LEAT is developing a miniature terminal that uses artificial intelligence for space IoT communication
- Research
- Innovation
Published on January 16, 2023
–
Updated on January 24, 2023
Dates
on the January 16, 2023
Location
Campus SophiaTech
Fabien Ferrero, Benoît Miramond and Laurent Rodriguez, professors at Sophia Antipolis' Laboratory for Electronics, Antennae, and Telecommunications (LEAT) and winners of the call for hardware projects issued in 2021 by the "Networks, Information and Digital Society" Academy (RISE) have acquired tailor-made hardware to develop their research aimed at embedding frugal AI in our connected objects.
In the next few years, our homes, cities, vehicles and industries will be filled with billions of connected devices. This concept is known as the Internet of Things (IoT). Many of these devices will use automatic machine learning models to extract relevant information from sensor data, derive accurate predictions from the data and make decisions.
The downside is that the massive data processing involving current artificial intelligence (AI) methods is power hungry. Another disadvantage is that data is processed in the cloud by machines often located at a great distance from the connected objects, which saturates network communications.
To address these issues, LEAT is pursuing a promising avenue of research that involves installing machine learning on the connected devices themselves, using low-power microcontrollers (miniature embedded computers) to keep computations close to the data source (edge computing) and reduce the energy and network footprint of the IoT.
However, to achieve the goal of embedding artificial intelligence directly into end-user IoT devices, problems of real-time computing, power consumption and memory footprint on these resource-constrained devices first need be solved. The SITH-Smart Internet of Things project was set up to deal with these issues by testing scenarios for implementing operational solutions using low-power electronic boards that include artificial intelligence algorithms and IoT communications.
The SITH project's electronic boards have demonstrated their capacity to communicate with them at 868 MHz using LR-FHSS (Long Range Frequency Hopping Spread Spectrum) technology.
Radio waves can travel over very long distances on this frequency band and use much less power than products that use 2.4 GHz bands, such as Wi-Fi or Bluetooth, and are not obstructed by human bodies or concrete walls.
Contacts were successfully established with three different satellites during the experimentation period.
The electronic boards developed as part of the SITH project serve as a new experimental platform for an innovative software solution based on MicroAI, which is currently being developed by Pierre-Emanuel Novac as part of his PhD research (under a LEAT and ELLCIE-HEALTHY CIFRE grant) supervised by Pr. Alain Pegatoquet and Pr Benoît Miramond, chair holder at the 3IA Côte d'Azur Institute.
MicroAI is used to deploy artificial neural networks on embedded platforms (such as the boards of the SITH project). This open-source software also converts and compresses deep neural networks to make them suitable for hardware that is highly limited in terms of memory, computing power and energy.
Students in the EIT-Digital Master's program (Autonomous Systems specialty), and 4th and 5th year students in computer science and electronics at Polytech Nice Sophia are given the opportunity to create, simulate and deploy their own deep neural networks, and at the same time discover the world of research by carrying out projects and practical work based on the immediate and concrete results of research conducted by the LEAT laboratory.
With the increasing number of edge AI applications, they are acquiring skills that are highly valued in the professional world, where the major companies are offering more and more solutions for deploying neural networks on embedded platforms (Google tensorflow Lite, STM Cube AI, etc.).
These projects could overcome technological barriers in different areas including antenna design, information routing algorithms, and the design and implementation of new models of distributed neural networks.
With this in sight, LEAT is now planning to carry out tests on a representative sensor network and then create an edge computing demonstrator using a single solar cell as an energy source.
The scientific results should allow us to understand the potential and the limits of LEAT's efforts for dealing from the growing energy consumption of digital systems. This research may interest various communities working on embedded artificial intelligence, sensors, telecommunications and wireless networks, to name just a few.
The downside is that the massive data processing involving current artificial intelligence (AI) methods is power hungry. Another disadvantage is that data is processed in the cloud by machines often located at a great distance from the connected objects, which saturates network communications.
To address these issues, LEAT is pursuing a promising avenue of research that involves installing machine learning on the connected devices themselves, using low-power microcontrollers (miniature embedded computers) to keep computations close to the data source (edge computing) and reduce the energy and network footprint of the IoT.
