Title

AWS re:Invent 2023 - Ground station digitalization with SDRs in the cloud (AES303)

Summary

• Introduction by Alistair McLean and Ajit Rajdeo Singh, discussing the agenda and the move from hardware-defined radios to software-defined radios (SDRs) in the cloud.
• Explanation of SDRs and their role in the digitalization of ground station components, with a focus on the use of Amazon EC2 accelerated compute capabilities.
• Discussion of use cases for SDRs in satellite communications (SATCOM) and Earth observation, highlighting the differences in requirements and benefits of cloud-based modems.
• Deep dive into FPGA programming, the EC2 F1 development workflow, and the advantages of using FPGAs for parallel processing in digital signal processing.
• Demonstration of high-performance network access using the Data Plane Development Kit (DPDK) and a packet generation demo.
• Invitation for one-on-one chats in the peer talk room for further discussion and questions.

Insights

• The industry is shifting towards the use of SDRs in the cloud, which allows for the virtualization of hardware processing stacks and offers flexibility and cost savings compared to traditional hardware modems.
• AWS EC2 instances with accelerated computing capabilities, such as AVX 512, GPUs, and FPGAs, are particularly suited for the parallel processing requirements of digital signal processing in SDRs.
• FPGAs are highlighted for their adaptability and high bandwidth memory, making them ideal for processing large amounts of data in parallel, which is essential for applications like satellite image processing and data transmission.
• The EC2 F1 instances are equipped with Xilinx VU9P FPGAs and offer a range of instance sizes to accommodate different performance needs, with up to 6,800 DSP engines per instance.
• The development workflow for FPGAs on AWS includes using an Amazon Machine Image (AMI) with pre-built FPGA development environments, compiling custom logic into an Amazon FPGA Image (AFI), and loading the AFI onto the FPGA board.
• The use of DPDK is recommended for network access to FPGA boards in EC2 instances, as it provides cloud-native network security, allows for advanced packet processing, and can achieve high network throughput.
• The demonstration showcased the ability to achieve up to 8 Gbps of throughput with jumbo frames, indicating the high performance of the EC2 F1 instances for network-intensive tasks.
• The session concluded with an open invitation for questions and further discussions, emphasizing the availability of code examples on the AWS GitHub repository for attendees to explore and experiment with.