Evaluation of IP Communication in FPGA-based Camera Platforms Felix - PowerPoint PPT Presentation

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Evaluation of IP Communication in FPGA-based Camera Platforms Felix Lampe, B. Sc. July 24, 2017 Chair of Network Architectures and Services Department of Informatics Technical University of Munich

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Contents Problem and Motivation Analysis Approach Evaluation Conclusion F. Lampe – IP in Camera Platforms 2

Chair of Network Architectures and Services Department of Informatics Technical University of Munich What kinds of cameras are we talking about? Figure 1: ARRI ALEXA Plus and RED Epic Dragon [1, 2] • Professional digital cameras for cinema and broadcast productions. • ASICs or FPGAs for image processing, CPU for UI etc. F. Lampe – IP in Camera Platforms 3

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Devices at film sets are highly interconnected Figure 2: ARRI Alexa with monitors and accessories [3] • Applications: monitoring, processing, synchronization, control, . . . • Usually Serial Digital Interface (SDI) on copper wires F. Lampe – IP in Camera Platforms 4

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Problems of SDI Infrastructures • Necessary data rates (Gbit/s) for uncompressed video: HD 4K 8K 30 fps 1.5 6.0 24.0 60 fps 3.0 12.0 48.0 120 fps 6.0 24.0 96.0 • SDI currently supports up to 12 Gbit/s per link • Low flexibility: SDI is unidirectional and circuit switched • Results in many point-to-point links F. Lampe – IP in Camera Platforms 5

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Possible Solution: IP Communication Research Questions: • What are the benefits and requirements of integrating IP communication into a motion picture camera platform? • How can such IP communication be realized in an FPGA-based camera and what are possible difficulties? F. Lampe – IP in Camera Platforms 6

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Advantages of IP • Data rates up to 100G, promoted by telecommunications industry • No restrictions to flows (only fixed stream sizes in SDI) • Arbitrary payload contents, not just audio/video/meta in fixed ratio • Bidirectional, packet-switched links • Easily extensible, cheap, widely used F. Lampe – IP in Camera Platforms 7

Chair of Network Architectures and Services Department of Informatics Technical University of Munich SDI over IP (SMPTE 2022) • Transition mechanism: wrap SDI in IP • High Bit-Rate Media Transport Protocol (HBRMT), RTP , UDP , IP • Overhead of 6.25% , units of 1376 bytes • Synchronization via PTP instead of coax cable F. Lampe – IP in Camera Platforms 8

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Quality of Service • Different QoS for different applications • E.g., low latency and guaranteed bandwidth for live video • DiffServ: simple and efficient, provides enough classes • Expedited Forwarding for live video, Best Effort for remote control, . . . F. Lampe – IP in Camera Platforms 9

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Designed Architecture for Camera Communication Image Chain CPU OS Kernel (De-)Packetizer Multilayer Switch MAC MAC MAC PHY PHY PHY SFP+ SFP+ SFP+ Figure 3: Block diagram of the proposed architecture F. Lampe – IP in Camera Platforms 10

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Prototype • Implements part of the proposed architecture • Packetizer limited to Ethernet/IP • No CPU, configuration via USB • Implemented in two steps Figure 4: The Altera/Intel Arria 10 development kit running the prototype F. Lampe – IP in Camera Platforms 11

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Early Prototype – Two Variants FIFO Generator Monitor Depacketizer Depacketizer Packetizer Packetizer MAC MAC PHY PHY SFP+ SFP+ Figure 5: The forwarding variant and the generating variant of the early prototype • Forwarding variant for latency measurements • Generating variant can measure its own throughput F. Lampe – IP in Camera Platforms 12

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Early Prototype – Throughput 16 14.882  eoretical Limit 14 Measured Rate Packet Rate (Mp/s) 12 10 8.446 8 6 4.529 4 3.094 2.350 1.586 2 1.197 0.962 0.813 0 64 128 256 384 512 768 1024 1280 1518 L2 Frame Size (Bytes), excluding Preamble, SFD and IFG Figure 6: The throughput achieved by the generating variant F. Lampe – IP in Camera Platforms 13