However, to achieve the goal of embedding artificial intelligence directly into end-user IoT devices, problems of real-time computing, power consumption and memory footprint on these resource-constrained devices first need be solved. The SITH-Smart Internet of Things project was set up to deal with these issues by testing scenarios for implementing operational solutions using low-power electronic boards that include artificial intelligence algorithms and IoT communications.
|
The SITH project meets three objectives of the Networks, Information and Digital Society Academy of Excellence of Université Côte d'Azur:
|
SITH project has already enabled several technological assessments
Long-distance communication with Lacuna Space satellites
To power connected objects in areas of the world where cellular and Wi-Fi communication coverage is lacking, Lacuna Space has deployed, in a 500 km orbit, shoebox-sized satellites that can transmit small amounts of data over a long distance.The SITH project's electronic boards have demonstrated their capacity to communicate with them at 868 MHz using LR-FHSS (Long Range Frequency Hopping Spread Spectrum) technology.
Radio waves can travel over very long distances on this frequency band and use much less power than products that use 2.4 GHz bands, such as Wi-Fi or Bluetooth, and are not obstructed by human bodies or concrete walls.
Contacts were successfully established with three different satellites during the experimentation period.
Map of the positions of the Lacuna Space satellites when they picked up information sent by the SITH project board, showing the path of the LS2B satellite (on the left with a polar orbit) and the LS2C satellite (on the right with an equatorial orbit) over one day. (@Université Côte d'Azur, LEAT).
Deployment of artificial intelligence code using MicroAI technology
The electronic boards developed as part of the SITH project serve as a new experimental platform for an innovative software solution based on MicroAI, which is currently being developed by Pierre-Emanuel Novac as part of his PhD research (under a LEAT and ELLCIE-HEALTHY CIFRE grant) supervised by Pr. Alain Pegatoquet and Pr Benoît Miramond, chair holder at the 3IA Côte d'Azur Institute.MicroAI is used to deploy artificial neural networks on embedded platforms (such as the boards of the SITH project). This open-source software also converts and compresses deep neural networks to make them suitable for hardware that is highly limited in terms of memory, computing power and energy.
SITH boards used as a teaching material to introduce graduate students to embedded AI
Université Côte d'Azur students use the electronic boards financed by the SITH project and the MicroAI solution as physical platforms and software bricks to experiment with embedded AI technologies.Students in the EIT-Digital Master's program (Autonomous Systems specialty), and 4th and 5th year students in computer science and electronics at Polytech Nice Sophia are given the opportunity to create, simulate and deploy their own deep neural networks, and at the same time discover the world of research by carrying out projects and practical work based on the immediate and concrete results of research conducted by the LEAT laboratory.
With the increasing number of edge AI applications, they are acquiring skills that are highly valued in the professional world, where the major companies are offering more and more solutions for deploying neural networks on embedded platforms (Google tensorflow Lite, STM Cube AI, etc.).
Students in the EIT-Digital Master’s program developing an artificial intelligence model for human activity recognition on SITH project electronic boards as part of an embedded AI course. (@Université Côte d'Azur, LEAT).
What's next?
These advances are sparking new research projects around embedded AI and IoT at LEAT. Researchers are seeking to create new models of neural networks distributed over a large set of boards that should be capable of exploiting the information extracted by the various sensors when the boards are deployed at different distances from each other.These projects could overcome technological barriers in different areas including antenna design, information routing algorithms, and the design and implementation of new models of distributed neural networks.
With this in sight, LEAT is now planning to carry out tests on a representative sensor network and then create an edge computing demonstrator using a single solar cell as an energy source.
The scientific results should allow us to understand the potential and the limits of LEAT's efforts for dealing from the growing energy consumption of digital systems. This research may interest various communities working on embedded artificial intelligence, sensors, telecommunications and wireless networks, to name just a few.
- Learn more:
-
- MicroAI software solution showcased by the 3IA Côte d’Azur institute, , http://3ia-demos.inria.fr/
- Research by the EDGE team at LEAT, https://leat.univ-cotedazur.fr/edge/
- https://lacuna.space/
- €8,900 of hardware acquired thanks to "Hardware 2021" Call for Projects launched by RISE Academy
-
- Communicating nodes with Edge AI capabilities.
- Sensors integrated to provide a large variety of data (9 axis IMU, microphone, Altimeter, gaz and luminosity sensor).
- A LORA sub-GHz wireless transceiver
- A GPS positioning module
- An advanced Micro Controller Unit with a large flash memory is in charge to control every component and to compute machine learning algorithm.
- A low consumption battery
- The low power platform with a battery support holder is set in an IP65 casing for outdoor experiments.
- A custom design with DS4H and Université Côte d'Azur logo has been plotted on the top PCB substrate.
