Project Description

Abstract

NEURO-RESPOND is a Pre-Phase-A mission study investigating how neuromorphic and event-driven onboard intelligence can support faster, more selective, and more autonomous handling of mission-relevant information under spacecraft constraints. Rather than treating neuromorphic computing as an isolated technology, the project evaluates its mission and system value through three complementary application lines: Transient Sentinel, Neuromorphic FDIR, and RF Watch.

Transient Sentinel addresses low-latency detection and compact reporting of fast Earth-observation transients, with lightning detection selected as the representative demonstrator.

Neuromorphic FDIR investigates advisory anomaly indication from spacecraft telemetry as a complement to conventional FDIR.

RF Watch assesses onboard RF monitoring and classification for compact interference-awareness outputs.

Across the three applications, the study considers a common processing chain from input acquisition and preparation to neuromorphic inference, compact result generation, and selective downlink or onboard use. The assessment is driven by mission- and system-level KPIs, including response latency, reduction of unnecessary downlink, onboard autonomy, service availability, application usefulness, and resource-efficient operation. The activity aims to establish a traceable mission and system baseline, assess representative demonstrators and platform feasibility, and identify the critical technologies and development roadmap needed for future neuromorphic-enabled space missions.

Funded by the Preparation Element of the Basic Activities.

Detailed description

Context and motivation

Future space missions are expected to process increasingly large and heterogeneous data streams while operating under strict constraints in power, processing capacity, communication availability, latency, and operational robustness. In many scenarios, the limitation is not the ability to acquire data, but the dependence on a ground-in-the-loop chain for interpretation and reaction. This can delay the availability of useful information, increase the volume of raw or weakly filtered data that must be downlinked, and reduce the value of short-lived events or early anomaly indications.

NEURO-RESPOND investigates whether neuromorphic sensing and processing can move part of this interpretation chain onboard. Neuromorphic and event-driven technologies are particularly attractive for sparse, dynamic, and time-critical workloads. Still, the project deliberately evaluates them from a mission and system perspective rather than only from a component or algorithm perspective. The central question is whether they can deliver measurable mission value once data preparation, resource sharing, scheduling, compact result generation, and spacecraft-level constraints are accounted for.

Project objective and common approach

The objective of NEURO-RESPOND is to assess, at Pre-Phase-A level, the feasibility and added value of neuromorphic onboard intelligence for a set of heterogeneous space applications. The project does not aim to deliver flight-qualified or operational products during the study. Instead, it defines and exercises bounded representative demonstrators that preserve the essential input-processing-output logic of each use case and provide evidence for KPI assessment, preliminary requirement derivation, and later architecture decisions.

The three applications are treated as complementary parts of one common mission concept rather than as independent demonstrations. Their nominal processing flow follows the same high-level sequence: onboard input acquisition, data preparation, neuromorphic inference, result handling, and generation of a compact mission-relevant output. Where appropriate, the compact result can then be used onboard or prioritized for selective downlink. This shifts the mission logic from continuous raw-data transfer and delayed ground interpretation towards local interpretation first and compact reporting second.

Because onboard processing resources are constrained, the project also investigates bounded coexistence of multiple workloads on a shared neuromorphic processing chain. The applications are therefore associated with different execution modes and priorities rather than unrestricted parallel execution. This enables the study to address realistic questions of scheduling, workload switching, service availability, and resource contention.

Target application 1 - Transient Sentinel

Transient Sentinel is a payload-driven Earth-observation application. It addresses low-latency detection and compact reporting of fast-transient phenomena observed from orbit. The broader application family includes lightning activity, wildfire ignition or evolution, and volcanic eruption signatures, where timely awareness can be more valuable than continuous downlink of raw sensor data. For the current representative demonstrator, the scope is deliberately narrowed to lightning detection, providing a bounded event-driven case supported by existing space-based event-camera heritage.

The representative processing logic uses event-driven optical input, onboard preparation of the selected event data, neuromorphic inference, and compact alert generation. The intended output is a concise event product rather than a continuous raw stream, potentially containing an event indication, time, approximate location, or observation context, confidence information, and supporting metadata.

Target application 2 - Neuromorphic FDIR

Neuromorphic FDIR is the spacecraft platform-health application. It investigates advisory anomaly indication from spacecraft housekeeping telemetry and the possibility of obtaining useful onboard awareness earlier or with lower processing burden. The function is explicitly conceived as complementary to conventional, flight-qualified FDIR rather than as a replacement for certified fault-management logic.

Representative telemetry windows can be processed periodically or adaptively in the background to identify abnormal behavior and produce a compact anomaly or diagnostic indication. This application stresses different system aspects from the payload use case, including continuous access to platform telemetry, low-overhead background execution, robustness, and interaction with existing spacecraft fault-management functions.

Target application 3 - RF Watch

RF Watch is the RF-environment-awareness application. It investigates onboard processing of representative RF measurement windows to identify relevant spectral-temporal patterns and generate compact classification or interference-awareness outputs. Its execution is expected to be scheduled or task-driven based on mission context and available resources, making it a useful third workload for assessing heterogeneous operations on a shared onboard processing platform.

Mission and system assessment

A common KPI framework provides the link between application performance and mission benefit. The project assesses

1. end-to-end alert or response latency,
2. reduction of unnecessary downlink volume
3. onboard autonomy and reduced ground dependency,
4. effective service availability or duty cycle,
5. application usefulness, and
6. resource-efficient mission benefit.

These indicators are intended to show whether the representative outputs remain meaningful under realistic onboard constraints, not simply whether an algorithm performs well in isolation.

The current mission and system baseline translates these objectives into user needs, mission requirements, and preliminary system requirements that cover functional, performance, interface, data, onboard processing, reliability, and robustness aspects. This traceability provides a common foundation for architecture definition, platform trade-offs, and representative demonstrator assessment during the subsequent project activities.

Expected evolution and future developments

The next project activities build on the mission and requirements baseline by defining the neuromorphic computing and onboard-intelligence architecture, assessing critical technologies, conducting application-level benchmarking and mapping, and performing mission-level feasibility and programmatic analysis. Quantitative KPI targets and platform constraints will be refined as demonstrator evidence becomes available.

Beyond the current study, the results can support follow-on activities aimed at higher-fidelity mission demonstration, more mature integration of event-based sensing and neuromorphic processing, and validation on space-relevant computing platforms. The broader Transient Sentinel trade space may also be revisited for other time-critical phenomena, such as wildfire or volcanic-event monitoring, once suitable sensing, datasets, and mission requirements are available.

Project Outcomes

Completed/established:

1. Mission concept, target-application baseline, KPI framework, user needs, mission requirements, and preliminary system requirements for NEURO-RESPOND.
2. Representative demonstrator scope for Transient Sentinel (lightning detection), Neuromorphic FDIR (advisory telemetry anomaly indication), and RF Watch (RF monitoring and classification).

Ongoing/upcoming:

1. Neuromorphic computing and onboard-intelligence architecture definition, including shared-resource operation and platform-feasibility assessment.
2. Critical-technology assessment and technology roadmap for neuromorphic-enabled onboard processing.
3. Application-level analysis, benchmarking, and representative demonstrator evaluation against the common KPI framework.
4. Mission and spacecraft conceptual design, feasibility and risk assessment, and preliminary programmatic analysis.