A Joint JSCC-Resource Allocation Framework for QoS-Aware Semantic Communication in LEO Satellite-based EO Missions

This paper proposes a joint source-channel coding and resource allocation framework for LEO satellite-based Earth observation missions that minimizes transmit power under quality-of-service constraints by approximating the complex relationship between compression, signal quality, and image reconstruction through a curve-fitting model and a dedicated optimization algorithm.

The Gist

Here is an explanation of the paper using simple language, creative analogies, and metaphors.

🛰️ The Big Picture: The "Space Delivery" Problem

Imagine a fleet of high-tech delivery trucks (Low Earth Orbit satellites) circling the Earth. Their job is to take high-resolution photos of the planet (like spotting a forest fire or monitoring a storm) and send them back to a central warehouse (ground stations).

The Problem:

1. The Cargo is Too Heavy: These photos are massive. Sending them all takes up a huge amount of space in the truck's cargo hold (bandwidth) and burns a lot of fuel (power).
2. The Truck is Running on a Battery: Satellites have very limited power. They can't just blast all that data at full speed; they need to be efficient.
3. The Road is Bumpy: As the satellite zooms around the Earth, the connection to the ground gets stronger and weaker (like driving through a tunnel or up a hill).
4. The Goal: The ground team doesn't always need every single pixel of the photo. They just need to know what is in the photo (e.g., "There is a fire here," or "The river is flooded").

The Old Way (JPEG):
Traditionally, the satellite takes the photo, compresses it into a smaller file (like a ZIP file), and sends the whole file. If the connection is bad, the file gets corrupted, and the image looks blurry or broken. It's like mailing a fragile glass vase; you have to wrap it in so much bubble wrap (redundant data) that the package becomes heavy and expensive to ship.

The New Way (Semantic Communication):
This paper proposes a smarter way. Instead of sending the whole "glass vase," the satellite acts like a smart translator. It looks at the photo, understands the meaning, and only sends the "essence" of the image.

• Analogy: Instead of sending a 100-page novel, the satellite sends a 1-page summary that captures the plot. If the ground team needs to know if there's a fire, they get the message "Fire detected" without needing the high-definition picture of the trees.

---

🧠 The Secret Sauce: The "Brain" (JSCC)

The paper uses a technique called Joint Source-Channel Coding (JSCC). Think of this as a super-smart AI chef.

• Traditional Cooking: You chop the vegetables (source coding), put them in a box (channel coding), and ship them. If the box gets crushed, the veggies are ruined.
• The AI Chef (JSCC): The chef tastes the dish while cooking it. They know exactly how much salt (data) is needed to make the flavor (image quality) perfect, even if the delivery truck hits a bump. They adjust the recipe in real-time based on the road conditions.

📉 The Challenge: The "Black Box"

The researchers faced a tricky math problem. They knew that:

1. Changing how much the AI compresses the image changes the quality.
2. Changing the signal strength (SNR) changes the quality.
3. But they didn't have a simple formula to predict exactly how these three things interact. It was like trying to bake a cake without knowing how the oven temperature affects the rise.

The Solution: The "Curve-Fitting" Map
To solve this, the researchers created a map (a curve-fitting model). They ran thousands of simulations, took notes on how the image quality changed with different settings, and drew a smooth line connecting the dots.

• Metaphor: Imagine you are trying to guess how far a ball will fly based on how hard you throw it. Instead of doing complex physics every time, you just look at a chart you made yesterday that says, "If I throw it at angle X with force Y, it lands here." This map turned a impossible math problem into a solvable one.

🚀 The Strategy: The "Smart Dispatcher" (JCRRA)

The paper proposes an algorithm called JCRRA (Joint Compression Ratio-Resource Allocation). Think of this as a Smart Dispatcher managing the delivery trucks.

The Dispatcher has to decide three things for every photo:

1. How much to summarize? (Compression Ratio): Do we send a 1-page summary or a 5-page summary?
2. Which road to take? (Resource Allocation): Which frequency channel is clearest right now?
3. How hard to push the engine? (Power): How much power do we use to send it?

