University of Twente Student Theses
Signal Processing with AMD Adaptive Compute Acceleration Platform (ACAP) for Applications in Radio Astronomy
Wijhe, Victor van (2024) Signal Processing with AMD Adaptive Compute Acceleration Platform (ACAP) for Applications in Radio Astronomy.
|
PDF
2MB |
| Abstract: | High-performance computing (HPC) has become essential due to the need to process enormous amounts of data quickly and accurately amongst others in the field of radio-astronomy. The Versal Adaptive Compute Acceleration Platform (ACAP) is a recent and promising alternative addition to the field of domain-specific accelerators. This thesis conducts a design space exploration of the Versal ACAP AI Engines using a polyphase filter. This study aims to assess the efficiency of utilizing the Artificial Intelligence engines within the ACAP for signal processing tasks in radio astronomy. We achieved this by deploying multiple streaming-based polyphase filter designs on the ACAP and conducting an evaluation of the AMD Tooling and AI Engines’ performance. A design space exploration was conducted to discover the existing libraries applicable to polyphase filters. The available FIR library from AMD reached a throughput of 208 M samples/s for a single branch but could not be effectively scaled to the requirements of the use case. The AMD FFT library managed an impressive throughput of 220 M samples/s on a single AI Engine and is used for the polyphase application. To overcome the AMD FIR library limitation, five distinct polyphase filter kernels have been designed ranging in effectiveness. A throughput of 312 M samples/s was reached when the filter is deployed on four AI Engines, however greater throughput could be reached by utilizing more AI Engines. Finally, a polyphase filter library is created adaptable to a wide range of use cases. Despite promising advancements such as the release of AIE-ML with enhanced performance and a 16-bit bfloat, the platform’s maturity of tooling remains a significant hurdle, as available features can not be optimally utilized. This work provides a foundation for future research and development when using the AMD Versal ACAP. |
| Item Type: | Essay (Master) |
| Clients: | ASTRON, Dwingeloo, The Netherlands |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Embedded Systems MSc (60331) |
| Link to this item: | https://purl.utwente.nl/essays/98354 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page