⚡ electrical engineering

Adaptive Bayesian Radar Scheduling for Hidden Space Debris Population Estimation

This paper demonstrates that an adaptive Bayesian radar scheduling approach, validated on EISCAT radar data and tested in simulations, significantly reduces uncertainty in estimating the population of untrackable space debris compared to fixed or random pointing strategies, thereby improving risk assessment for critical satellite services.

Original authors: Aahan Sachdeva, Richard Linares, Edward Oughton

Published 2026-10-13

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Aahan Sachdeva, Richard Linares, Edward Oughton

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

High above the Earth, in the silent vacuum where satellites orbit, a hidden danger drifts unseen. While we can track large pieces of space junk, like defunct satellites or spent rocket stages, there are millions of smaller fragments, ranging from the size of a paint fleck to a soda can, that are too small to be cataloged individually. These invisible shards travel at speeds of thousands of miles per hour, capable of punching through spacecraft hulls and disabling critical technology that supports life on the ground, from global internet access to GPS navigation. Because we cannot see them all, we do not know exactly how many are out there or where they are most concentrated. This uncertainty makes it difficult to assess the true risk to the satellites that keep our modern world running. To solve this, scientists have long used radar to sweep the sky, but traditionally, these radars have been passive observers, pointing in one fixed direction and waiting for debris to drift through their beam. The question researchers now face is whether a radar could do better by actively choosing where to look next, learning from each detection to find the hidden population more efficiently.

A team of researchers, led by a high school student working with university scientists, set out to answer this question by combining real-world radar data with a sophisticated computer model. They began by analyzing recordings from two powerful European radars, one located in Norway and the other on an island in the Arctic Circle. These radars had been pointed at the same spot in the sky for 24 hours, collecting raw electrical signals as debris passed through their beams. The team first had to teach their computer how to distinguish a real piece of space junk from the static noise of the universe. They tested several methods for spotting these faint signals, eventually selecting a technique that adjusts its sensitivity based on the immediate background noise, much like a person adjusting their eyes to a dark room. This method proved highly effective, identifying 155 potential debris events in the Norwegian data and 222 in the Arctic data. Crucially, when they tried to match these events to known lists of space objects, a significant number—106 from the first radar and 119 from the second—did not match any known catalog. These were the "hidden" pieces of debris the researchers needed to study.

To understand how to find these hidden objects more quickly, the team built a virtual world inside a computer. They created 300 different simulated universes, each with its own unique distribution of space debris, and then ran five different strategies for how a radar should point its beam. The first strategy was to keep the radar fixed in one spot, just as the real radars had done. The second was to point the radar at random angles. The third was to follow a pre-planned list of angles decided before the experiment began. The fourth and fifth strategies were "adaptive," meaning the radar would look at where it had already looked and decide where to point next based on what it had just learned. The fifth strategy added a special rule: it would prioritize looking at altitudes where important satellites for communication, navigation, and military use were flying.

The results of these simulations showed that the old way of doing things—keeping the radar fixed—was the least efficient. The adaptive strategies, where the radar changed its aim based on new information, learned about the hidden debris population much faster. By the end of the 24 simulated observations, the adaptive approach that prioritized critical satellite altitudes reduced the uncertainty about the debris population by more than 40 percent compared to the fixed strategy. It also outperformed the random and pre-planned strategies, though the improvement over the pre-planned list was smaller. Interestingly, the extra step of prioritizing specific satellite services provided only a tiny, statistically insignificant boost over a simpler adaptive approach that treated all altitudes equally. This suggests that simply changing direction based on what the radar finds is the most powerful tool, and that the specific goal of protecting certain services does not drastically change the best path to finding the debris.

The study concludes that while a pre-planned schedule of different viewing angles is far better than staring at one spot, a radar that can adapt in real-time offers a measurable advantage in reducing our ignorance about space debris. However, the researchers are careful to note that these findings come from computer simulations, not a live radar campaign where the antenna actually moves. The next step, they suggest, is to test these ideas in the real world with a radar capable of changing its aim, to see if the theoretical gains hold up when faced with the complexities of actual atmospheric conditions and the time it takes to move a massive antenna. Until then, this work provides a strong mathematical argument that the way we look at the sky matters just as much as the technology we use to look.

Technical Summary: Adaptive Bayesian Radar Scheduling for Hidden Space Debris Population Estimation

Problem Statement
A significant portion of the space debris population, particularly fragments between 1 and 10 cm, lacks reliable orbital records in public catalogs. These "hidden" objects pose a collision risk capable of disabling critical satellite services (e.g., broadband, GPS), yet their abundance and distribution remain uncertain. Traditional space surveillance often focuses on tracking known objects or using fixed-pointing "beam park" observations, which do not actively optimize for reducing uncertainty about the hidden population. The core challenge addressed is how to efficiently estimate the size and distribution of this uncataloged debris population using limited radar observation time, and whether adaptive scheduling—choosing where to point next based on prior data—outperforms fixed, random, or preplanned schedules.

