Understanding AI to Maintain Control

Artificial intelligence opens new possibilities in signal processing and communications optimization. But faced with the risk of “black box” behavior, understanding how a system reaches its decisions is essential to maintaining control over them. In aerospace — where criticality is high and embedded systems operate autonomously — AI must remain explainable, verifiable, and trustworthy. 

What Is Trustworthy AI in Aerospace?

Trustworthy AI, applied to aerospace, must meet unique requirements in terms of safety, certification, and robustness under extreme conditions.

Reliability and consistency:
An embedded AI must deliver stable, predictable results — even under extreme conditions such as jamming, noise, or signal loss. In a radar or communications system, reliability is measured by performance stability: packet delivery rate (PDR), target detection accuracy, trajectory estimation.

Security and confidentiality:
Aerospace communications handle sensitive data. Trustworthy AI must guarantee the protection of that information and resist cyberattacks, while complying with applicable safety standards.

Criticality and certification:
In an environment where the slightest failure can compromise a mission, AI must be certifiable. This requires algorithms that are explainable, traceable, and auditable — compliant with aeronautical and space standards such as DO-178C and ECSS.

Adaptability and robustness:
Radio environments are inherently dynamic — subject to interference, mobility, and jamming. Reliable AI must adapt in real time and remain robust in the face of the unexpected.

How to Build Trustworthy AI: Deep Learning for Communications and Radar Systems

In the communications and radar domain, building trustworthy AI means mastering every stage of the lifecycle: data, training, usage scoping, traceability, explainability, and certification.

Data quality and diversity:
Representative datasets covering a wide range of propagation, jamming, and radar detection scenarios.

Rigorous annotation and validation:
Ensuring label accuracy — signals, targets, interferences — to produce reliable training data.

Bias correction:
Identifying and correcting biases to ensure consistent performance in real-world conditions.

Usage scoping:
A good local AI is not one that can do everything — it is one that does one specific task well.

• Example: a communications AI can be certified to optimize spectral allocation within a defined frequency range and under specified load scenarios, but not for all unknown situations.
• Example: a radar deep learning network can be validated for target detection within a specific range of distances and angles, but cannot be extrapolated to untested edge cases.

Traceability and explainability:
Ensuring that every decision made by the neural network can be traced and justified. This involves inference logging, analysis of internal weights and activations, and the use of Explainable AI (XAI) methods such as saliency maps, feature attribution, and sensitivity analysis.

Continuous evaluation and certification:
Testing models in simulation and under operational conditions to ensure their validity, stability, and compliance with applicable standards.

Go further: To learn more about the four pillars of our solution, visit the ARKANE Technology page.

A Hybrid Architecture for On-Board Decision-Making

We design hybrid architectures that combine traditional signal processing with neural networks optimized for embedded deployment. The goal is straightforward: enabling our systems to process and interpret complex signals directly at the source.

In practice, this means:

• Real-time operation, even in resource-constrained environments
• Significantly reduced processing latency
• Low power consumption and more efficient resource management
• Dynamic adaptability to changing signal or environmental conditions

Concrete Use Cases

In radar systems:
Our technology detects, tracks, and classifies targets in jammed or cluttered environments. Embedded AI improves resolution, strengthens tracking reliability, and automates the analysis of RF signatures.

In GNSS systems:
By combining deep learning with adaptive beamforming, our solutions deliver robust geolocation under jamming conditions — filtering out interfering signals and reinforcing useful signal paths, all in real time.

In critical telecommunications:
Our multi-antenna processing improves link stability in dense or mobile networks, while optimizing dynamic radio resource allocation.

Toward Truly Autonomous Systems

Embedded deep learning fundamentally transforms the way a system operates. By integrating artificial intelligence directly into a radar or GNSS device, we give it the ability to:

Go Further

To learn more about the four pillars of our solution, visit the ARKANE Technology page.

What Is the Hiroshima AI Process for AI?

Launched by the G7 in 2023, the Hiroshima AI Process resulted in an International Code of Conduct for Organizations Developing Advanced AI Systems, built around eleven actions covering risk management across the AI lifecycle, transparency, security, content authentication, governance, and data protection.

In February 2025, the OECD translated these principles into a Reporting Framework — a voluntary, standardized tool enabling organizations to publicly report on how they implement these principles. It is neither a label nor a certification, but a transparency initiative designed to make AI practices comparable across organizations.

For ARKANE, Trustworthy AI Starts with Engineering

ARKANE designs frugal embedded AI for signal processing, antenna systems, and wireless protocols, primarily serving customers in the aerospace and defense sectors.

In these industries, trust cannot simply be declared — it has to be proven. Robustness under real-world operating conditions, risk management across the full product lifecycle, model security, and explainability are engineering requirements, aligned with standards such as DO-178C and the EASA AI Roadmap. Here, explainable AI (XAI) is used as a reliability tool rather than a retrospective justification.

