The overall goal of the project AIComS is to develop AI/ML-based SW/HW platforms for future products of an integrated satellite and 5G&beyond communication network. It focusses on the development of ML-based 5G NR PHY-Layer components and 5G&beyond compliant ML-based routing, network slicing, and security components.
The following developments are targeted by the project activities:
5G NTN Baseband Processing Platform: Development of data driven baseband technologies for 5G-NTN RAN with different functional splits.
5G AI Satellite Packet Router: Development of data-driven network and service technologies.
5G AI Satellite Packet Router: Development of data-driven IT security concepts.
AI based Fault Detection, Identification and Recovery (AI-FDIR) Software: Development of data driven prediction methods.
AI based Formation Control (AI-FC) Algorithms: Development of data driven formation control algorithms.
AI based VLEO Orbit Control (AI-VLEO-OC): Communication aware development of data driven antenna beam pointing, navigation and AOCS algorithms as well as drag compensation.
It will be an outcome of the project to gain knowledge about the appropriate degree of autonomy by means of AI or how much control is still required.
AIComS aims to replace so far model-based designed functionalities with their data-driven AI-counterparts. The key challenge is on developing and implementing AI/ML concepts for HW/SW platforms for LEO satellite connectivity leading to an improved cost-benefit ratio. In this context the effect of (un-)explainable AI on the implementation of communication network functionalities and safety critical functions is challenging.
AIComS will provide partial integration of AI/ML-based components for satellite-based communication systems following the 3GPP work on Non-Terrestrial-Networks (NTN). The baseband unit will enable 5G NTN baseband processing with functional blocks using AI/ML- algorithms on-board the satellite. In contrast to classical approaches, ML techniques do not rely on model knowledge such that ML is a promising tool for applications where the underlying model is unknown or very complex. Furthermore, ML can flexibly approximate classical approaches with good performance and low complexity. In addition, the routing hardware will provide processing power to enable the features needed for NTN and boosting routing and security in space nodes. AI approaches will be used to design novel security mechanisms as well as control plane mechanisms. Furthermore, efficient and adaptive approaches for flight control for formations of satellites exploiting the advantages of a data driven design will be developed.
The main target products developed in AIComS will implement key functionalities of a space based 5G&beyond network such as baseband processing, routing, security for communications and (v)LEO satellites using appropriate ACS. For integration with 5G and beyond 5G networks, compliance of the developed functionalities with 3GPP standardization will be targeted.
The target main communication products are:
5G NTN Baseband Processing Platform: lower layer gNB NR PHY functionality for service link (NR-Uu) in (v)LEO-Satellite.
Layerscape: higher layer gNB NR PHY functionality for service link (NR-Uu) in gNB (DU/CU) based on NXP Layerscape® processor.
5G AI Satellite Packet Router: Security, QoS and traffic aware IP Routing (GTP-U and F1-AP) and IP-level network slicing functionality for Inter-Satellite-Links and service-feeder link in (v)LEO-Satellite (satellite based backhaul network) and NTN gateway.
This is supported by target main products of flight control with corresponding functionality
AI based Fault Detection, Identification and Recovery (AI-FDIR) Software: continuous satellite service capabilities providing “radiation protection by software”.
AI based Formation Control (AI-FC) Algorithms: ensuring accurate positioning of satellites in formation.
AI based VLEO Orbit Control (AI-VLEO-OC): ensuring satellite lifetime and communication service provisioning as required, i.e., antenna beam pointing, navigation and AOCS as well as drag compensation.
AIComS considers a regenerative satellite-based NG-RAN architecture following the ongoing discussion in 3GPP. The figure shows the assignment of the target main products to the network elements to be developed in AIComS.
The project consists of two development steps.
The first step is planned to last for 12 months and covers the investigation of basic concepts for the targeted main products. The first step includes two reviews. For MS1 the preliminary design was provided in Q2/23, whereas MS2 in Q4/23 delivers preliminary conclusions and recommendations of this development step.
The second step targets to advance these concepts towards development, evaluation and verification of the corresponding products. The second step encompasses four reviews and is planned to last for 24 months.
The project was kicked off February 2023, with prior work started in November 2022.
The identified uses cases, scenarios and potential architecture have been reported in the first milestone meeting in May 2023. The basic concepts for the main target products are investigated.
Work currently in progress.