⚡ electrical engineering

A Knowledge-Centric Communication For Autonomous Cislunar Networks

This paper proposes a knowledge-centric communication framework for autonomous cislunar networks that utilizes a digital twin to integrate delayed observations and uncertainty quantification into an evolving Operational Knowledge State, introducing metrics like Knowledge Entropy and Knowledge Freshness to optimize mission success under light-speed and orbital constraints.

Original authors: Afan Ali, Daniel Benevides da Costa, Ali Arshad Nasir

Published 2026-08-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Afan Ali, Daniel Benevides da Costa, Ali Arshad Nasir

Original paper licensed under CC BY 4.0 (http://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

Imagine you are the captain of a spaceship, but you are so far away from Earth that your radio messages take seconds, sometimes minutes, to arrive. By the time you hear a reply, the situation has already changed. In the old days of space exploration, missions were short trips where you could talk to mission control, get instructions, and go. But the future is different: we are building a permanent "cislunar" neighborhood—a busy ecosystem of habitats, robots, and satellites orbiting between Earth and the Moon. In this new world, communication isn't just about sending messages; it's about keeping a living, breathing map of reality that updates itself even when the radio goes silent.

To understand the challenge, think of three tools we usually use to navigate the unknown. First, there's Communication, which is like a courier service trying to deliver letters as fast as possible. Second, there's the Digital Twin, which is like a perfect, real-time video game copy of your spaceship that shows exactly where everything is. Third, there's Artificial Intelligence (AI), the smart brain that tries to guess what to do next. Usually, we assume these tools work together perfectly: the courier delivers the letter, the video game updates instantly, and the brain makes a decision based on the latest truth. But in deep space, the "instant" part is broken. Light takes time to travel, so your video game is always showing you the past, and your courier is always late. If you try to make a decision based on "what is happening right now," you are actually guessing based on old news. This paper asks a big question: How do we make smart decisions when we can never know the exact present moment?

The authors of this paper, researchers from King Fahd University of Petroleum and Minerals, propose a new way to think about this problem called a Knowledge-Centric Communication framework. Instead of trying to force a perfect, instant video of the universe (which is physically impossible due to the speed of light), they suggest we should build a "Digital Twin" that admits what it doesn't know. Imagine your spaceship's computer isn't just a map, but a detective who keeps a notebook. When the radio is working, the detective writes down facts. But when the radio goes silent because a planet blocks the view, the detective doesn't freeze the page. Instead, the detective starts thinking: "Okay, the last time I saw the robot, it was moving this fast. Since I haven't heard from it for ten minutes, it's probably moved this far, but I'm not 100% sure. My confidence is dropping."

This is where the paper introduces two new ways to measure how "smart" the computer is feeling. The first is Knowledge Entropy. Think of this as a "Confusion Meter." When the radio is working, the meter is low because the computer has clear facts. When the radio goes silent, the meter slowly climbs up, showing that the computer is getting more confused and uncertain about where things are. The second is Knowledge Freshness. This is like a "Staleness Clock." It doesn't just tell you when you last got a message; it tells you how much that message has probably "rotted" or become outdated since you got it.

The researchers tested this idea using computer simulations of a future Moon network and even checked it against real data from a tiny satellite called Longjiang-2 that has been orbiting the Moon. They found that their "detective" approach works much better than the old way. In the old "State-Centric" way, the computer would just freeze its map at the last known location and pretend nothing changed, which is dangerous because things do change. In the new "Knowledge-Centric" way, the computer keeps updating its "Confusion Meter." If the meter gets too high (too much confusion), the system knows to be extra careful, perhaps switching to a different radio link or slowing down a robot, rather than blindly guessing.

One of the most exciting findings is that this system gets smarter when it combines information from different sources. The paper shows that if one satellite sees a robot and another satellite sees the same robot from a different angle, fusing those two delayed, blurry pictures creates a much clearer, more confident picture than either one could do alone. In fact, in some tricky situations, no single satellite had enough information to make a safe decision, but when they combined their "detective notebooks," they could meet the mission's safety requirements. This proves that for future space exploration, we don't need perfect, instant data; we need a system that is honest about its uncertainty and knows how to reason even when the lights go out.

