# Chandra Absolute Timing Accuracy

## Definition of Terms

To properly understand the various elements involved in establishing absolute time, one has to be aware of the different components and time systems. In the Chandra FITS files, one will find the following quantities, either as header keywords or table columns.

```Time         Time stamp, in MET seconds (i.e., elapsed seconds since MJDREF)
TIMEZERO     First order clock correction
MJDREF       The MJD(TT) of 1998.0(TT): 50814 d
TIMESYS      Time system; either TT or TDB
CLOCKAPP     Boolean: any correction applied?
TIERABSO     Absolute time accuracy
TIMEPIXR     Relative position of time stamp in bin
```
Time systems:
```TT           Terrestrial Time
TDB          Barycentric Dynamical Time
TCG          Geocentric Coordinate Time
TCB          Barycentric Coordinate Time
```
MET (Mission Elapsed Time) is counted as elapsed seconds since 1998.0 (TT). MJD (Modified Julian Day) is defined as JD - 2400000.5. A more detailed description of time scales and time systems can be found in the Chandra Time Tutorial.

Time can be expressed in the following ways:

```MET (accuracy 4 s)      Time
MET (accuracy 100 µs)   Time + TIMEZERO
MET (accuracy 5 µs)     Time + TIMEZERO + clockCorr;
MJD(UTC)                MET + 50814 d - leap seconds
MJD(TT)                 MET + 50814 d
MJD(TDB)                TT + baryCorr(TDB) = TT + tau + (TDB - TT)
MJD(TCB)                TT + baryCorr(TCB) = TT + tau + (TCG - TT) + (TCB - TCG)
```

## Clock Calibration

The DSN (Deep Space Network) performs clock calibrations about twice a week. The accuracy is required to be better than 100 µs. We have performed a scientific calibration of absolute time in Chandra data and concluded that it is possible to achieve an absolute time accuracy of about 4 µs. See the note on absolute time calibration.

## Barycenter Corrections

The pathlength corrections (tau) applied by the tool axBary are accurate to better than a fraction of a µs, based on JPL ephemerides DE-200 or DE-405. The tool faseBin uses the same code.

The values for TDB-TT that these tools derive are not to be trusted to better than a few µs. However, if one were to improve on these, one should really switch from using TDB to TCB.

## Sampling and the Position of Time Stamps

In Chandra FITS files, three types of data are encountered:

1. Integrated data.

Also histogram or binned data. All science data (including event lists) belong to this category. A certain quantity, such as an event count, is integrated over a specific time range (bin). In binned modes, a histogram is presented with contiguous time bins. In event modes, individual time bins are enumerated, each with an implied event count of 1 and a width equal to the time resolution.

The keyword TIMEPIXR indicates the relative position of the time stamp in the bins: a value of 0.0 indicates that the time stamp refers to the beginning of the bin; 0.5 the center of the bin; 1.0 the end of the bin. If TIMEPIXR is absent, a value of 0.5 is to be assumed, conform the FITS standard. Most Chandra HRC FITS files have TIMEPIXR=0.0; i.e., for HRC, the time stamps always refer the the start of the bin. For ACIS files, TIMEPIXR usually is 0.5.

2. Sampled data with known sampling interval

A physical quantity, like a voltage or a temperature, or a bit mask, such as a command, is sampled at regular, fixed, and known intervals. This is the case for housekeeping files (particularly instrument housekeeping files) that have a non-zero value for the keyword DELTAT. All that can be said is that the reading took place in the time interval of DELTAT seconds following the time stamp.

3. Sampled data with unknown sampling interval

A physical quantity, like a voltage or a temperature, or a bit mask, such as a command, is sampled at unknown intervals. This is the case for housekeeping files (particularly spacecraft housekeeping files) that have a zero value for the keyword DELTAT. All that can be said is that the reading took place in the time interval between the previous and the current time stamp. In retrospect, a value of 1.0 might have been more appropriate for TIMEPIXR, in this case.

In the last two cases, it may be possible to be more precise, but this requires detailed knowledge of the subsystem, as well as particulars of the data, and no general rule can be given.