The satellite industry is experiencing a revolution, with thousands of spacecraft operating simultaneously in low Earth orbit (LEO). This scale brings unprecedented challenges, particularly in managing the vast and complex telemetry data generated by these fleets. The concept of the 'cardinality wall' emerges as a critical issue, representing the point at which traditional telemetry infrastructure fails to support the growing volume and complexity of data.
Telemetry, once manageable with a few hundred parameters, now involves tens of thousands of signals streaming at sub-second intervals. This includes battery voltages, thermal sensor readings, reaction wheel speeds, and more. With hundreds of spacecraft, operators face the challenge of ingesting millions of measurements per second, each with intricate structure and metadata. This high cardinality, or unique data streams, poses a significant problem for ground system databases designed for smaller missions.
The issue lies in the limitations of traditional ground system databases. These systems, often built on relational databases or log analytics platforms, struggle with high-cardinality telemetry. They rely heavily on indexing, which becomes inefficient as the number of identifiers grows, leading to performance degradation. Additionally, they are ill-suited for the continuous telemetry streams and time-based queries that are now commonplace in aerospace systems.
The problem intensifies when operators attempt long-term telemetry retention. Satellite programs store data for years or decades, requiring systems to support both real-time ingestion and historical analysis. This dual requirement challenges general-purpose databases, often forcing operators to simplify data to maintain operational stability. For instance, Loft Orbital, a microsatellite operator, faced this challenge, eventually transitioning to a time series-oriented architecture to handle high-frequency telemetry and maintain context across missions.
The consequences of losing context are severe. Engineers rely on correlating events across subsystems to detect anomalies. Removing this context makes anomaly detection more difficult, and it severely impacts machine learning systems designed to predict component failures. As satellite constellations grow in size and autonomy, telemetry infrastructure becomes integral to mission success, influencing operational visibility and resilience.
Addressing the cardinality wall requires a strategic approach. Teams should identify specific areas where cardinality impacts operations, such as delayed anomaly detection or data gaps. Decoupling the telemetry pipeline into high-throughput ingestion and analytical workloads can help stabilize real-time monitoring. Additionally, preserving full context should be a priority, as downsampling or stripping metadata can introduce blind spots during anomaly investigations and system validation.
The solution lies in treating telemetry systems as distributed infrastructure rather than centralized databases. Data arrives out of order and in bursts, requiring systems that can adapt to this reality. By isolating and redesigning strained architecture components, teams can make progress. Incremental tuning is no longer sufficient; the next generation of LEO infrastructure demands telemetry architectures designed for scale, distribution, and context preservation from the outset.
In conclusion, the cardinality wall poses a significant challenge to the satellite industry. By recognizing the limitations of traditional approaches and embracing distributed telemetry systems, operators can ensure the reliability and resilience of their LEO missions. This transformation is essential as the industry continues to scale and evolve, shaping the future of space operations.
