FOOL: Addressing the Downlink Bottleneck in Satellite Computing with Neural Feature Compression

TL;DR

FOOL tackles the downlink bottleneck in LEO nanosatellite constellations by introducing a task-agnostic neural feature compression method that operates on shallow representations of foundational models. It combines Shallow Variational Bottleneck Injection with inter-tile dependencies and a context-aware encoding strategy, guided by a profiler to maximize throughput under intermittent links, while enabling a single encoder to support multiple backbones and predictors. The approach also includes a reconstruction pathway to map compressed features back to human-interpretable imagery, and the system is designed to operate within CubeSat-like hardware constraints with concurrent task execution. Empirical results show FOOL can boost downlinkable data by over 100x compared with bent-pipe baselines, achieve substantial bitrate reductions relative to SVBI, and provide competitive image-reconstruction quality at lower bitrates, underscoring a practical path for scalable Orbital Edge Computing in EO applications.

Abstract

Nanosatellite constellations equipped with sensors capturing large geographic regions provide unprecedented opportunities for Earth observation. As constellation sizes increase, network contention poses a downlink bottleneck. Orbital Edge Computing (OEC) leverages limited onboard compute resources to reduce transfer costs by processing the raw captures at the source. However, current solutions have limited practicability due to reliance on crude filtering methods or over-prioritizing particular downstream tasks. This work presents FOOL, an OEC-native and task-agnostic feature compression method that preserves prediction performance. FOOL partitions high-resolution satellite imagery to maximize throughput. Further, it embeds context and leverages inter-tile dependencies to lower transfer costs with negligible overhead. While FOOL is a feature compressor, it can recover images with competitive scores on quality measures at lower bitrates. We extensively evaluate transfer cost reduction by including the peculiarity of intermittently available network connections in low earth orbit. Lastly, we test the feasibility of our system for standardized nanosatellite form factors. We demonstrate that FOOL permits downlinking over 100x the data volume without relying on prior information on the downstream tasks.

Figures (19)

• Figure 1: Comparing Effect of Codec and Additive Noise
• Figure 2: Head Distillation Distortion Loss
• Figure 3: Discarding Information in Discriminative Tasks
• Figure 4: High-Level Inference Request Flow
• Figure 5: Contrasting Throughput Measures