Algorithm

The input time series is first low passed to remove any power above the new Nyquist frequency. There are two available low pass filters:

1. defaultButterworth The input time series has a time domain low pass filter applied by the LALDButterworthREAL4TimeSeries() function from the tdfilters package. The filter order is $20$ and the attenuation is $0.1$ at the new Nyquist frequency. Since the butterworth filter is applied forwards and backwards there is no time delay or phase shift of the output data. The disadvantage of the butterworth filter is that since it is an IIR filter, it is not possible to determine the exact length of corrupted data at the start and end of the output time series. Care should be taked to avoid using these regions of data in subsequent filtering.

2. LDASfirLP The input time series has a time domain low pass filter applied by the LALDIIRFilterREAL4Vector() function from the tdfilters package. This applies an FIR filter with coefficents generated by the LDAS firlp() dataconditioning API action. FIR coefficents are available for downsampling by a factor of 2, 4 or 8. An attempt to downsample by any other ratio will cause an error. The FIR coeffients are equivalent to the default resample() action in the dataconditioning API.

   The FIR filter causes the a time delay of 10 points in the output time series and corruption of the first $n$ points. $n$ is given by the order of the filter by $n = 2 \times 10 \times q$, where $q$ is the resampling ratio. To account for this, we do the following:

   1. The first $n/2$ points of the time series are deleted.
   2. The whole time series is shifted $n/2$ points ``left'' to remove the time delay.
   3. The first and last $n/2$ points of the output time series are set to zero to make sure that the corrupted data contains sensible numbers and not garbage or Inf.
   This means that there is no time delay in the output time series, but the first and last $n/2q$ points are set to zero. Care should be taked to avoid using these regions of data in subsequent filtering. If the debug level is set to LALWARNING, a message is printed reporting how many points at the start and end of time series are corrupted.

   The filter coefficents used were produced by LDAS-CIT running version 0.7.0 of LDAS. See the LDAS dataconditioning API documentation for more information.