Typhoon Edge: Bringing Real-Time Intelligence Closer to the Sensor

Traditional Earth Observation missions have largely depended on a collect–store–downlink–process model. Typhoon Edge is designed to move part of that intelligence directly into orbit by combining the deterministic processing capabilities of a Microchip PolarFire FPGA with the high-performance AI and GPU capabilities of the NVIDIA Jetson AGX Orin.

From Data Collection to Actionable Intelligence

In an increasingly connected world, the value of Earth Observation is no longer measured only by how much data a satellite can collect. The real value lies in how quickly that data can be transformed into actionable intelligence.

For applications such as environmental monitoring, disaster response and public safety, waiting for raw imagery to be downlinked, processed on the ground and analysed may introduce delays at precisely the moment when rapid decisions matter most.

Typhoon Edge, CAVU Aerospace UK’s high-performance edge computing OBC architecture, is designed to address this challenge. By combining the flexibility and deterministic processing of a Microchip PolarFire FPGA with the powerful GPU-based AI capabilities of an NVIDIA Jetson AGX Orin, Typhoon Edge creates a heterogeneous computing platform capable of bringing advanced processing directly into the spacecraft.

Microchip’s PolarFire architecture is particularly well suited to power- and thermally constrained embedded systems, offering deterministic processing, flexible FPGA logic and high-speed connectivity.

A New Era for Intelligent EO

Conventional Earth Observation architectures typically follow a familiar sequence:

Capture → Store → Downlink → Process → Analyse → Decide

This approach remains essential for many missions, but it can be inefficient when the spacecraft collects significantly more data than can be transmitted immediately, or when a rapid response is required.

Typhoon Edge introduces the possibility of moving intelligence closer to the point of data acquisition:

Capture → Process On-Board → Detect → Classify → Prioritise → Downlink Intelligence

Instead of treating the satellite simply as a remote sensor, the spacecraft can become an intelligent edge node operating directly in orbit.

The NVIDIA GPU can execute AI and machine-learning workloads, while the PolarFire FPGA can manage deterministic interfaces, high-speed sensor connectivity, pre-processing, data movement and mission-specific hardware acceleration. The result is a complementary architecture in which each processor can focus on the tasks for which it is best suited.

The Power of a Heterogeneous Architecture

Typhoon Edge is based on the principle that no single processing architecture is ideal for every task in a modern EO mission.

The PolarFire FPGA can provide a highly programmable and deterministic layer between multiple sensors and the AI computing subsystem. It can be used for functions such as:

• High-speed sensor interfacing
• Image and data acquisition
• Data formatting and pre-processing
• Real-time signal processing
• Sensor synchronisation
• Multi-sensor data routing
• Hardware-accelerated algorithms
• Mission-specific interface implementation

At the same time, the NVIDIA Jetson AGX Orin provides a powerful environment for GPU-accelerated computing and AI inference.

Together, these devices can create an architecture in which sensor data is acquired, prepared and delivered efficiently to AI algorithms, allowing the spacecraft to perform increasingly sophisticated analysis without waiting for ground processing.

This combination of deterministic FPGA processing and AI acceleration is particularly relevant to space systems, where power, thermal capacity and communications bandwidth are all limited resources. PolarFire devices are specifically designed around low-power operation, with Microchip highlighting their thermal efficiency and deterministic computing capabilities.

Multi-Sensor Intelligence from Orbit

Modern Earth Observation increasingly relies on more than a single optical camera.

A mission may combine visible imaging, infrared imaging, thermal sensors, hyperspectral or multispectral instruments, RF sensors, radar-derived information, telemetry & GNSS and positional data to create live tracking EO intelligence. Typhoon Edge’s architecture can support the concept of a multi-sensor intelligence platform, where information from different sources can be collected, aligned and processed as part of a unified on-board workflow.

The PolarFire FPGA can act as the flexible and deterministic interface layer connecting sensors and data sources, while the Jetson AGX Orin can perform higher-level AI processing and sensor-fusion algorithms. This creates opportunities for the spacecraft to move beyond simple image capture.

For example, an EO satellite could potentially combine visible imagery with thermal information to distinguish between different environmental conditions. Multiple observations collected over time could be analysed to identify changes, moving objects or developing events. Rather than transmitting every frame with equal priority, the system can support intelligent decisions about what information is most important to transmit first.

Enabling Live Tracking from Space

One of the most exciting possibilities enabled by reliable on-board GPU processing is near-real-time detection and tracking. A conventional imaging satellite may capture an image of an area and send that data to the ground for later analysis. With AI processing available directly on-board, a different approach becomes possible. The process chain operates as follows:

Sensor Acquisition
↓
FPGA-Based Data Handling and Pre-processing
↓
GPU-Based AI Inference
↓
Object Detection or Event Identification
↓
Tracking Across Sequential Observations
↓
Generation of Metadata, Alerts or Priority Data

Depending on the mission architecture, sensor characteristics, orbital geometry, revisit opportunities and communications availability, this approach could support near-real-time tracking and event monitoring directly from orbit.

Instead of downlinking only large volumes of raw imagery, the satellite could potentially generate valuable information such as object location, object classification, direction of movement, speed estimation, event detection, change detection, confidence levels & time and geographic metadata.

The resulting metadata package can be dramatically smaller than the original image data, potentially allowing critical information to reach operators more quickly. The original imagery can still be stored and transmitted when required for validation or more detailed analysis.

Environmental Monitoring: Turning Observations into Early Insight

Environmental changes are often gradual, widespread and difficult to monitor continuously from the ground.

An intelligent EO platform equipped with Typhoon Edge could support automated monitoring of wildfires and smoke development, flood expansion, coastal changes, deforestation, ice and glacier movement, oil spills and marine pollution, agricultural conditions, water quality indicators & land-use changes. AI algorithms operating on-board could search for specific patterns or changes between observations.

For example, instead of requiring a ground team to review every image, an AI model could identify regions that exhibit characteristics associated with a developing wildfire or flood. The spacecraft could then assign a higher priority to the relevant data. This does not eliminate the role of human analysts. Instead, it allows the satellite to act as an intelligent first layer of analysis, helping ground teams focus their attention on the information that matters most.

Faster Disaster Response

During natural disasters, time is one of the most critical resources. Floods, wildfires, earthquakes, volcanic events and severe storms can evolve rapidly. Emergency responders need timely situational awareness to understand what is happening, where the greatest impact is occurring and how conditions are changing. Typhoon Edge can support an architecture in which AI algorithms process sensor data directly on the spacecraft and identify potentially significant events.

Wildfire detection is a great example. AI can analyse imagery for smoke, thermal anomalies or rapidly expanding fire boundaries. Once a possible event is detected, the satellite can prioritise the relevant imagery and generate event metadata.