The Algorithm's Logic:
Instead of guessing, the Dispatcher uses the "Map" (the curve-fitting model) to calculate the perfect balance.

• If the road is bumpy (bad signal), it sends a shorter summary (more compression) but uses just enough power to get it through.
• If the road is smooth (good signal), it sends a slightly longer summary to get better quality, but saves power because the signal is strong.

🏆 The Results: Saving Fuel

The researchers tested their "Smart Dispatcher" against two other methods:

1. Greedy JSCC: A smart chef who picks a random summary size and sticks with it.
2. Greedy JPEG: The old-school method of compressing the whole file and hoping for the best.

The Outcome:

• Fuel Savings: The new method saved a massive amount of power (up to 6–7 dBm, which is a huge difference in satellite terms).
• Flexibility: It handled the "bumpy roads" (changing satellite positions) much better than the old methods.
• Quality: It ensured the ground team got exactly the quality they asked for, no more, no less.

💡 The Takeaway

This paper is about teaching satellites to be smarter, not just louder. Instead of shouting the whole story to the ground, they whisper the most important parts. By using AI to understand the meaning of the images and math to optimize the delivery, we can send more data, faster, using less battery power.

In short: It's the difference between mailing a heavy, fragile box of rocks (old way) and sending a text message that says, "The rocks are here, and they are red" (new way). The text message gets there faster, cheaper, and still tells you everything you need to know.

Technical Summary

Here is a detailed technical summary of the paper "A Joint JSCC-Resource Allocation Framework for QoS-Aware Semantic Communication in LEO Satellite-based EO Missions."

1. Problem Statement

The paper addresses the critical challenge of transmitting massive volumes of high-resolution Earth Observation (EO) data from Low Earth Orbit (LEO) satellites to ground users under strict constraints:

• Limited Power Budget: LEO satellites have constrained on-board power, making energy-efficient transmission crucial.
• Dynamic Environment: LEO movement causes time-varying channel conditions (fading, slant range changes), complicating resource allocation.
• QoS Requirements: Downstream tasks (e.g., disaster response) require a specific level of image reconstruction quality, quantified by the Structural Similarity Index Metric (SSIM), rather than just raw bit accuracy.
• Optimization Complexity: The goal is to minimize total downlink transmit power while satisfying SSIM requirements. This requires jointly optimizing:
  1. User Association: Which sub-channel serves which user at which time slot.
  2. Power Control: Transmit power allocation.
  3. Semantic Compression: Selecting the optimal Joint Source-Channel Coding (JSCC) compression ratio.

The core difficulty lies in the implicit, non-analytical relationship between JSCC parameters (compression ratio), channel conditions (SNR), and the resulting image quality (SSIM), combined with the mixed-integer nature of the variables (discrete compression options and binary association).

2. Methodology

The authors propose a comprehensive framework to solve this non-convex optimization problem:

A. System Model

• Architecture: A LEO satellite serves $K$ users over $T$ time slots using $S$ sub-channels.
• Semantic Communication (SemCom): Instead of transmitting raw bits, the system uses a Deep Learning-based JSCC encoder to extract semantic features from images and a decoder to reconstruct them.
• Constraints:
  • Orthogonality (one user per sub-channel per time slot).
  • Power budget limits.
  • Delay constraints (images must be transmitted within a specific window).
  • QoS constraints (Expected SSIM $\geq$ Target $Q_k$).

B. Curve-Fitting Approximation

To make the problem tractable, the authors address the lack of a closed-form expression linking SSIM, compression ratio ($\rho$), and SNR ($\gamma$).