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Early Prototype – Latency No Background Tra ffi c 4500 Mbit/s Background Tra ffi c 50 50 Mean: 571.7 ns Mean: 581.6 ns Occurrence (%) Occurrence (%) 40 40 StdDev: 4.2 ns StdDev: 27.7 ns 30 30 20 20 10 10 0 0 550 560 570 580 590 550 560 570 580 590 Latency (ns) Latency (ns) Figure 7: The latency of the early prototype F. Lampe – IP in Camera Platforms 14

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Final Prototype Generator Monitor (De-)Packetizer Multilayer Switch MAC MAC PHY PHY SFP+ SFP+ Figure 8: The structure of the final prototype (simplified) F. Lampe – IP in Camera Platforms 15

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Final Prototype – Latency 1000 Mbit/s BG Tra ffi c 4500 Mbit/s BG Tra ffi c 7 7 6 6 Mean: 1685.3 ns Mean: 1656.4 ns 5 5 Occurrence (%) Occurrence (%) StdDev: 35.2 ns StdDev: 33.5 ns 4 4 3 3 2 2 1 1 0 0 1550 1600 1650 1700 1750 1800 1550 1600 1650 1700 1750 1800 Latency (ns) Latency (ns) Figure 9: The latencies measured with the final design of the prototype F. Lampe – IP in Camera Platforms 16

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Final Prototype – Latency (QoS) 4500 + 4500 Mbit/s BG Tra ffi c 4500 + 4500 Mbit/s BG Tra ffi c 5 5 Mean: 1657.6 ns Mean: 1657.5 ns 4 4 Occurrence (%) Occurrence (%) StdDev: 32.2 ns StdDev: 33.6 ns 3 3 2 2 1 1 0 0 1500 1550 1600 1650 1700 1750 1800 1500 1550 1600 1650 1700 1750 1800 Best E ff ort Latency (ns) Expedited Forwarding Latency (ns) Figure 10: The latencies measured with the final design of the prototype F. Lampe – IP in Camera Platforms 17

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Conclusion • Ethernet/IP is suitable for communication • Higher flexibility than compression, multi-link SDI, . . . • Developed architecture is promising • Latency performance needs more accurate evaluation F. Lampe – IP in Camera Platforms 18

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Thanks for listening! Time for questions F. Lampe – IP in Camera Platforms 19

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Early Prototype – Latency (Loopback) 100 No Background Tra ffi c 100 Mbit/s Background Tra ffi c 80 1000 Mbit/s Background Tra ffi c Occurrence (%) 4500 Mbit/s Background Tra ffi c 60 40 20 0 407 408 409 410 411 412 413 414 415 416 417 418 Loopback Latency (ns) Figure 11: The latency of the test environment without the FPGA F. Lampe – IP in Camera Platforms 20

Chair of Network Architectures and Services Department of Informatics Technical University of Munich Final Prototype – Latency (Loopback) 1000 Mbit/s BG Tra ffi c 1000 Mbit/s BG Tra ffi c 100 20 10 Mean: 327.7 ns Occurrence (%) Occurrence (%) StdDev: 32.1 ns 15 1 10 0.1 5 0.01 0 0.001 200 250 300 350 400 450 500 550 200 250 300 350 400 450 500 550 Loopback Latency (ns) Loopback Latency (ns) Figure 12: The latency of the test environment without the FPGA F. Lampe – IP in Camera Platforms 21

Chair of Network Architectures and Services Department of Informatics Technical University of Munich References I [1] S. P . Anderson, “Arri Alexa,” Apr. 2014, accessed June 16, 2017. Licensed under https://creativecommons.org/licenses/by/2.0/legalcode. [Online]. Available: https://www.flickr.com/photos/seanpanderson/14390225393/ [2] V. Hyvönen, “RED Epic Dragon 6k,” Dec. 2013, accessed June 16, 2017. Licensed under https://creativecommons.org/licenses/by-sa/2.0/legalcode. Brightness reduced. [Online]. Available: https://www.flickr.com/photos/villehoo/ 11216966276/ [3] “ARRI Alexa rig,” Dec. 2013, accessed July 23, 2017. [Online]. Available: https://i.imgur.com/AXPMX.jpg F. Lampe – IP in Camera Platforms 22