SPAICE - Satellite Signal Processing Techniques using a Commercial Off-The-Shelf AI Chipset
Funding Source: ESA – European Space Agency
Team: Symeon Chatzinotas (PI), Jorge Querol (VPI), Eva Lagunas, Lei Lei, Flor Ortiz Gomez, Jan Thoemel, Juan Carlos Merlano Duncan, Jorge Luis Gonzalez Rios
Starting date / Duration: Nov 2021 / 26 months
Summary
The SPAICE project aims to study, develop, and validate Artificial Intelligence (AI)-based signal processing techniques for satellite communications in scenarios and use cases where specific AI processors can provide a significant performance improvement with respect the current state-of-the-art. The SPAICE project has as its principal outcome the AI Satellite Telecommunications Testbed (AISTT), which will be the platform to test and demonstrate the AI-accelerated scenarios. We have already identified and traded-off some prospective scenario candidates such as interference detection, interference localization, link adaptation and error correction for regenerative satellites, and the suitable ML architectures, framework and off-the-shelf chipsets. We believe that the University of Luxembourg team has the required technical knowledge gained through the participation in different AI/ML-related research projects, the experimental expertise and available facilities (CubeSat Laboratory, Satellite Channel Emulator, End-to-end Satellite Communications Testbed) to complete successfully the design and training of the ML algorithm, the implementation and validation of the AISTT, and the evaluation of its potential road-to-the-market.
Description
The objective of the activity is to develop and validate Artificial Intelligence (AI)-based signal processing techniques for satellite communications such as signal identification, spectrum monitoring, spectrum sharing or signal demodulation using an Off-the-Shelf AI chipset. The activity will develop a laboratory testbed to validate and demonstrate the developments in both on-ground and on-board scenarios.
The targeted improvement of this activity with respect to the existing state-of-the-art is enabling on-board real-time satellite signal processing techniques currently not possible without specific AI processors/accelerators.
We propose several candidate scenarios and targeted applications in the field of satellite telecommunications, which are computationally challenging in terms of required on-board processing. In the vast majority, these applications are also delay-sensitive, in the sense that real-time processing and decisions are needed for effective operation/communication.
Existing solution for the proposed use cases usually exhibit inherent limitations in translating theory to practice when handling the computational complexity and/or the latency required for the outcomes specially when dealing with large search space or high degrees of freedom. This has been typically the motivation for the use of AI, since it has been shown to have a strong potential to overcome this challenge via data-driven solutions. In this project, we will consider these criteria and evaluate what are the expected gains, both in terms of latency, complexity and performance that can be achieved with the application of AI-based techniques in each of the considered use cases.
Table 1: Proposed candidate scenarios
| Scenario |
Benefit of AI-based technique |
|
| 1 |
Interference Detection |
Reduce the expert human intervention in order to generate alarms when interferences are occurring to perform operations in an automated fashion |
| 2 |
Interference Classification |
|
| 3 |
FEC in regenerative payloads |
Reduction of the implementation complexity and power consumption the FEC algorithms using ML is of particular importance in regenerative satellite payloads. |
| 4 |
Link Adaptation / ACM Optimization |
Improve typical channel model-based link adaptation to better predict the best ACM for given channel conditions. |
| 5 |
Flexible payload reconfiguration / Spectrum sensing |
Complexity reduction when addressing the optimal configuration of a flexible payload. |
| 6 |
User demand (congestion) prediction |
ML-based method can predict network loads and detects congested areas before they actually experience congestion. |
| 7 |
User allocation and network optimization |
Use ML to reduce complex optimal algorithms to speed-up the convergence to a solution with good performance. |
| 8 |
Active antenna array satellite optimization |
Reduction of complexity of beam and radio resource optimization achieving good performance. |
The setup of the data generation may slightly change depending on the implemented scenario. From the ones considered as potential candidates in the previous Section, the #1 Interference Detection and #2 Interference Classification can use the setup summarized in Figure 1. For this scenario, the satellite communications setup may be implemented using the DVB Gateway and Channel emulator available at UniLu to generate a realistic uplink waveform (e.g. DVB, 3GPP). The interference generation can be easily implemented using a generic SDR device with a predefined battery of interference signals. The channel emulator will generate the additive noise at the receiver side and control the interference power according to the selected geometry, orbit, antenna gain, etc. The output of the channel emulator will be down-converted and digitalized using the Cubesat SDR. The output baseband samples will serve as the input data set for the AI-enabled application.
