//
How does RTC facilitate telemedicine? Where is the differentiation of 5G? How does SMT maximize remote control? On July 29th, LiveVideoStackCon2023 Shanghai Station Transmission Network Optimization Session will answer your questions.
Transport Network Optimization
In response to the increasing acceleration requirements of the Internet, complex network environments, and various video services, technicians are constantly exploring how to achieve accurate and extremely low-latency network transmission capabilities. It is also actively exploring aspects such as application layer flow control, transport layer protocol design, and cross-layer optimization to bring users a better network experience.
TOPIC1 "Adaptive Streaming Media Intelligent Transmission Optimization"
Huang Tianchi Ph.D., Tsinghua University
Adaptive streaming video transmission is an important network application that improves user viewing experience by dynamically selecting bit rate gears. Due to the limitations of fixed parameters and traditional modeling, traditional heuristic bit rate adaptive algorithms are gradually replaced by data-driven intelligent bit rate adaptive algorithms. The algorithm significantly improves the experience by building models to assist traditional algorithms, make decisions, or generate intelligent strategies. However, problems such as network time-varying, heterogeneity, and user demand diversity require intelligent algorithms to rely on closed-loop systems to continuously optimize performance.
This report focuses on the four aspects of efficient training models, reducing execution overhead, quickly adapting to the environment, and meeting diverse needs. The user experience in the scene improves the feasibility of intelligent algorithms.
The first part introduces the intelligent code rate adaptive optimization algorithm based on lifelong imitation learning, which improves the model training efficiency of the server;
The second part introduces a low-overhead intelligent code rate adaptive fusion scheme combined with a heuristic algorithm mechanism, which improves performance and reduces overall overhead;
The third part introduces the intelligent code rate fast adaptive method based on meta-reinforcement learning, which improves the user experience through the fast optimization strategy;
The fourth part introduces the demand-driven intelligent adaptive video transmission algorithm based on the self-play theory, which meets the diverse user needs.
TOPIC2 "Technical Practice of NetEase Yunxin 4K/8K RTC Assisting Telemedicine"
Zhu Mingliang Netease Yunxin Audio and video engine development expert, video engineering team leader
At present, the distribution of domestic medical resources is uneven, and medical collaboration is an important way to solve the uneven distribution of medical resources and improve the efficiency and value of limited resources. With the implementation of "hierarchical diagnosis and treatment", remote surgery teaching, video conferencing, cross-domain medical training Collaboration can bridge the gap in medical resources between urban and rural areas, which creates a strong demand for ultra-high-definition real-time audio and video communication technology. NetEase Yunxin provides 4K/8K ultra-high-definition real-time audio and video communication capabilities for the medical industry, and has in-depth cooperation with a number of well-known medical software and hardware ecological partners in the industry to form a scenario-based solution, which has been used in many public tertiary hospitals The telemedicine and telesurgery teaching scenarios successfully landed. This sharing is mainly divided into three parts:
The first part introduces the needs of the industry and the solutions and applications provided by NetEase Yunxin for the medical industry;
The second part introduces the practice of Yunxin's 4K/8K RTC ultra-high-definition audio and video communication technology, including technical implementation and optimization;
The third part introduces the general application of 4K/8K RTC in telemedicine.
TOPIC3 "Caton Media Xstream: Redefining Real-time Content Delivery Service"
Wei Ling Keteng Technology Network Protocol and Algorithm Leader
As the public Internet becomes more and more complex, the basic prototype of best effort can no longer meet the growing number of real-time content delivery services that require QoS guarantees. However, traditional solutions such as leased lines and satellites have problems such as high deployment costs and long cycles, and cannot quickly respond to various needs.
In order to provide low-cost, highly reliable transmission services, Caton, the leader of next-generation IP network transmission solutions, launched the Caton Media Xstream platform. Through AI-based intelligent routing algorithms and high-reliability, high-quality transmission strategies, the platform has achieved broadcast-level transmission standards on the low-cost public Internet, with an SLA exceeding six nines.
Caton Media Xstream is based on the global distributed cloud architecture Caton Cloud, which supports hybrid network deployment and efficient expansion. It not only realizes efficient connection and forwarding of traffic, but also supports rapid deployment, real-time traffic monitoring and other functions. Caton Media Xstream built its own big data platform. Based on big data technology, the platform continuously iteratively optimizes intelligent routing algorithms through mining and learning of real-time network data to improve transmission performance.