Methodology
The study employs a two-pronged approach combining empirical radar analysis with Bayesian simulation.

Radar Data Processing and Validation:
- Data Source: The authors analyzed 24 hours of simultaneous beam park data from the EISCAT Tromsø (930 MHz) and Svalbard (500 MHz) radars collected in January 2018.
- Detection Pipeline: Raw voltage data were processed using a Generalized Matched Filter (GMF) to search for moving objects. To distinguish true signals from noise, the authors compared fixed Signal-to-Noise Ratio (SNR) thresholds against Constant False Alarm Rate (CFAR) methods.
- Detector Selection: Through testing with 4,000 simulated echoes (including Swerling fluctuation models) and shuffled real data to estimate false alarm rates, the Cell-Averaging CFAR (CA-CFAR) detector was selected. It was tuned to a threshold of 1.75 times the local mean to maximize recovery of weak, short signals while maintaining a false alarm rate below one event per 24 hours.
- Event Association: Detected signals were grouped into events and cross-referenced with cataloged objects using SGP4 propagation. Events without confident matches were classified as "unresolved." The Tromsø data yielded 155 events (106 unresolved), while the Svalbard data yielded 222 events (119 unresolved) using the same detector settings without retuning.
Bayesian Population Model and Simulation:
- Model Structure: A Bayesian model was constructed to estimate the hidden debris population across ten 15-km altitude bands (915–1,065 km). The model estimates a multiplier ( $\theta_j$ ) for each band relative to the Orbital Debris Engineering Model (ORDEM) baseline.
- Adaptive Scheduling: The study simulated 300 "worlds" with varying debris populations and radar responses. Five scheduling strategies were compared over 24 one-hour observations:
  - Fixed: Radar points at a single elevation (68.5°).
  - Random: Random selection of 49 possible elevations.
  - Preplanned: A sequence of elevations chosen in advance to maximize information gain.
  - Adaptive (Unweighted): Selects the next elevation to maximize expected information gain regarding the total population.
  - Adaptive (Service-Weighted): Similar to the unweighted version but prioritizes altitude bands hosting satellites providing critical services (navigation, communications, Earth observation, military/intelligence).
- Optimization: The adaptive scheduler uses a Gaussian approximation to predict the expected reduction in posterior uncertainty (weighted by service importance) for each potential elevation, selecting the one that minimizes future uncertainty.

Key Contributions

Detector Transferability: Demonstrated that a CA-CFAR detector tuned on Tromsø UHF data could be applied to Svalbard VHF data without retuning, successfully identifying 222 events, though the false alarm rate on shuffled Svalbard data slightly exceeded the strict Tromsø selection criteria.
Adaptive vs. Static Scheduling: Provided simulation evidence that adaptive pointing significantly reduces uncertainty in hidden debris population estimates compared to fixed or random pointing, even with the same observation budget.
Service-Aware Scheduling: Investigated whether weighting observations toward altitudes critical to satellite infrastructure yields additional benefits over general population estimation, finding a marginal but statistically insignificant improvement.

Results

Uncertainty Reduction: After 24 observations, adaptive scheduling with service weights reduced the remaining weighted uncertainty by:
- 41.2% compared to fixed pointing.
- 15.4% compared to random pointing.
- 5.35% compared to a preplanned schedule.
Accuracy: The adaptive schedules also produced lower Root Mean Square Error (RMSE) in population estimates compared to fixed and random schedules.
Service Weighting Impact: The additional benefit of adding service weights to the adaptive algorithm was estimated at 1.55% reduction in uncertainty (95% interval: −0.47% to 3.53%). This difference was not statistically significant ( $p = 0.400$ ), suggesting that for the specific geometry and debris distribution modeled, general population learning and service-weighted learning converge on similar optimal elevations.
Efficiency: Adaptive schedules reached a 25% uncertainty reduction target approximately 2–3 observations faster than fixed or random schedules.

Significance and Claims
The paper claims that adaptive radar pointing is a viable method for reducing uncertainty regarding hidden space debris populations using the same number of observations as traditional methods. The results suggest that the majority of the performance gain over fixed pointing can be achieved by a well-chosen preplanned schedule, with adaptive adjustments offering a smaller, incremental improvement.

The authors remain modest regarding the specific value of "service-aware" scheduling in this context; while the concept is theoretically sound for prioritizing risk, the simulation did not yield a statistically significant advantage over unweighted adaptive scheduling. The study concludes that the primary value lies in the shift from static to adaptive (or carefully preplanned) observation strategies. The authors note that these findings are based on simulation and fixed-pointing historical data, and they call for a future real-world campaign with a steerable radar to validate these gains under actual operational constraints, such as beam movement time and varying sensitivity.

More like this