Figure 1: Interference Detection and Classification Setup
Figure 2 shows the ACM optimization setup for the candidate scenario #4. In this case, both gateway and terminal devices are connected to the channel emulator, and the satellite contains a regenerative payload. The estimated SNR time series will be the data to be collected in this case. Eventually, the predicted SNR values will be the outputs for the regenerative payload.
Figure 2: ACM optimization setup.
Figure 3 shows the FEC decoder setup for the candidate scenario #3. The setup is similar to the one in the previous case, but in this case the AI-enabled FEC decoder is part of the regenerative payload.
Figure 3: FEC decoder setup.
Once the targeted application is selected, we will use our established CubeSat design methodology, tailored here, for the design of the AISTT. The SPAICE Module assumes the role of a satellite payload to this end and drive the system design. The SPAICE module carrying the AI Accelerator will be connected to the AISTT through a development interface and a satellite interface both implemented on the breakout board. The latter interface will also be used for integrating the module into a satellite for flight. The two-fold interface architecture allows an IOD and market ready development of the SPAICE module. Common software interfaces will be adopted such as CAN, I2C and UART for the satellite connections and 10-1000 Base-T Ethernet, USB and RS422. The AISTT will be implemented in our mobile CubeSatLab payload development bench where we will test the non-AI functions. The AI functions will then undergo validation in the SatComLab.
Figure 4: UniLu/SnT CubeSatLab with CubeSat communication equipment and payload engineering bench space (bottom right).
Figure 5: End-to-end satellite communications testbed available at the SatComLab.
Project Team:
SIGCOM Group:
- Prof Symeon Chatzinotas
- Dr Jorge Querol
- Dr. Eva Lagunas
- Dr. Lei Lei
- Dr. Flor Ortiz Gomez
- Dr Juan Carlos Merlano Duncan
- Dr. Jorge Luis Gonzalez Rios
RSA Group:
- Dr Jan Thoemel
Relevant Projects:
Machine Learning, Artificial Intelligence, On-board and on-ground satellite signal processing for several applications such as spectrum sensing, cognitive radio, spectrum coexistence, reconfigurable payloads and interference cancelation techniques:
- AGNOSTIC - Actively Enhanced Cognition based Framework for Design of Complex Systems (see more).
- APSIM - Antennas and Signal Processing Techniques for Interference Mitigation in Next Generation Ka Band High Throughput Satellites (see more).
- SeMIGod - Spectrum Management and Interference Mitigation in Cognitive Hybrid Satellite Network (see more).
- SATSENT-SATellite SEnsor NeTworks for Spectrum Monitoring (see more).
- PROSAT - on-board PROcessing techniques for high throughput SATellites (see more).
- CoRaSat - Cognitive Radio for Satellite Communication (see more).
- SANSA - Shared Access terrestrial - Satellite Backhaul Network enabled by Smart Antennas (see more).
High TRL signal processing activities for both terrestrial and satellite communication systems:
- 5G-SpaceLab - 5G Space Communications Lab (see more).
- SERENADE - Satellite Precoding Hardware Demonstrator (see more).
- LiveSatPreDem – Live Satellite Precoding Demonstration (see more).
- CGD - Prototype of a Centralized Broadband Gateway for Precoded Multi-beam Networks (see more).
- DISBuS - Dynamic Beam Forming and In-band Signalling for Next Generation Satellite Systems (see more).
Contact: jorge.querol@uni.lu
