UC Berkeley

Over the 60 years since the first published search for radio technosignatures, relatively established methods have come about for the detection and analysis of narrowband radio signals in the Search for Extraterrestrial Intelligence (SETI). Generally, this involves using position-switching to take multiple observations on and off the target of interest and detecting raw narrowband signals via a matched filter that linearly fits Doppler accelerations to each signal. High quality candidates are identified in those cases in which detected signals appear to persist through all ``ON'' observations and do not appear in any ``OFF'' observations, implying that the signal source is localized on the sky.

While these techniques are used commonly across the field, they are by no means perfect. Typical signal detection methods can struggle to detect all signals present when there are regions of time-frequency space that are densely populated, which means potential technosignatures may be missed. Furthermore, since we are fundamentally searching for a type of signal that has never been found before, it is difficult to quantify the accuracy of detection algorithms. Even the sky localization technique is not necessarily sufficient for distinguishing against radio frequency interference (RFI), which takes on many unknown morphologies and various intensity modulations.

In this thesis, we aim to push the bar forward for both signal detection and candidate identification (filtering). First, we develop a machine learning (ML) methodology for localizing narrowband signals in frequency and Doppler drift rate. Not only is this procedure faster over datasets than the standard tree detection algorithm, we train ML models to identify up to 2 signals within each stretch of 1024 frequency bins, whereas the standard algorithm can only identify 1 in the same stretch. From this work, we develop and independently present \setigen, an open source Python library for the synthesis and injection of artificial narrowband signals into real observational data, both directly in the form of Stokes I intensities in time-frequency space and in the form of raw complex voltages taken by baseband recorders. \setigen can and has been used for creating large datasets used in ML training, validating detection algorithms using injection-recovery analysis, and developing new candidate filters.

Then, we propose a new candidate identification strategy based on plasma scattering from the interstellar medium (ISM). Theoretically, narrowband radio signals traveling through ionized plasma in our galaxy will exhibit strong intensity scintillations from multi-path scattering. As technosignature searches are typically tuned towards continuous narrowband signals, these scintillations should be imprinted on the received intensities and therefore detectable under the right observing parameters. Finally, we conduct a dedicated search for scintillated technosignatures towards the Galactic center and Galactic plane, for which the timescales of scintillation will be contained within individual observations. In addition to the specific scintillation analysis, we apply the sky localization filter to identify technosignature candidates. Though we do not find evidence of technosignatures, we set limits on their presence and comment about the feasibility of detection in the future.