Technical details for Advanced LIGO and Virgo event data releases

Data files span a range of lengths from 32 second to 4096-second data associated with Advanced LIGO and Virgo events. The data is expressed as a time series of strain values at 4 kHz or 16 kHz, and is accompanied by two bitfield integers at 1 Hz, for Data Quality and Injections. Thus for each second several quality flags are combined into an integer with binary bits, for example the flags PASS FAIL FAIL PASS PASS PASS would be encoded as binary 100111, or 39 in decimal.

Data quality for event data release

Data quality categories are used to characterize the quality of the data at a given time. Data quality categories for this data release are defined and described in the paper "Characterization of Transient Noise in Advanced LIGO Relevant to Gravitational Wave Signal GW150914" (available from LIGO DCC or arXiv).
The meaning of the bits for Data Quality are:
bitnamedescription
0DATAData is available
1CBC_CAT1Data passes CBC CAT1 test
2CBC_CAT2Data passes CBC CAT2 test
3CBC_CAT3Data passes CBC CAT3 test
4BURST_CAT1Data passes Burst CAT1 test
5BURST_CAT2Data passes Burst CAT2 test
6BURST_CAT3Data passes Burst CAT3 test

In particular, the DATA bit channel defines the seconds when LIGO or Virgo data are available: if the DATA flag is not set, the strain data will have NaN values. The majority of the released data pass all data quality categories, so the data quality bitfield integer is 1111111 binary = 127 decimal. However there are instances where the data fail a particular data quality category and as such the bitfield deviates from 1111111 binary = 127 decimal for a few seconds. See the LOSC tutorials page for information about how to access these, or the tutorial on segment lists.

For a description of Advanced LIGO and Virgo data quality see: Generally:
  • DATA (Data Available): Failing this level indicates that LIGO or Virgo data is not available at this time.
  • CAT1 (Category 1): Failing a data quality check at this category indicates a critical issue with a key detector component not operating in its nominal configuration. Since these times indicate a major known problem these times are identical for each data analysis group. Times that fail CAT1 flags are not available as open data -- not available at this web site.
  • CAT2 (Category 2): Failing a data quality check at this category indicates times when there is a known, understood physical coupling to the gravitational wave channel. This might include times of high seismic activity.
  • CAT3 (Category 3): Failing a data quality check at this category indicates times when there is statistical coupling to the gravitational wave channel which is not fully understood.

Injections for the event data releasers

The second bitfield integer is to show injection status, where a 1 indicates NO injection.
bitnamedescription
0NO_CBC_HW_INJNo CBC hardware injections
1NO_BURST_HW_INJNo BURST hardware injections
2NO_DETCHAR_HW_INJNo DETCHAR hardware injections
3NO_CW_HW_INJNo CW (pulsar) hardware injections
4NO_STOCH_HW_INJNo Stochastic hardware injections

During any of the O2 event data releases, there were no transient hardware injections, and therefore all these bits are also equal to 1. However, there were (non-transient) pulsar injections in both LIGO detectors. Therefore the injection bitfield integer is 10111 binary = 23 decimal. In Virgo O2 data, this may instead be 11111 = 31 decimal, indicating pulsar injections were not active in Virgo. Note that the amplitudes of these pulsar injections are such that they are probably too small to detect over the noise during the 32-seconds to a maximum of 4096-seconds of data being released. If you need more information, please contact us at losc@ligo.org.

Spectral Lines

There are in addition a variety of instrumental spectral lines in the data. These include harmonics of the mirror suspension resonances at ~500 Hz; electrical power lines at 60 Hz and harmonics; and calibration lines throughout the spectrum. Such lines are not relevant to the detection or analysis of this event.

Advanced LIGO and Virgo data are not calibrated under 10 Hz

Advanced LIGO and Virgo detector noise rises rapidly below 10 Hz; it is many orders of magnitude larger than any conceivable gravitational wave strain signal. Further, the data made available on LOSC are aggressively high-pass-filtered at 8 Hz in order to avoid downstream signal processing problems, so it doesn't properly represent either signal or noise at those low frequencies. In particular, any DC offsets have no significance. In the Initial LIGO (S5 and S6) datasets, data below 40 Hz is similarly uncalibrated and suppressed. The 8 Hz roll-off is visible in this example low frequency spectrum.

Version 2 files

For events GW150914, LVT151012, and GW151226, the tables point to version 2 of the data files (v2). These v2 data files were posted in October of 2016. They differ from v1 in that they use an updated version of the LIGO calibration (C02 instead of C01). Also, the v1 4096 Hz files included a minor time-offset, roughly 1 ms, introduced during the down-sampling process. This has been corrected in the v2 files.

Data before and after noise subtraction

Some events include data files both before and after noise subtraction. In these cases, the filenames include the string "C00" to refer to data before noise subtraction, or the string "CLN" to indicate data after noise subtraction.

Advanced Virgo data noisier above 2 kHz during O2

During O2 the Advanced Virgo data are contaminated at frequencies above 2 kHz by a significant amount of spectral and transient noise, which are not fully characterized. The gravitational-wave signals detected during O2 do not overlap with this noisier frequency band in the upper part of the observed spectrum.

Other Details

LOSC strain data have been decimated from 16384 Hz to 4096 Hz. This has been done with the decimate method in the scipy package, with the "zero-phase" option.

For events released before August of 2017, files of duration 32 sec and 4096 sec are computed independently, by decimation, from the 16 kHz data. There may be a slight difference between them, at the level of one part in 10^8. This is because the decimation filter used was an infinite impulse response filter. This is also true for the data "after noise subtraction", as with GW170814.