AI, Radio Astronomy, and the Search for Life Beyond Earth

Speaker’s Background and Passion for Science Communication

Ramiro Caisse Saide, affiliated with the University of Manchester, has a long-standing commitment to education, particularly in deep learning and astronomy. His work includes leading deep learning workshops in Mozambique for flood detection using satellite data, and teaching the Search for Extraterrestrial Intelligence (SETI) at the Hartebeesthoek Radio Astronomy Observatory in South Africa. Beyond academia, he’s a prolific software builder, having developed platforms like Mashamba (for agricultural diagnosis using AI), Mosbrain (a no-code AI model builder for students), and wildlife monitoring systems using computer vision. He also founded Matisu, an open-source data labeling platform, and directs the Mozambican Astronomical Society, where he built a full-stack application for planet and star observation. His passion for understanding life in the universe drives his research.

The Quest for Extraterrestrial Life

The fundamental question driving SETI is the apparent uniqueness of life on Earth, despite billions of stars and planets in the observable universe. Preserving life on Earth is a key motivation, with astronomy providing insights into potential threats like comets. The “Pale Blue Dot” image from Voyager highlights Earth’s isolation and the necessity for detectable “technosignatures” or other intelligent activity for us to be found from space. Ramiro notes Microsoft co-founder Paul Allen’s crucial role in funding the first instrument dedicated to SETI, the Allen Telescope Array (ATA). The scientific basis for interstellar communication was established by Cocconi and Morrison in 1959, demonstrating the detectability of signals through interstellar medium.

Conditions for Life and Detection

Searching for life involves identifying planets capable of sustaining it, where detectable technology could develop, and where signatures persist over long periods. Mars, despite past potential, lacks a strong magnetic field crucial for protecting life from ionizing radiation. Venus showed traces of phosphine, a potential biosignature linked to anaerobic activity on Earth. Exoplanet discoveries, especially systems like Trappist-1, show planets in the “habitable zone” or “Goldilocks zone” where liquid water can exist. Terrestrial atmospheres, whose compositions are largely driven by biological activities, offer another avenue for detecting life through chemical ratios. Over 6,000 exoplanets have been discovered and confirmed to date.

Radio Astronomy and SETI Methodologies

Radio astronomy studies the universe by detecting radio waves, which penetrate Earth’s atmosphere, unlike most of the electromagnetic spectrum. SETI primarily employs a passive listening approach rather than actively broadcasting (Active SETI or METI). Key facilities include the Green Bank Telescope and China’s FAST, the largest single-dish radio telescope. The Breakthrough Listen project, primarily privately funded, is a leading initiative in this field, providing open-source data.

Key Concepts in Radio Astronomy:

• Angular Resolution: Large telescopes are needed for high resolution (governed by $\lambda/D$), but mechanical limits necessitate other techniques.
• Interferometry: Combining signals from multiple smaller telescopes (like “smaller buckets” collecting raindrops) to achieve the resolution of a much larger virtual telescope. This technique was crucial for imaging the M87 black hole, requiring global synchronization with hydrogen clocks.
• Beamforming: A technique used by multi-antenna arrays (like the ATA with 42 antennas) to focus on specific regions within the primary field of view by correcting signal delays. This is analogous to 5G towers’ “bolt beam” technology and is vital for mitigating Radio Frequency Interference (RFI) by comparing different beams.

Radio astronomy generates big data, exemplified by the ATA processing 1.4 terabits/second (equivalent to 45 HD movies/second) and the M87 black hole image requiring two years to process five petabytes of data transported via hard drives, making AI essential.

Earth’s Own Technosignatures: Leakage Studies

Understanding Earth’s radio leakage helps predict what an advanced civilization might detect from us. Studies show that RFI is a significant problem, even for protected spectrum users like NASA’s soil moisture satellites. Ramiro’s research investigates:

• Mobile Tower Leakage: Simulating aggregated radio output from global mobile towers revealed that most radiation escapes the atmosphere when the planet is rising or setting relative to a particular direction. The northern hemisphere showed significant leakage, suggesting a focus for observations from potential northern stars.
• Radar Leakage: Tracking planes at airports generates highly detectable signatures from space. Simulations indicate this is one of Earth’s most prominent technosignatures, detectable even from nearby stars like Barnard’s Star.

AI for Technosignature Detection

Radio astronomy pipelines produce millions of hits (potential signals), making manual inspection impractical. This is where AI becomes critical. Ramiro’s current research focuses on deep learning strategies to discriminate between genuine technosignatures and RFI.

SETI Search Criteria (leveraging AI for discrimination):

1. Narrowband Features: Natural phenomena produce wide spectral features; artificial signals are expected to be extremely narrow (e.g., < 3 Hz bandwidth).
2. Doppler Effect/Acceleration: Terrestrial RFI has no relative acceleration, thus no frequency shift. True extraterrestrial signals would exhibit Doppler shifts due to relative motion.
3. Spatial Filtering (On/Off Strategy): A true extraterrestrial signal should only be present when the telescope is pointed directly at the source, disappearing when pointed away (e.g., using beamforming in interferometry).
4. Instrumental Artifacts: AI can be trained to recognize patterns from non-linear instrument responses (harmonics).
5. Satellite Mimicry: Satellites can produce narrowband signals with relative acceleration, making them difficult to distinguish without sophisticated RFI mitigation techniques, which AI can significantly enhance through simulation and classification.

Multimodal Deep Learning Approach:

• Modalities: Spectrograms (frequency, power, time) and time series (in-phase and quadrature - I/Q - signal representations).
• Fusion Strategies:
  • Feature Fusion: Concatenating features from both modalities before the final prediction layer.
  • Decision Fusion: Running independent models for each modality and then averaging their predicted probabilities.
• Models: CNN, EfficientNet, and 1D CNN for fusion layers.
• Preliminary Results: Surprisingly, adding the time series modality did not improve generalization; the single spectrogram model achieved the best performance (around 95% accuracy in preliminary tests).

Future Work

Ramiro plans to expand the research by:

• Increasing the number of simulated RFI modulations (e.g., radar, GSM, GPS) to improve discrimination against diverse interference sources.
• Implementing different sequential models like XLSTM and Temporal Convolutional Networks (TCNs) to potentially improve generalization, especially for time-series data.

He concludes with Carl Sagan’s profound statement: “The absence of the evidence is not the evidence of the absence,” underscoring the ongoing nature and philosophical underpinnings of the SETI search.

• Tags:
• Machine Learning
• Deep Learning
• Big Data
• SETI
• Radio Astronomy
• Technosignatures
• Signal Processing
• Multimodal Learning
• RFI Mitigation
• MeerKAT
• Exoplanets
• Interferometry
• Beamforming