The paper also showed that this "detective" system is flexible. They trained it on one type of orbit and then tested it on a completely different one without re-teaching it the basics, and it still worked well. However, they also warned that if the computer's "Confusion Meter" isn't calibrated correctly—if it thinks it's more sure than it really is—the whole system can fail. So, the key isn't just having a smart AI, but having an AI that knows exactly how much it doesn't know.

In short, this paper suggests that the future of space travel isn't about faster radios or perfect maps. It's about building systems that are comfortable with the unknown. By measuring how "confused" and "stale" our knowledge is, we can make safer, smarter decisions even when we are light-minutes away from home, turning the silence of space from a dangerous void into a manageable gap in our understanding.

Technical Summary: A Knowledge-Centric Communication Framework for Autonomous Cislunar Networks

1. Problem Statement

Future lunar infrastructure is shifting from mission-specific, short-duration expeditions to a persistent, heterogeneous cislunar ecosystem (e.g., NASA's Artemis, Lunar Gateway, commercial services). While communication, digital twin, and artificial intelligence (AI) research have advanced independently, they share a critical assumption: that autonomous decisions can be supported by sufficiently accurate operational information reconciled with ground truth.

This assumption fails in cislunar environments due to three fundamental constraints:

Propagation Delay: Bounded by the speed of light, preventing real-time state synchronization.
Intermittent Visibility: Dictated by orbital geometry rather than deployment choices, leading to unpredictable communication outages.
Decision Latency: Autonomous actions must often be executed before corroborating observations arrive.

In terrestrial or near-Earth domains, synchronization delays are treated as engineering nuisances to be minimized because a "ground truth" reference is eventually available. In cislunar space, this reference is structurally unavailable. Consequently, existing state-centric paradigms (which assume the last observation remains valid until the next update) and standard digital twins (which rely on continuous synchronization) are insufficient. They fail to represent the growing uncertainty during communication gaps, leading to overconfident and potentially unsafe autonomous decisions.

2. Methodology: The Knowledge-Centric Digital Twin (KCDT)

The authors propose a Knowledge-Centric Communication framework centered on a Knowledge-Centric Digital Twin (KCDT). Unlike traditional approaches that optimize channels or synchronize states, the KCDT treats the Operational Knowledge State itself as the object of reasoning.

Core Architecture:
The KCDT continuously fuses four heterogeneous streams into an evolving belief state for each communication link:

Delayed Observations: Past measurements of signal-to-noise ratio (SNR), Doppler shift, and propagation delay.
Communication Physics: Deterministic models (e.g., Friis equation, two-body orbital propagation) that constrain plausible states independent of specific observations.
Learned Models: Neural networks trained to predict residual dynamics (multipath, scintillation, hardware imperfections) and their associated uncertainty.
Mission Objectives: Constraints defining the required precision and freshness for specific infrastructure elements.

Mathematical Formulation:
The system maintains a posterior belief $b_{ij}(t) \approx \mathcal{N}(\mu_{ij}(t), \Sigma_{ij}(t))$ for each link $(i, j)$ .

Prediction Step: Between observations, the belief evolves via physics models, with uncertainty ( $\Sigma$ ) growing according to process noise ( $Q_{proc}$ ) to reflect the aging of knowledge.
Correction Step: Upon receiving a delayed observation, the belief is updated using Bayesian precision weighting, where the observation's uncertainty (predicted by the learned residual model) acts as the measurement noise covariance.

Key Metrics:
To quantify the quality of the Operational Knowledge State, the authors define two metrics over the fused Bayesian belief:

Knowledge Entropy: The differential entropy of the posterior belief. It quantifies "how much remains unknown." Unlike state-centric uncertainty (which remains constant during outages), Knowledge Entropy increases during communication gaps and decreases only when delayed observations are fused.
Knowledge Freshness: A generalization of Age of Information (AoI) that measures the temporal validity of the fused state, accounting for the time since the last evidence was collected.

