NXP Semiconductors GmbH

Smarter SOS: Generative AI at the Edge for Emergency Communication over Satellite Links

Description of the company

NXP Semiconductors is a global leader in secure connectivity, embedded processing, and edge-computing solutions. The CTO System Innovation Competence Center develops forward-looking concepts for the next generation of connected systems. With a focus on AI-Enabled Future Connectivity, the center explores innovative hardware and software solutions that enable the use of artificial intelligence in safety-critical applications such as emergency communication.

NXP technologies play a key role in industrial, medical, and automotive domains, where reliable and efficient communication systems are essential. By combining hardware expertise with advanced AI capabilities, NXP drives innovation in areas ranging from connected cars and intelligent healthcare devices to resilient industrial infrastructure.

Situation

Modern emergency systems, such as the SOS service on smartphones or the BMW eCall demonstrator over satellite, allow emergency calls to be transmitted via satellite links. However, due to the severe bandwidth limitations of Non-Terrestrial Networks (NTN), only minimal data such as GPS coordinates or a standardized SOS activation message can be transmitted. Rich, semantic communication (e.g., speech, contextual details, medical information) is currently not possible.

So far, generative AI models have mainly been deployed in the cloud, which is not practical in emergency situations without stable connectivity. With modern edge hardware platforms (e.g., NXP i.MX95), the possibility arises to bring Generative AI to the Edge. In this setup, classical Speech-to-Text (STT) and Text-to-Speech (TTS) systems act as enabling technologies to convert spoken language into text and back. The real added value comes from Generative AI models, which can compress, condense, and reconstruct the semantic content of messages, enabling meaningful communication even with minimal bandwidth.

Problem

Existing SOS systems over satellite currently allow only very limited information transfer.

• Lack of contextual information: Rescue services often receive only location data, but no details about the situation (e.g., type of accident, injuries, number of people affected).
• High uncertainty: This leads to delays, misallocation of rescue resources, or inefficient deployments.
• Technological gap: While the infrastructure allows basic emergency signaling, it does not support intelligent, semantically condensed communication over narrowband NTN links.
• Lack of Edge optimization for AI: Current AI models are often too large or computationally intensive for efficient execution on edge devices, highlighting a need for adaptation and optimization.

This creates a critical gap between technical feasibility (satellite connectivity, cloud AI) and the practical need for reliable, real-time AI at the edge in emergency scenarios.

Aims of the project

The project aims to design and prototype an AI-assisted emergency communication system that enables meaningful, reliable, and context-rich communication over satellite links, even with minimal data rates.

Sub-goals:

1. Integration of Speech-to-Text (STT) and Text-to-Speech (TTS) pipelines as enabling interfaces for transforming spoken language into text and vice versa.
2. Combination with Generative AI at the Edge, responsible for semantic compression, condensation, and reconstruction of information.
3. Exploration of methods to adapt and optimize generative language models for efficient execution on edge hardware (e.g., quantization, pruning, model compression).
4. Integration on the NXP i.MX95 platform and coupling with an NTN simulation using an open-source 5G stack.
5. Demonstration of an end-to-end emergency communication prototype built by the student team.
6. Improvement of the reliability of emergency calls through semantically meaningful information transfer, supporting more efficient and targeted rescue operations.

Scopes

Emergency situations can occur in very different environments – often in places without terrestrial mobile network coverage. Relevant scenarios include:

• Bicycle or e-bike accidents: often on remote trails with no cellular coverage.
• Hiking and mountain sports: accidents in forests or mountainous regions where mobile networks are unavailable.
• Accidents at sea: sailors, fishermen, or small boats in distress far from terrestrial networks.
• Traffic accidents in remote regions: car crashes on rural roads, offroad tours, or in sparsely populated areas.

Industrial operations: accidents on offshore platforms, in mines, or at remote construction sites.

Emergency situations can occur in very different environments – often in places without terrestrial mobile network coverage. Relevant scenarios include:

• Bicycle or e-bike accidents: often on remote trails with no cellular coverage.
• Hiking and mountain sports: accidents in forests or mountainous regions where mobile networks are unavailable.
• Accidents at sea: sailors, fishermen, or small boats in distress far from terrestrial networks.
• Traffic accidents in remote regions: car crashes on rural roads, offroad tours, or in sparsely populated areas.
• Industrial operations: accidents on offshore platforms, in mines, or at remote construction sites.

In such cases, Generative AI at the Edge can be critical, ensuring that precise and semantically relevant information reaches emergency centers despite extremely low bandwidth. The goal is to enable higher survival chances, faster response times, and more efficient use of rescue resources.

The project focuses on the conceptual design and prototypical implementation of an AI-assisted emergency communication system over satellite links with Generative AI at the Edge.

In scope (included):

• Development of a demonstrator on the NXP i.MX95 platform.
• Integration of Speech-to-Text and Text-to-Speech as enabling technologies.
• Adaptation and optimization of Generative AI models for edge computing.
• Simulation of NTN links using an open-source 5G stack.
• Scenario-based testing (e.g., bicycle accident, hiking accident, sea accident).
• Documentation and presentation of results.

Out of scope (not included):

• Development of new satellite hardware or end devices beyond the demonstrator.
• Establishing a real satellite connection (simulation only).
• Medical assessments or integration with real emergency centers.
• Full-scale development of new large language models (only adaptation and optimization of existing ones).

Target group (students)

The project addresses students from the following study programs:

• Computer Science, Technical Computer Science
• Computer Engineering
• Information and Electrical Engineering
• Computer Science Engineering
• Data Science
• Communications and Network Engineering

Desired skills:

• Programming (e.g., Python, C/C++)
• Foundations in Machine Learning / AI
• Interest or first experience with Generative AI and language models
• Experience in Embedded Systems / Edge Computing is beneficial
• Basic knowledge in communication systems (5G, NTN) is helpful but not required

The project is designed to be interdisciplinary and suitable for students with either a software or hardware focus.

Dates
Please save these dates: Fishing for Experience Termine

Registration
You can apply for Fishing for Experience online.