TOPIC4 "The Development Practice of SMT in Remote Production and Virtual Cloud Production"
Zhai Shuang Shanghai Radio and Television Technology Center Deputy Director of Broadcasting Department of Shanghai Oriental Media Co., Ltd.
With the development of streaming media technology, virtualization technology and cloud technology, as well as objective factors such as the epidemic, the application of remote production has become more and more extensive, virtualization and cloud production have gradually begun to take off, and a new broadcast mode is also emerging. It is accepted and recognized by all major competitions in the world. What can remote production and virtualized cloud production bring to customers? Is it the improvement of production quality, similar to the new audio-visual experience brought to us by HDR, 4K, 8K, and three-dimensional sound, or the substantial reduction in production costs? worth discussing.
This sharing is divided into three parts. The first part introduces several important remote production cases of SMT, the second part introduces a preliminary attempt of virtualization and cloud production, and finally some thoughts on the future fusion production mode.
TOPIC5 "Thoughts on 5G Differentiated Experience and Industry-Network Collaboration"
Zhu Jiayue, Deputy Director of Business Planning, Huawei Wireless Application Scenario Laboratory
The 3G era mainly carries graphic content, and the 4G era mainly carries video content. What is the differentiated experience in the 5G era? What requirements and challenges do these differentiated experiences bring to network transmission? This sharing will be divided into three parts. The first part will introduce the differentiated experience and target experience in the 5G era; the second part will introduce in detail the value and significance of GAP and industry-network collaboration where the current network and application layers are separated from the target experience; Some of the current mainstream industry-network collaboration solutions will be explained in detail.
TOPIC6 "Multipath Transmission Optimization for Ultra-Low Latency"
Wen Nian Tencent Senior Engineer
With the development of the Internet industry, from text, voice, pictures, and video, to the very popular live broadcast and video conferencing in the past two years, the entire industry is developing in the direction of increasing traffic and increasing real-time performance. Compared with the most demanding video call scene in the industry, cloud games have higher requirements, the delay is reduced from 100 milliseconds to 20 milliseconds, and the bit rate is increased from 2Mbps to 30Mbps, which poses a new challenge to the real-time performance of network transmission .
The sharing of this article is mainly divided into three parts. The first part introduces the challenges of data reliability and network uncertainty to low-latency and high-bit-rate systems in real-time audio and video scenarios; the second part introduces mobile phones based on Wi-Fi + Cellular dual-chain The transmission scheme of the road, and how to find the best balance under the delay, bit rate and low traffic consumption; the third part introduces how to support flexible multi-path transmission in the system architecture design.
TOPIC7 "Audio and Video Experience of AR Scene with Device-Cloud Integration"
Party Yubo Rokid application platform audio and video director
It mainly introduces large scenes rendered in the cloud. Through real-time audio and video and command transmission, combined with 6dof positioning, the augmented reality experience of device-cloud integration is realized, and the application field of audio and video is expanded. First of all, it is necessary to achieve the same ultra-low-latency audio, video and data transmission as cloud games. Second, it is necessary to define interactive events, such as pose updates, gestures, rays, buttons, etc.; in order to better superimpose cloud images and real images, the third is Real-time translucent video transmission, alpha channel data only adds a small amount of bandwidth.
TOPIC8 "Intelligent Congestion Control and Its Practice in Real-time Audio and Video Transmission"
Liu Feida Nanjing University Ph.D.
With the continuous emergence of emerging applications (such as VR/AR, live broadcast services) and the continuous evolution of new networks (such as 5G, WiFi6), the existing Internet congestion control algorithms can no longer adapt to the diversity of upper-layer applications and the differences of underlying networks. The current mainstream classic congestion control algorithms have the advantages of stable performance, high practicability, and low overhead, and are widely deployed in operating system kernels. However, they are often adapted to specific scenarios and cannot meet the challenges brought by upper-layer applications and underlying networks. Algorithms based on machine learning have great potential for strong adaptability, but there are defects such as inability to guarantee fairness and convergence, and high overhead. We complemented the advantages of the two types of algorithms, designed an adaptive congestion control framework, and carried out preliminary practice in WebRTC live broadcast scenarios. Experiments show that the scheme we designed can effectively improve the anti-weak network performance and improve the QoE in real-time audio and video transmission.
▲Scan the QR code in the picture or click " Read the original text " ▲
Check out more highlights of LiveVideoStackCon 2023 Shanghai Station