3. Key Contributions

Paradigm Shift: The paper argues for a shift from "state-centric" communication (faithful reproduction of symbols or states) to "knowledge-centric" reasoning, where the system reasons about what it knows and how confident it is, rather than just what it observes.
Theoretical Properties:
- Fusion Dominance: Proved that fusing independent observations via Bayesian precision weighting strictly reduces posterior uncertainty compared to any single source.
- Fusion Necessity: Demonstrated scenarios where no single observation source meets mission requirements, but a fused state does, making distributed fusion a necessity rather than an optimization.
- Knowledge Aging: Proved that without new observations, uncertainty (entropy) monotonically increases, preventing the system from maintaining false confidence during outages.
Metrics: Introduction of Knowledge Entropy and Knowledge Freshness as the primary objects of optimization, replacing traditional metrics like Bit Error Rate (BER) or standard AoI for autonomous decision-making.

4. Results

The framework was evaluated through Monte Carlo simulations in cislunar environments and validated against real flight data from the Longjiang-2 lunar micro-satellite.

Distributed Fusion Performance: Across 5,000 randomized realizations, Bayesian fusion consistently reduced posterior uncertainty by an average of 31.5% compared to the best individual source. In "Fusion Necessity" scenarios, fusion achieved 100% mission success where individual sources achieved 0%.
Transferability: A residual model trained on one orbital regime (Regime A) generalized to other regimes (Regime B) with minimal fine-tuning (few-shot learning), provided the predictive variance was physically bounded to prevent extrapolation failure during long outages.
Decision Layer Comparison: The KCDT outperformed state-centric heuristics, physics-only predictors, and data-driven Kalman filters in terms of Root Mean Square Error (RMSE) and calibration in Regime A. However, in Regime B/hop2, the KCDT performance was conditional on the residual model's calibration; when the model's parameters were mismatched to the target regime (using Regime A parameters on hop2), the KCDT exhibited significantly higher RMSE (2.464 dB vs. 0.397 dB) and calibration error (29.531 vs. 0.176) compared to the state-centric baseline. The Probability of Mission Completion (PMC) for KCDT (0.983) was slightly lower than the state-centric baseline (0.987) in this specific regime, where visibility was near-continuous and the advantage of knowledge-centric reasoning was marginal. The results highlight that knowledge-driven decision-making remains fundamentally dependent on an appropriately calibrated uncertainty model.
Real-World Validation (Longjiang-2):
- The framework successfully modeled the actual orbital geometry and link budget of Longjiang-2.
- It correctly identified an in-flight anomaly (a temperature-compensated oscillator jump) by showing a rise in Knowledge Entropy, whereas a state-centric model showed flat entropy (structural blindness).
- Real-world multi-path and multi-station reception data confirmed that fusing independent observations reduced RMSE compared to single-source processing.
Scalability: The computational complexity scales linearly with the number of active links ( $O(|E(t)|)$ ), making it feasible for large-scale cislunar networks.

5. Significance and Claims

The paper claims that the proposed framework is essential for the next era of autonomous space infrastructure. Its significance lies in:

Bridging the Gap: It unifies communication physics, digital twin theory, and AI to address the specific structural limitations of cislunar operations (light-speed delay and intermittent visibility).
Enabling Autonomy: By explicitly quantifying uncertainty and knowledge freshness, the KCDT allows autonomous systems to make calibrated decisions (e.g., proactive link switching, resource hedging) even when the physical state is unknown or stale.
Existence Proof: The validation against Longjiang-2 data serves as an existence proof that knowledge-centric behaviors (entropy growth during outages, fusion benefits) hold true in real-world space environments, not just simulations.

The authors conclude that this work opens a research direction where "operational knowledge" becomes a central object of communication theory, moving beyond the transmission of bits to the management of what the infrastructure knows and how it acts on that knowledge. They acknowledge limitations, such as the use of reduced-order orbital models for the Lunar Gateway and the need for further work on distributed asynchronous fusion and foundation model trustworthiness, but assert that the core theoretical properties hold regardless of specific orbital geometries.