMon performance for S3 data

Conditions:

• code version: preliminary release, 04/27/2004
  
• input data H1:LSC-AS_Q channel, GPS start time 751974320 sec
• data length: 4096 sec
• sampling rate after decimation: 8192 Hz
• frequency resolution: 64 Hz
  
• injection waveform: Sine-Gaussian 235 Hz, Q=9 (duration=0.0085 s)
• time stride: 60 sec
  
• number of simultaneous injections at the same 1 minute interval: 10
• number of repetitions per time stride: 10
• number of different injection amplitudes for efficiency meausurements: 5
• total number of injections per time stride: 10x10x5=500

    The picture shows 1 minute  trend (blue line) for h₅₀ calculated by BurstMon  for actual H1:LSC-AS_Q channel data stretch from S3 run. Initial  approximation amp0 is also shown (magenta line). The average relative standard error for h₅₀ calculated from Least Square fit  is  4%, this value is achieved using total 500 number of injections. The series of  h₅₀ shows variations with relative RMS  26% so that  the error for  h₅₀ is reasonably small to show actual variations of detector sensitivity. The statistical correlation between these series of  h₅₀ and amp0 is 0.76.

    For the sake of  calculation speed 10 injections with the same amplitude but at different moments of time were made at each 60 sec time stride. In order to achieve required accuracy this procedure were repeated 10 times for each stride with random selection of injection moments. Then another 4 series of 100 injections each were performed with different waveform amplitudes. Fitting all 5 measurements by appropriate curve (sigmoid) allowed to calculate final estimation of  h₅₀. Calculations for each 60 seconds  of data stretch consumed 37 seconds of CPU time for BurstMon running on Pentium III 1 GHz computer operated by Linux.