BLISS is a toolkit for finding narrowband doppler-drifting signals. This is frequently used to search for technosignatures.
BLISS is able to use cuda-accelerated kernels with deferred execution and memoization for flagging, noise estimation, integration, and hit search.
Running bliss requires
- libhdf5-cpp-103 (double check!)
- hdf5-filter-plugin
- bitshuffle
Prebuilt wheels are available for the following runtimes:
- cpu only:
pip install dedrift - cuda 11:
pip install dedrift-cuda11x - cuda 12:
pip install dedrift-cuda12x
This project builds with cmake and is set up to be built as a python package with pyproject.toml. Building and running depend on the following libraries/tools:
- cmake
- gcc / clang capable of C++17 and supporting your version of cuda
- libhdf5-dev
- (optional) libcapnp-dev
The standard cmake workflow should configure everything for a build:
mkdir build
cd build
cmake .. # -G Ninja # if you prefer ninja
make -j $(($(nproc)/2)) # replace with make -j CORES if you don't have nproc
The python package is partially set up in bliss/python/bliss. During the build process, the C++ extensions are built and placed in this package. In cmake development mode, this is placed in build/bliss/python/bliss and configured to be updated with any file changes as they occur (each file is symlinked) and new files will be added at the next build.
pyproject.toml configures the python package and uses py-cmake-build as the build backend to get the required C++ extensions built and packaged appropriately. You can build this package with standard tools such as pip install . and python -m build.
Inside build/bland/tests you will have a test_bland executable. You can run those tests to sanity check everything works as expected.
Inside build/bliss/ you will have a justrun executable. You can pass a fil file to that, but right now this is a useful debugging and sanity check target, the output will be underwhelming.
Inside build/bliss/python you should have a pybliss.cpython-311-x86_64-linux-gnu.so or similarly named pybliss shared library. This can be imported in python and exposes functions and classes for dedoppler searches. Inside the notebooks directory there is an rfi mitigation visual.ipynb jupyter notebook that walks through several of the functions with plots showing results. That is the best way to get a feel for functionality.
The following is example usage for Voayger-1 recordings from the Green Bank Telescope
import bliss
data_loc = "/datag/public/voyager_2020/single_coarse_channel/old_single_coarse/"
cadence = bliss.cadence([[f"{data_loc}/single_coarse_guppi_59046_80036_DIAG_VOYAGER-1_0011.rawspec.0000.h5",
f"{data_loc}/single_coarse_guppi_59046_80672_DIAG_VOYAGER-1_0013.rawspec.0000.h5",
f"{data_loc}/single_coarse_guppi_59046_81310_DIAG_VOYAGER-1_0015.rawspec.0000.h5"
],
[f"{data_loc}/single_coarse_guppi_59046_80354_DIAG_VOYAGER-1_0012.rawspec.0000.h5"],
[f"{data_loc}/single_coarse_guppi_59046_80989_DIAG_VOYAGER-1_0014.rawspec.0000.h5"],
[f"{data_loc}/single_coarse_guppi_59046_81628_DIAG_VOYAGER-1_0016.rawspec.0000.h5"]])
cadence.set_device("cuda:0")
working_cadence = cadence
working_cadence = bliss.flaggers.flag_filter_rolloff(working_cadence, .2)
working_cadence = bliss.flaggers.flag_spectral_kurtosis(working_cadence, .05, 25)
noise_est_options = bliss.estimators.noise_power_estimate_options()
noise_est_options.masked_estimate = True
noise_est_options.estimator_method = bliss.estimators.noise_power_estimator.stddev
working_cadence = bliss.estimators.estimate_noise_power(working_cadence, noise_est_options)
int_options = bliss.integrate_drifts_options()
int_options.desmear = True
int_options.low_rate = -500
int_options.high_rate = 500
working_cadence = bliss.drift_search.integrate_drifts(working_cadence, int_options)
working_cadence.set_device("cpu")
hit_options = bliss.drift_search.hit_search_options()
hit_options.snr_threshold = 10
cadence_with_hits = bliss.drift_search.hit_search(working_cadence, hit_options)
hits_dict = bliss.plot_utils.get_hits_list(cadence_with_hits) auto voyager_cadence = bliss::cadence({{"/datag/public/voyager_2020/single_coarse_channel/old_single_coarse/single_coarse_guppi_59046_80036_DIAG_VOYAGER-1_0011.rawspec.0000.h5",
"/datag/public/voyager_2020/single_coarse_channel/old_single_coarse/single_coarse_guppi_59046_80672_DIAG_VOYAGER-1_0013.rawspec.0000.h5",
"/datag/public/voyager_2020/single_coarse_channel/old_single_coarse/single_coarse_guppi_59046_81310_DIAG_VOYAGER-1_0015.rawspec.0000.h5"
},
{"/datag/public/voyager_2020/single_coarse_channel/old_single_coarse/single_coarse_guppi_59046_80354_DIAG_VOYAGER-1_0012.rawspec.0000.h5"},
{"/datag/public/voyager_2020/single_coarse_channel/old_single_coarse/single_coarse_guppi_59046_80989_DIAG_VOYAGER-1_0014.rawspec.0000.h5"},
{"/datag/public/voyager_2020/single_coarse_channel/old_single_coarse/single_coarse_guppi_59046_81628_DIAG_VOYAGER-1_0016.rawspec.0000.h5"}});
auto cadence = voyager_cadence;
cadence.set_device("cuda:0");
cadence = bliss::flag_filter_rolloff(cadence, 0.2);
cadence = bliss::flag_spectral_kurtosis(cadence, 0.1, 25);
cadence = bliss::estimate_noise_power(
cadence,
bliss::noise_power_estimate_options{.estimator_method=bliss::noise_power_estimator::STDDEV, .masked_estimate = true}); // estimate noise power of unflagged data
cadence = bliss::integrate_drifts(
cadence,
bliss::integrate_drifts_options{.desmear = true,
.low_rate = -500,
.high_rate = 500,
.rate_step_size = 1});
cadence.set_device("cpu");
auto cadence_with_hits = bliss::hit_search(cadence, {.method=bliss::hit_search_methods::CONNECTED_COMPONENTS,
.snr_threshold=10.0f});
auto events = bliss::event_search(cadence);
bliss::write_events_to_file(events, "events_output");