PAGE CONTENTS
Objectives
The primary objectives include designing and testing an AI model for physical layer processing tasks, such as symbol soft de-mapping and channel decoding, taking as reference the CCSDS 131.2-B-2 standard. This involves a critical assessment of the state of the art and the definition of a new suitable topology for an AI model, as well as adequate training and validation strategies. The performance is then compared with a standard demodulator in terms of cost, computational complexity, power consumption, and bit error rate.
Benefits
The complex algorithms for symbol soft de-mapping and decoding involved in satcom receivers require high computational power to achieve the desired performance in terms of bandwidth and bit error rate.
In particular, dedicated processing platforms based on FPGA accelerators are commonly used to speed-up these complex recievers. This is usually not problematic for ground station receivers because their design can rely on commercial components and there are no stringent constraints on power consumption, complexity, and weight. However, high-performance represent a big challenge for on-board satellite applications, where only rad-hard components can be used and mass is a major concern.
Established literature on Artificial Neural Networks (ANN) shows that they generally provide a good approximation to complex non-linear functions and therefore are considered a technology with the potential to bring significant improvements to physical layer processing in satcom receivers. In particular, a lightweight AI-enhanced receiver is expected to enable a better trade-off between bit error rate and bandwidth at a reduced computational cost and power consumption.
Features
The system deployed at the end of the project consists of a CCSDS 131.2-B-2 transmitter, channel noise emulator, and receiver, the latter featuring the AI-enhanced demodulator. Using this model of an end-to-end communication system, the performance of the innovative demodulator is exemplified.
Challenges
The following are the key challenges to be addressed:
- Designing and fine-tuning an AI model for physical layer processing tasks like symbol soft de-mapping and channel decoding;
- Generating a large and diverse dataset that accurately represents the possible different conditions of the communication channel to effectively train the AI model;
- Measuring the performance and comparing it to a traditional demodulator while considering various metrics (cost, power consumption, BER, implementation loss, etc.).
System Architecture
The architecture of the system was composed of a CCSDS 131.2-B software simulator supported by GPU accelerators. These accelerators speed up both the inference of the designed AI model and the entire CCSDS 131.2-B chain.
During the project, practical hardware constraints led to a reassessment of the original plan to port the model onto the GPU@SAT IP‑core. In response, IngeniArs adopted a more flexible approach, developing a higher‑complexity model unconstrained by edge‑computing memory limitations. This enabled successful operation in low‑noise conditions and compliance with the CCSDS 131.2‑B‑2 standard. The results in additive white Gaussian noise (AWGN) scenarios highlighted clear areas for further optimisation and provide valuable insight for guiding future model refinements.
Plan
The project followed a well‑structured development flow, starting with requirements definition and the identification of a preliminary AI model topology, culminating in the System Requirements Review. This was followed by the planning of the AI model design and dataset generation activities, which supported a successful Preliminary Design Review. The subsequent phases focused on AI model implementation, performance evaluation, and consolidation at Critical Design Review level. While initial efforts targeted porting the model onto the GPU@SAT platform, the stringent hardware constraints prompted a strategic refinement of the approach, enabling a re‑iteration of the design phase to better align performance objectives with implementation feasibility. The final phase centred on a critical evaluation of the achieved results and the definition of a clear roadmap for the future utilisation of this technology in space applications, concluding with the Final Review.
Current Status
The project is completed.