For ARKANE, joining the HAIP reporting initiative means formalizing, at an international standard, a practice that is already central to how the company operates.

A Natural Extension of Our Commitments

This commitment builds naturally on ARKANE’s existing positioning:

• “AI Pioneers” award winner (France 2030 / P2IA) — recognition of ARKANE’s innovation capacity in AI applied to demanding use cases.
• Bpifrance Deep Tech label — a mark of strong technological and scientific depth.
• AI for critical infrastructure resilience — a priority identified by the tech industry in the “Cybersecurity and Critical Infrastructure Resilience” section of the TECH7 recommendations to the G7.

ARKANE also took part in the TECH7 Summit (28 May 2026), the forum bringing together the leading digital-industry associations from the G7 and the EU, which issues annual recommendations to G7 leaders.

Our Practices, in Concrete Terms

Our approach to trustworthy AI translates into concrete, day-to-day practices:

• Lifecycle risk governance: identifying, assessing, and mitigating risks from design through embedded deployment.
• Robustness and reliability: validation under representative operational conditions, consistent with our sectors’ certification requirements.
• Explainability (XAI) used as an engineering lever to strengthen hybrid AI / signal-processing pipelines.
• Frugality: optimized models (compression, quantization) designed to run under tight embedded resource constraints (Size, Weight, and Power — SWaP).
• Data security and governance: internal data-protection and security rules tailored to the sensitivity of our projects.

What Comes Next

This declaration is the first step. The next will be completing our HAIP report, to be published on the OECD platform, mapping our practices against the Code of Conduct’s eleven actions and turning our commitment made in Bercy into demonstrated proof.

Further reading:

• Press release from the French Directorate General for Enterprises (DGE)
• OECD HAIP Reporting Framework

A strategic issue: securing PNT signals

Positioning, navigation and synchronization (PNT) signals are now essential for the operation of many critical infrastructures: transport, energy, telecommunications, autonomous systems.

However, these signals are exposed to increasing threats:

• Jamming
• Spoofing
• Desynchronization

In an increasingly tense geopolitical and cyber context, the ability to guarantee the integrity, authenticity and continuity of service of PNT signals becomes a challenge for European sovereignty.

The P2IA project: an embedded AI at the service of cyber-resilience

To meet this challenge, ARKANE is developing P2IA, an embedded artificial intelligence solution capable of:

• Detect attacks in real time
• Identify their nature
• Neutralize their effects
• Maintain service continuity

Our approach is based on frugal and explainable AI, designed to be directly integrated into embedded systems, with strong constraints on performance, latency, and energy consumption.

A differentiating technological approach

1. AI hybridization / signal processing
ARKANE combines the robustness of classical signal processing methods with the adaptive power of artificial intelligence.

This hybridization allows:

• Better resilience against unknown attacks
• A dynamic adaptability to contested environments
• Increased explainability of algorithmic decisions

2. Generative AI attack scenario generation
We integrate generative AI mechanisms to automatically produce complex and realistic attack scenarios.

Objective:
Continuously train, test and reinforce our defense systems against evolving and increasingly sophisticated threats.

3. A sovereign and certifiable firmware architecture

From phase 1 of the project, P2IA’s architecture is designed to meet European regulatory requirements:

• Compliance with the AI Act (high-risk AI systems—critical infrastructures)
• Alignment with NIS2 and REC frames
• Taking into account the recommendations of ANSSI

This “compliance by design” approach guarantees a certification path compatible with the requirements of European institutional and industrial actors.

A strategic recognition

The selection of ARKANE among the first winners of ‘Pioneers of AI’ is a strong recognition:

• From our vision of a sovereign AI
• Of our commitment to an explainable AI
• From our positioning on the challenges of cyber-resilience of critical infrastructures

We thank our institutional partners as well as all of our teams, whose expertise in signal processing, embedded AI and secure architectures makes this ambition possible!

In 2025, ARKANE devoted its efforts to validating and strengthening its value proposition in artificial intelligence applied to connected and autonomous systems for the aeronautics, space, and defense sectors.

This trajectory is based on close collaboration between ARKANE’s teams, its partners and its ecosystem. ARKANE particularly thanks the CNES, the European Commission’s JRC, IMT Atlantique and the Brittany Region for their continuous support.

The entire ARKANE team wishes you an excellent 2026!

Arkane had the opportunity to participate in one of the highlights of the French space ecosystem: the CNES’s Rencontres Techniques et Numériques, dedicated to Machine Learning applications for signal processing. This event brought together experts, industry and researchers around the technological issues related to navigation systems, telecommunications and next-generation embedded architectures.

CRPA antenna and robust navigation: Arkane’s feedback

Embedded Machine Learning on FPGA: a decisive performance lever

One of the key points of the presentation focused on the integration of machine learning models embedded on FPGA. This approach takes advantage of both:

• the flexibility and adaptability power of AI models,
• real-time performance 

By jointly optimizing algorithmic processing and their hardware implementation, Arkane develops solutions capable of offering better responsiveness to disturbances, while respecting the strict constraints of critical embedded systems.

