Cooperative Multigroup Broadcast 360° Video Delivery Network: A Hierarchical Federated Deep Reinforcement Learning Approach

Toward immersive communications in 6G

“…Even if viewpoints are known in advance, dynamic network environments such as data traffic load and processing time require adaptive resource management to ensure playback performance. With stochastic decision-making methods, such as reinforcement learning, it is possible to identify the dynamics of user viewpoint movement and determine which tiled videos to deliver to the corresponding VR device (Hu F. et al, 2022). In addition, the portion of tiled videos with different video qualities transmitted in a given time interval can be adjusted according to the viewpoint movement of a user.…”

“…Immersive communication: Open issues and future directions Despite an increasing amount of studies and solutions for supporting XR, haptic communication, and holographic communication, there exist many open issues to address before immersive communications can popularize. To name a few, synchronization of multi-modal communications, user QoE modeling and enhancement, and intelligent network (Hu M. et al, 2022) and machine learning (Hu F. et al, 2022) methods…”

Toward Immersive Communications in 6G

“…Even if viewpoints are known in advance, dynamic network environments such as data traffic load and processing time require adaptive resource management to ensure playback performance. With stochastic decision-making methods, such as reinforcement learning, it is possible to identify the dynamics of user viewpoint movement and determine which tiled videos to deliver to the corresponding VR device [185]. In addition, the portion of tiled videos with different video qualities transmitted in a given time interval can be adjusted according to the viewpoint movement of a user.…”

“…In addition, lower-layer approaches can take into account application-related information for efficient network resource management, e.g., timely adjusting the amount of radio resources allocated to a user in response to the dynamic sensitivity of the user's perception. Since multi-modal perception data in immersive communications can include personal biometric information of individual users, privacy challenges can arise in Balancing the robustness of viewpoint prediction error and video quality by optimization [186] and machine learning [185] methods…”

Hierarchial-DQN Position-Aided Beamforming for Uplink mmWave Cellular-Connected UAVs

“…DQN is a value-based approach, learning an optimal approximated policy of states mapping to actions π(s) = a by parameterizing and estimating state-action value function Q(s, a; θ) where θ denotes the weight matrix of the primary deep neural networks (DNN) [13]. The hDQN framework integrates hierarchical action-value functions operating at different temporal scales using DQN approach and learns optimal approximated policies π 1 (s) = a, a ∈ A B and π 2 (s) = a, a ∈ A N , respectively [14], [15]. Under our hDQN framework, we consider a broad beam (BB) and narrow beam (NB) DQN agents over the same state space S but different action spaces A B and A N , respectively as shown in Figure 3.…”