Integrated Computing Platform for Detection and Tracking of Unidentified Aerial Phenomena

Brief

Cloete et al. (2023) describe the full-stack computing infrastructure for the Galileo Project's first observatory-class UAP monitoring station, spanning sensor data ingestion, edge AI inference, event-driven targeting of a narrow-field PTZ camera, and cloud archival. The sensor suite, seven LWIR FLIR Boson 640 cameras, one NIR all-sky camera, one optical all-sky imager, passive FM-reflection radar, acoustic arrays, and an RF spectrum analyzer, collectively generates 9.52 TB/day, of which 8.79 TB originates from the camera cluster alone. Real-time processing runs on NVIDIA Jetson Xavier units (384-core GPU, 8–16 GB RAM) deployed at the edge; a ZeroMQ publisher-subscriber bus routes detection events to PTZ actuator control within the latency window required to slew onto a moving target. Full-scale future phases are projected at 9–10 TB/day and would require an estimated 58–65 simultaneous Starlink connections for complete data egress, a logistical constraint the team acknowledges will force selective event-windowed upload.

Metadata

Category

Hub & Overview

Venue

Journal of Astronomical Instrumentation

Type

Peer-reviewed

Year

2023

Authors

Richard Cloete, Phillip Bridgham, Sergei Dobroshinsky, Carson Ezell, Andriy Fedorenko, Frank Laukien, Sarah Little, Abraham Loeb, Eric Masson, Matthew Szenher, Wesley Andres Watters, Abigail White

DOI

10.1142/S2251171723400081

Access

Open access

Length

3.5 M

Programs

Galileo Project

Instruments

FLIR Boson 640 LWIR camera (7-camera Dalek array), ZWO ASI462 NIR all-sky camera, Alcor OMEA 9C optical all-sky camera, Beacon 8 PTZ narrow-field optical camera, Skywatch passive FM-reflection radar, AMOS acoustic array (infrasonic/audible/ultrasonic), Spectre RF spectrum analyzer, NPACKMAN environmental/particle/magnetic sensor, NVIDIA Jetson Xavier edge compute nodes, ThinkMate Threadripper PRO offline analysis server

Data sources

ADS-B aircraft transponder feeds, Starlink satellite internet (egress), GPS time synchronization, FM broadcast signals (passive radar illuminator)

Tags

UAP-instrumentation, technosignature, anomaly-detection, edge-computing, computer-vision, passive-radar, multi-sensor-fusion

Key points

• The Dalek camera dome houses seven 50° LWIR FLIR Boson 640 cameras in a hemispherical array plus one 95° zenith camera, providing continuous all-sky thermal infrared coverage at 30–60 FPS and 640×512 pixels per channel.p.3
• Total instrument data production is 9.52 TB/day; the Dalek camera cluster alone accounts for 8.79 TB/day, while the NPACKMAN environmental sensor contributes only 0.000049 TB/day.p.6
• Passive radar system Skywatch has a detection range of approximately 150 km, far exceeding any wide-field camera, and records 300 seconds of narrowband FM radio data at 1 MHz sample rate, cycling every 1024 seconds (~85 acquisition cycles per day).p.3
• Phase 1 network egress relies on a single Starlink node providing 8–25 Mbps upload, supporting roughly 158 GB/day; full-throughput future phases would require an estimated 58–65 simultaneous Starlink connections.p.6
• Edge computing nodes are NVIDIA Jetson Xaviers (6-core ARM, 384-core Volta GPU, 8–16 GB LPDDR4x RAM, 1 TB SSD), selected specifically for on-site AI inference to minimize round-trip latency for PTZ camera targeting.p.7
• Inter-process event routing uses ZeroMQ in a brokerless publisher-subscriber pattern to pass object-detection events from the vision pipeline to PTZ actuator control with minimal computational overhead.p.8
• Time synchronization is implemented via NTP at millisecond precision, with a planned upgrade path to PTP/IEEE 1588 for nanosecond precision, a prerequisite for multi-site triangulation.p.5
• ADS-B aircraft transponder data is ingested as an external reference layer to identify and filter known aircraft from the detection pipeline in real time.p.4

Verbatim

• “The Galileo Project (GP) was established in the summer of 2021 to investigate anomalous aerospace phenomena including ambiguous interstellar objects (ISOs) such as Oumuamua (Siraj et al. , 2022) and UAP (Loeb, 2022).”

  p.2

• “it is estimated that approximately 58 – 65 Starlink connections would be necessary to manage this volume of data.”

  p.6

Most interesting

• The NPACKMAN environmental sensor generates 0.000049 TB/day, less than 50 MB, making it more than 179,000 times smaller a data producer than the camera cluster, yet it is the instrument monitoring magnetic field anomalies and energetic particle counts.
• Full-bandwidth data egress at projected future throughput (9–10 TB/day) would require 58–65 simultaneous Starlink satellite internet nodes, a cost and physical footprint the authors explicitly call 'prohibitively expensive.'
• The passive radar (Skywatch) works entirely off ambient FM radio broadcast reflections, no active transmitter, and can detect objects at ranges up to ~150 km, blind-penetrating cloud cover that defeats all optical and infrared cameras.
• The NIR all-sky camera (ZWO ASI462) operates at 136 FPS, more than twice the frame rate of any LWIR channel, providing sub-10 ms temporal resolution for fast transient events in the visible-to-near-infrared band.
• The acoustic system (AMOS) spans more than four decades of frequency: infrasonic sampling at 50 Hz, audible at 44.1 kHz, and ultrasonic at 512 kHz, enabling detection of sound signatures invisible to standard audio recording.