This mastery of embedded machine learning on FPGA is now a differentiating advantage for the operational deployment of AI solutions in space and defense.

Arkane participated, alongside CNES, in JammerTest 2025, a unique campaign dedicated to the evaluation of GNSS solutions against intentional interference.

Organized in Andøya, Norway, these tests allowed our team to test its antenna electronics and CRPA anti-jamming treatments under different interference scenarios, in order to demonstrate their robustness and effectiveness in real conditions. 

These campaigns are also a valuable source of real acquisition data, used for training and continuous improvement of embedded AI models, thus enhancing the performance of adaptive treatments in disturbed environments. 

—–

Real-life tests for Arkane at the Jammer Test 2025 in Andøya

GNSS technologies, such as GPS or Galileo, are essential for location and timing applications, but are particularly sensitive to jamming and interference. In order to ensure reliable navigation and operation without the risk of signal loss, Arkane develops space-time algorithms that nullify these sources of interference in order to recover robust GNSS signals. By integrating Deep Learning-type adaptive AI solvers, Arkane strengthens the ability of its processes to self-adjust in the face of increasingly complex interference, thus ensuring stable and secure navigation.

In the middle of a test campaign, we are validating and evaluating our anti-jamming and anti-luring solution. We thus push our solution to its limits, evaluating its performance against various levels of JSR and multiple sources of interference. These full-scale tests allow us to test the robustness and effectiveness of our technology in real conditions.

Accompanied by our historical partner CNES, Joan Gomes and Stéphane Oriol, we meet up with our partners from the JRC ISPRA European Commission and the European Space Agency – ESA to face fog and jammers!  

Focus on the anti-jamming solutions developed by Arkane

ARKANE solutions respond to new interference threats by optimizing adaptive space-time processing through the integration of artificial intelligence software components. Lighter in embedded resources and more efficient, they provide enhanced protection of GNSS signals and increased resilience for critical systems subject to strong operational constraints. 

Designed to operate in dynamic environments with high acceleration, these solutions are now applied to autonomous navigation and autonomous launcher backup, where the continuity of GNSS guidance is essential, while responding to the increasingly present and sophisticated interference threats in the space, drone, and maritime domains.

Balance sheet and results of the JammerTest 2025?

This campaign allowed an initial evaluation of the performance of our existing anti-jamming solutions in real conditions, through a wide variety of scenarios: different waveforms (spoofing, meaconing, broadband jamming and complex mixtures), low to high power jammers, multi-transmitter configurations to test the performance of CRPA systems, as well as dynamic tests with onboard drone jammers.

It also constitutes a valuable database for the continuous improvement of our treatments, the calibration of our simulators and the training of our machine learning models.

Finally, it was an opportunity to strengthen our network in the field of anti-interference and GNSS acquisition, and to prepare for future test campaigns

The JRC of the European Commission: a reference centre for scientific innovation

Arkane in GNSS testing campaign at the JRC: a key step towards validation under real conditions

Our team took advantage of the summer to launch a first laboratory test campaign at the JRC, as part of the development of its anti-jamming GNSS solution adapted to the navigation of future space launchers.

Objectives of the current phase:

• Conducting tests in the pipeline via a Spirent bench,
• Acquisition of IF/RF signals using our datagrabber, then post-processing on FPGA via our proprietary algorithms,
• Validation of the acquisition chain for the jammertests scheduled in September,
• In parallel, characterization of our CRPA antenna (Controlled Reception Pattern Antenna) to prepare for tests in a realistic environment.

Next steps:

• Real-time transition of the entire chain,
• Radiated tests in an anechoic chamber, with real antennas, to simulate representative operational scenarios.

We warmly thank Luca Cucchi and the JRC teams for their welcome and technical expertise, as well as Joan Gomes and Stéphane Oriol from CNES for their support during this crucial phase. Thank you also to the National Frequency Agency (ANFR) and to Jérôme André for facilitating this European collaboration.

Arkane : des solutions d’intelligence artificielle pour les communications aérospatiales

This test campaign is a continuation of Arkane’s work in the field of intelligent signal processing for aeronautics, space and defense.

Among our solutions in development:

• Adaptive beamforming algorithms, capable of separating signal sources in noisy or interfered environments,
• Onboard multi-antenna processing chains, optimized for constrained systems (FPGA)
• GNSS interference detection tools for critical applications (satellites, drones, autonomous vehicles),
• Solutions to reduce interference and improve the robustness of communications in complex electromagnetic environments.

Thanks to our expertise in AI applied to the radio frequency signal, Arkane offers a new generation of embedded processing designed to enhance the resilience of navigation and communication systems.

Would you like to know more about our GNSS technologies and solutions for critical systems?

Contact us to discuss your specific needs!

Juillet 2025