An Attempt to Search for Unintended Electromagnetic Radiation from Starlink Satellites with the 21 Centimeter Array: Methodology and RFI Characterization

Imagine the night sky as a giant, quiet library where astronomers are trying to read the faintest whispers of the universe—specifically, the ancient radio signals from the very beginning of time. Now, imagine thousands of noisy delivery trucks (Starlink satellites) zooming overhead, honking their horns and playing loud music. This is the challenge facing modern radio astronomers today.

This paper is essentially a report card from a team of astronomers in China who tried to listen to these "noisy trucks" using a specific radio telescope called the 21 Centimeter Array (21CMA). Here is the story of what they did, what they found, and what they learned, explained in simple terms.

1. The Mission: Catching a Ghost in the Machine

The astronomers wanted to find Unintended Electromagnetic Radiation (UEMR). Think of this as "radio static" or "leakage" that the satellites accidentally send out while they are trying to do their real job (sending internet data).

The Problem: These satellites are like bright flashlights in a dark room. If they leak even a tiny bit of light, it can blind the sensitive cameras (telescopes) trying to look at the dark corners of the universe.
The Tool: They used the 21CMA, a massive radio telescope in Xinjiang, China. However, for this specific experiment, they only used one small section (called a "pod") of the whole telescope.
- Analogy: Imagine trying to hear a whisper in a stadium by holding a single, tiny ear trumpet. It's a very difficult task!

2. The Strategy: Predicting the Traffic

Since they couldn't watch the sky 24/7 (it would take too much computer memory), they acted like traffic controllers.

They used a digital map (called TLE data) to predict exactly when and where the Starlink satellites would fly over their "ear trumpet."
They set up their recording to only turn on when a satellite was predicted to pass directly overhead.
Analogy: Instead of recording a whole day of traffic, they only hit "record" when they knew a specific delivery truck was about to drive past their house.

3. The Result: The "Silent" Satellite

Did they hear the Starlink static?
No. They found nothing.

Why?
It wasn't because the satellites were silent; it was because their "ear trumpet" wasn't sensitive enough.

Analogy: The astronomers were trying to hear a mosquito buzzing in a hurricane. The Starlink leakage is like that mosquito. The 21CMA's single pod is like a person with slightly muffled ears. The "noise" of the universe (sky background) was too loud, and the telescope wasn't sensitive enough to pick up the faint satellite leak.
They calculated that they would need to upgrade their equipment (by linking many "ear trumpets" together) to hear these faint signals clearly.

4. The Twist: The "Lightning" Noise

Even though they didn't hear the satellites, they did hear something else: Power Line Arcing.

What happened: They saw sudden, sharp bursts of radio noise that looked like they could be from a satellite.
The Investigation: They analyzed the rhythm of the noise. They found it happened exactly 100 times a second (matching the 50Hz electricity cycle).
Analogy: It was like hearing a crackling sound and thinking, "Is that a spaceship engine?" But then they realized, "Oh, that's just a loose wire on a high-voltage power line nearby sparking in the wind."
They used a fancy new AI tool (called SAM 2) to automatically spot these sparks in their data, proving that the noise was from Earth, not space.

5. The Success Story: Catching a Different Satellite

While they missed the Starlink leakage, they successfully caught a different kind of satellite: ORBCOMM.

These are older satellites that send data for tracking ships and trucks. They are louder and easier to hear.
The team built a custom software tool (a "decoder ring" called orbdemod) to translate the satellite's radio language.
Analogy: They tuned their radio to a specific station, and instead of static, they heard a clear message: "I am Satellite #108."
Why this matters: This proved their "traffic controller" strategy worked perfectly. They correctly predicted when and where the satellite would be, and their software correctly decoded its message. If they could catch the loud ORBCOMM, they know their system is working; they just need a bigger "ear trumpet" to catch the quiet Starlink.

The Big Picture: What's Next?

The team concludes that:

Starlink leakage is real, but it's currently too faint for this specific telescope setup to hear.
Earthly noise (power lines) is a major distraction that needs to be filtered out.
The Solution: In the future, they plan to link all the "pods" of the telescope together to create a super-sensitive "super-ear." This will allow them to finally hear the faint whispers of the satellites and understand exactly how much they are interfering with our view of the universe.

In short: They tried to catch a ghost with a small flashlight, found a spark from a nearby wire instead, but proved they could catch a loud bird. Now, they are building a giant spotlight to finally catch that ghost.

Here is a detailed technical summary of the paper "An Attempt to Search for Unintended Electromagnetic Radiation from Starlink Satellites with the 21 Centimeter Array: Methodology and RFI Characterization."

1. Problem Statement

The rapid expansion of Low-Earth-Orbit (LEO) megaconstellations, particularly SpaceX's Starlink, poses a significant threat to radio astronomy through Unintended Electromagnetic Radiation (UEMR). While previous studies (e.g., by LOFAR and SKA-Low prototypes) have detected broadband and narrowband UEMR from Starlink satellites in the 40–188 MHz range, observational data from East Asia remains limited.

Challenge: Detecting UEMR is difficult because satellite signals can be orders of magnitude stronger than the faint 21-cm Epoch of Reionization (EoR) signal, yet the UEMR itself can be weak (tens to hundreds of Jy) and easily masked by terrestrial Radio Frequency Interference (RFI) or limited by telescope sensitivity.
Goal: This study aims to search for Starlink UEMR using the 21 Centimeter Array (21CMA) in Xinjiang, China, while simultaneously developing and validating a robust observation pipeline and RFI characterization framework.