• Data-Driven Modeling: Using the EUROSAT dataset and a pre-trained JSCC model (from [14]), they generate data points mapping $(\rho, \gamma) \to \text{SSIM}$.
• Approximation Function: They propose a curve-fitting model (Eq. 7) to approximate the expected SSIM as a function of compression ratio and SNR:
  $$f_Q(\rho, \gamma) = \bar{a} + \sum_{j=1}^{J} (a_{0,j}e^{a_{4,j}\rho} + a_{1,j}f_{es,j}(\rho, \gamma))$$
  This transforms the implicit QoS constraint into an explicit mathematical function.

C. Optimization Algorithm (JCRRA)

The original problem (P0) is non-convex due to discrete variables and the non-convex nature of the fitted function. The authors propose the Joint Compression Ratio-Resource Allocation (JCRRA) algorithm:

1. Problem Transformation: The problem is reformulated (P1 $\to$ P2) by introducing a target SNR variable and relaxing the discrete constraints.
2. Discrete Relaxation: Binary association variables ($\alpha$) and discrete compression ratios ($\rho$) are relaxed to continuous domains.
3. Successive Convex Approximation (SCA):
   • The non-convex QoS constraint is convexified using first-order Taylor approximations (lower bounds for convex terms, upper bounds for concave terms).
   • Slack variables are introduced to handle the complex function structure.
4. Iterative Solution: An iterative algorithm solves the convexified sub-problem.
5. Recovery: The continuous solutions are mapped back to discrete values (binary association via thresholding, compression ratio via nearest neighbor selection).

3. Key Contributions

1. Novel Framework: First integration of JSCC-based semantic communication specifically for LEO satellite EO missions, focusing on power minimization under QoS constraints.
2. Data-Driven Modeling: Development of a curve-fitting model to approximate the complex relationship between image quality, compression, and channel SNR, enabling analytical optimization.
3. Algorithm Design: Proposal of the JCRRA algorithm using SCA to solve the mixed-integer non-convex problem efficiently.
4. Benchmarking: Introduction of two greedy baselines for comparison:
   • Greedy JSCC: Fixed compression ratio, greedy resource allocation.
   • Greedy JPEG: Traditional JPEG compression followed by Shannon-capacity-based resource allocation.

4. Numerical Results

Simulations were conducted using the EUROSAT dataset with realistic LEO orbital parameters (Starlink orbit, 20 GHz frequency, 40 MHz bandwidth).

• Power Savings vs. Greedy JSCC: The proposed JCRRA algorithm achieved an additional 0.7 – 1.8 dBm power reduction compared to the Greedy JSCC scheme by jointly optimizing compression and resources.
• Power Savings vs. Conventional (JPEG): The JSCC-based approach significantly outperformed the traditional JPEG method.
  • Under similar greedy strategies, JSCC saved approximately 6 dBm.
  • With full optimization (JCRRA), the total power saving over JPEG was substantial, highlighting the efficiency of semantic transmission.
• Impact of Delay: As delay requirements were relaxed (allowing more time slots), transmit power decreased for all schemes, but JCRRA consistently maintained the lowest power consumption by exploiting time-varying channel gains.
• Impact of SSIM Requirements: JSCC showed superior performance at lower SSIM requirements. While JPEG power savings were marginal when lowering SSIM, JSCC power consumption dropped sharply because the required SNR decreased significantly for lower quality thresholds.

5. Significance and Conclusion

This work demonstrates that Semantic Communication (SemCom) combined with Joint Source-Channel Coding (JSCC) is a viable and highly efficient paradigm for future LEO satellite networks.

• Energy Efficiency: By transmitting only task-relevant semantic features rather than raw bits, the system drastically reduces the energy required for high-resolution image transmission.
• Adaptability: The framework effectively handles the dynamic nature of LEO channels and varying user QoS requirements.
• Practicality: The use of curve-fitting allows the integration of deep learning-based JSCC models into traditional optimization frameworks without requiring complex end-to-end differentiable optimization of the entire physical layer.

The paper concludes that the proposed JCRRA algorithm offers a robust solution for minimizing power consumption in satellite EO missions, paving the way for more sustainable and high-capacity satellite communication systems.