Due to the intensive computing resource requirements, real-time video analytics applications typically need to transmit video to a server. However, the transmission inevitably suffers from network bandwidth limitations and fluctuations, making it challenging to guarantee video analytics performance. In this paper, we aim to maximize video analytics accuracy while maintaining low latency and frame loss rate, and propose a joint segment and frame bitrate adaptation (JSFBA) framework for real-time video analytics, which incorporates two reinforcement learning-based algorithms to adapt to bandwidth at both the segment and frame levels. Initially, considering the effect of video encoding on video analytics, we employ a bitrate control method to design a segment-level bitrate adaptation (SLBA) algorithm with a unique reward function. Based on the historical information of the video segments, SLBA selects the appropriate bitrate for each segment. Subsequently, by leveraging the ability to generate multiple bitrates in scalable video coding (SVC), we design a frame-level bitrate adaptation (FLBA) algorithm, which adapts to bandwidth in a more fine-grained manner by determining the number of layers sent for each frame. Extensive experiments on large-scale network traces reveal that JSFBA effectively balances various video analytics performance metrics and achieves maximum utility compared to state-of-the-art solutions.

Due to the huge file size of virtual reality (VR) videos, providing high-quality and low-latency live VR videos to multiple users using existing network architectures becomes exceptionally challenging. Although some tile-based multicast schemes can reduce the amount of transmitted data, users with differentiated channel conditions and interaction behaviors still need to be effectively grouped. In this paper, we first model VR video multicast scheduling in the case of orthogonal multiple access (OMA), where the objective of maximizing the sum of user experience qualities forms an NP-hard integer nonlinear programming problem. By analyzing the time complexity source of the exhaustion-based global optimal algorithm and the impact of two factors on grouping, we propose a scheduling algorithm with very low complexity based on dynamic programming in the OMA case and a field-of-view ratio sorting and grouping method for non-orthogonal multiple access (NOMA) scenarios. In extensive actual viewport experiments, the proposed algorithm outperforms the state-of-the-art multicast scheme adopting OMA by 17.86% with negligible time complexity, making it suitable for deployment in realistic environments. The proposed sorting method increases the system utility of NOMA’s optimal solution by 13.13%.

By providing an immersive experience with panoramic views, 360-degree video streaming has gained increasing popularity recently. In many cases, videos are transmitted to mobile users over cellular networks. However, due to the high bandwidth requirement of 360-degree videos and the growing number of users, it is challenging to provide high-quality live streaming services to all users with limited bandwidth. Improving spectral efficiency and reducing bandwidth consumption are two major approaches to address this issue, which can be achieved with non-orthogonal multiple access (NOMA) and scalable video coding (SVC), respectively. In this paper, we apply NOMA and SVC to 360-degree video streaming over a cellular network and propose a multicast scheme called SVCast, aiming to maximize the sum quality of experience of users served by a base station. Such a problem is formulated as an NP-hard problem, and we decompose it into two levels of subproblems. The lower-level subproblem is inter-group spectrum allocation, which is solved by a knapsack approach. The higher-level subproblem is intra-group multicast scheduling, and we propose a recursive algorithm to solve it. Simulation results demonstrate that SVCast improves the system utility by 31.9% on average. Furthermore, SVCast eliminates the need for viewport prediction by aggregating the contents from the viewports of multiple users.