TransientX is a one command line high performance transient search software (filtterbank and psrfits (--psrfits) are both supported).
- boost > 1.56
- PlotX (https://github.com/ypmen/PlotX)
- XLibs (https://github.com/ypmen/XLibs)
- sofa (https://www.iausofa.org/2020_0721_C/sofa_c-20200721.tar.gz)
- ./bootstrap
- ./configure --prefix=[install_path] LDFLAGS="-L/path_to_sofa" CPPFLAGS="-I/path_to_sofa"
- make
- make install
docker pull ypmen/pulsarx
- export YMW16_DIR=/install_path/src/ymw16
- transientx_fil -h for help
transientx_fil -v -o test -t 4 --zapthre 3.0 --fd 1 --overlap 0.1 --ddplan ddplan.txt --thre 7 --maxw 0.1 --snrloss 0.1 -l 2.0 --drop -z kadaneF 8 4 zdot -f *.fil
or using a json file
transientx_fil -v -o test -t 4 -l 2.0 -c examples/transientx_config.json -f *.fil
- --saveimage (save image for AI classifier)
replot_fil -v -t 4 --zapthre 3.0 --td 1 --fd 1 --dmcutoff 3 --widthcutoff 0.1 --snrcutoff 7 --snrloss 0.1 --zap --zdot --kadane 8 4 7 --candfile test.cands --clean -f *.fil
-
When searching in high-resolution data (e.g., <50us), please apply multiple search strategies to detect bursts with various timescales: (1) For microsecond bursts, use a smaller data block size (-l), a smaller DBSCAN radius (-r), a smaller maximum search pulse width (--maxw), etc;(2) For millisecond bursts, you can use larger values for the above parameters. This approach can help avoid slowing down the clustering process, as discussed in the Benchmark Section of the TransientX paper TransientX.
-
use the option --iqr when there are many strong bursts, which can improve the estimation of baseline and noise level.
Please cite https://arxiv.org/abs/2401.13834, if you are using TransientX.
Thanks for very helpful suggestions and bug reports from Guðjón Henning Hilmarsson, Emma Carli and PKU pulsar group.