Tools for applying the Karhunen-Loève Transform (KLT) to SETI radio observations, developed as part of my BSc thesis in Physics.
Applicazione della trasformata di Karhunen-Loève per la ricerca di segnali telemetrici extraterrestri Giovanni Nicola D'Aloisio — Laurea Triennale in Fisica, Università di Napoli "Federico II", A.A. 2024/2025
This work was carried out during an internship at the Osservatorio Astronomico di Cagliari, within the Breakthrough Listen project, using data from the Sardinia Radio Telescope (SRT), Green Bank Telescope (GBT) and Parkes Observatory.
The KLT (proposed for SETI by Claudio Maccone in the 1990s) decomposes a stochastic signal into an optimal eigenbasis, making it effective at isolating strong, narrowband radio-frequency interference (RFI) from the noise floor — RFI that would otherwise mask faint, Doppler-drifting technosignatures. Here the KLT is integrated into the turboSETI pipeline as a pre-cleaning step, operating on filterbank (.fil) data.
The method was validated on archival data (bright FRBs, Voyager 1) and tested on ON-OFF observations of exoplanet targets such as TOI-1422 b (SRT, C-band 5.7-7.7 GHz and K-band 18.5-18.8 GHz), with synthetic technosignatures injected via setigen to quantify detection performance before and after KLT cleaning.
| Script | Description |
|---|---|
01a_test_env.py |
pytest suite checking conda env, Python/library versions, and basic functionality of numpy, scipy, astropy, setigen, your, blimpy |
01b_set_env.sh |
activates the klt conda environment and runs 01a_test_env.py plus a smoke test on a sample filterbank |
03a_clean_filterbank.py |
core cleaning script: spectral-kurtosis RFI flagging and/or KLT-based RFI template subtraction (PCA on the covariance matrix, -var_frac controls the eigenvalue threshold, -klt_win sets the channel window) |
03b_run_klt_from_list.sh |
runs 03a_clean_filterbank.py -klt over a list of filterbank files |
03c_run_cleaning.sh |
example invocations of 03a_clean_filterbank.py with fixed parameters on stored experiment data |
04_compare_filterbanks.py |
checks whether two filterbanks contain the same data (regression testing) |
05a_signal_injector.py |
injects synthetic technosignatures into a filterbank, either via setigen or a custom FakeTelemetry Gaussian-drift model |
05b_setigen_for_klt.sh |
example runs of 05a_signal_injector.py (setigen and telemetry-simulation modes) |
07_run_plotting.sh |
batch-runs 08_plot_data.py over all .fil files in a directory for a set of channel ranges |
08_plot_data.py |
plots a filterbank waterfall (with optional time/frequency range selection) |
11_check_hits.sh |
computes the average number of turboSETI hits per target from .dat output files |
13_run_turboseti.sh |
runs turboSETI on a list of filterbank files |
14_check_memory.sh |
reports current RAM usage on the observation server |
Python 3.11.4, conda env klt. Key dependencies: astropy, blimpy, your, sigpyproc, setigen, turbo-seti, numpy, scipy, matplotlib.
unset PYTHONPATH
source /opt/conda/bin/activate klt- https://thepetabyteproject.github.io/your/0.6.6/ipynb/Reader/
- https://sigproc.sourceforge.net/sigproc.pdf
- https://sigproc.sourceforge.net/
- https://docs.astropy.org/en/stable/coordinates/index.html
- https://blimpy.readthedocs.io/en/latest/overview.html
- https://turbo-seti.readthedocs.io/en/latest/
- C. Maccone, The KLT (Karhunen-Loève Transform) to extend SETI searches to broad-band and extremely feeble signals, Acta Astronautica 67 (2010) 1427-1439