2. Methodology

The researchers employed a targeted observation strategy using a single pod (E8) of the 21CMA, rather than the full array, to test the methodology before implementing full beamforming.

Observation Strategy:
- Targeting: Used Two-Line Element (TLE) data and the Skyfield Python library to predict satellite transits.
- Selection Criteria: Defined a transit event as a satellite having an apparent declination $> 85^\circ$ relative to the 21CMA (which points at the North Celestial Pole with a $5^\circ$ field of view).
- Data Acquisition: Recorded raw voltage data (480 Msps, int16) only during predicted transit windows to maximize efficiency. Five observation sessions were conducted on April 15, 2025, targeting Starlink (v1.0 to v2-mini D2C) and Chinese Qianfan (G60) satellites.
Sensitivity Analysis:
- Calculated the System Equivalent Flux Density (SEFD) for a single pod.
- Parameters: Effective area ( $A_{eff}$ ) $\approx 218 \, m^2$ at 150 MHz; System temperature ( $T_{sys}$ ) $\approx 350 \, K$ (dominated by sky noise).
- Resulting SEFD: $\approx 4.43 \, kJy$ .
- Detection Threshold: After time/frequency averaging (35 ms time resolution, 0.234 MHz frequency resolution), the theoretical sensitivity was $\Delta S \approx 48.95 \, Jy$ , yielding a 3 $\sigma$ detection limit of 146.85 Jy.
Signal Processing & RFI Analysis:
- Modulation Spectrum Analysis: Applied a secondary FFT to time-series power data to identify periodic modulation.
- AI-Based RFI Identification: Tested SAM 2 (Segment Anything Model 2), a video-based AI model, to track transient broadband features across consecutive spectral frames, leveraging its memory module to distinguish non-stationary signals.
- Demodulation: Developed a custom Python package, orbdemod, to demodulate ORBCOMM satellite downlinks (SDPSK modulation) to validate the transit prediction accuracy.

3. Key Results

A. Absence of Detectable Starlink UEMR

Null Result: No broadband UEMR from Starlink satellites was detected in the 110–188 MHz range.
Cause: The study attributes this to insufficient sensitivity. The 3 $\sigma$ detection threshold (146.85 Jy) significantly exceeds the typical UEMR flux densities reported in other studies (which range from a few to ~100 Jy).
Frequency Dependence: At lower frequencies (<100 MHz), sensitivity degrades further due to increased galactic synchrotron noise and filter roll-off effects, making detection even less likely with a single pod.
Future Potential: The authors estimate that implementing Tied-Array Beamforming (TAB) with 12 coherent pods would improve sensitivity to $\approx 6.23 \, Jy$ , making detection feasible.

B. RFI Characterization: Power Line Arcing

Observation: The dynamic spectra revealed impulsive, broadband bursts (tens to >100 MHz bandwidth) lasting ~34 $\mu$ s, primarily in the 50–100 MHz range.
Identification: These were not satellite UEMR. Modulation spectrum analysis revealed a dominant 100 Hz peak (with a 50 Hz harmonic), characteristic of power line arcing from nearby high-voltage infrastructure.
AI Validation: The SAM 2 model successfully tracked these transient arcing events across time sequences, demonstrating its potential for automated, real-time RFI flagging in high-cadence radio data.

C. Validation via ORBCOMM Demodulation

Detection: Narrowband signals from ORBCOMM satellites were detected near 137 MHz.
Verification: The orbdemod package successfully demodulated the SDPSK signal.
- The recovered satellite ID was 0x26 (Decimal 38).
- Based on mapping logic, this corresponds to ORBCOMM FM 108.
Significance: This matched the satellite with the highest apparent declination (63.9 $^\circ$ ) in the vicinity, confirming the accuracy of the TLE-based transit prediction and identification framework, even though the satellite did not strictly enter the main $5^\circ$ FoV (signal likely entered via side lobes).

4. Key Contributions

Methodological Framework: Established a robust pipeline for predicting and observing satellite transits using TLE data and fixed-antenna arrays, validated by successful ORBCOMM signal recovery.
Sensitivity Benchmarking: Provided a quantitative analysis showing that single-pod observations of 21CMA are insufficient to detect typical Starlink UEMR, defining the specific sensitivity gap (146 Jy vs. ~50 Jy signals).
RFI Characterization: Identified and characterized local power line arcing as a major source of impulsive broadband interference in the 50–100 MHz band, distinguishing it from satellite emissions.
AI Integration: Demonstrated the feasibility of using SAM 2 for tracking transient RFI in radio dynamic spectra, offering a new tool for automated interference mitigation.
Software Release: Open-sourced the orbdemod Python package for demodulating ORBCOMM downlinks.

5. Significance and Future Outlook

Scientific Impact: This study clarifies that the lack of detected Starlink UEMR in East Asia is likely due to instrumental sensitivity limits rather than the absence of emissions. It sets a clear path for future observations using multi-beam coherent beamforming to achieve the necessary sensitivity.
Regional Data: It fills a geographical gap in UEMR studies, providing crucial data from East Asia for global interference modeling.
Mitigation Strategies: The identification of local power line arcing and the successful testing of AI-based RFI tracking provide immediate strategies for cleaning data in future low-frequency radio astronomy experiments.
Next Steps: The authors plan to upgrade 21CMA to full beamforming capabilities to detect UEMR definitively and extend the study to other constellations (e.g., China's Qianfan/G60) and varying operational modes of Starlink satellites.